Built from firsthand experience growing and scaling more than 20 vertical SaaS platforms, this article explores what high-performing software companies do differently to increase merchant adoption, convert existing customers, and create long-term enterprise value. Read the article here.

Alright, let’s get to it…

The market is not paying for AI features. It is paying for AI re-architecture.

Every pre-AI vertical SaaS company is running the same experiment right now. Some are launching a separate AI product on top of their legacy data. Some are rebuilding the UI inside the core app. Some are adding a chat panel. Some are trying to sell an AI Agent marketplace. Some are buying their way in through M&A.

A few are already getting rewarded. A few are getting punished. Some strategies are too early to tell if they will work out or not, but we can all agree the ceiling is enormous…

Pre-AI Incumbents Have Five Paths Forward. Here Is What Is Working, What Is Not, and What Is Just Early.

Path #1: Build a net new AI platform on top of your legacy data, separated from the legacy product

This is the most interesting path in the market today and it is not close.

The move is simple to describe and hard to execute. You take a pre-AI incumbent with a real data spine and you spin the AI product out as its own brand, SKU, pricing model, and often its own GTM. The legacy product still runs. The legacy customers still pay. But the AI platform sits alongside it, feeds off proprietary workflow data, and gets sold with an AI-native pricing model.

You keep the moat. You escape the multiple.

Intercom is the reference case. They renamed the entire company Fin. Fin surpassed $100M in ARR growing 350% YoY, resolving nearly 2M queries per week across 7,000+ customers. NRR moved from 112% to 146% after outcome-based pricing. Salesforce agreed to acquire Fin in June 2026 for ~$3.6B, roughly 9x ARR. Same company. Different re-architecture.

The critical unlock was that Fin became sellable outside Intercom. Fin for Platforms runs on top of Zendesk or Salesforce. The AI product does not just defend the installed base, it expands TAM to everyone using a competitor’s SoR.

In vertical software, the pure Fin move is rarer but starting to happen. RealPage launched Lumina AI Workforce with OpenAI as multifamily’s first agentic AI platform. AppFolio launched Realm-X Performers as a distinct product line and reported nearly 500% QoQ Performer adoption growth, with Leasing Performer customers seeing 35%+ more showings booked.

The reason this path is winning is structural. Markets know how to price AI-native usage growth. They do not know how to price a legacy per-seat business with an AI chatbot glued on. Architect the AI platform as its own product with outcome-based pricing and you tell two stories at once: durable legacy ARR underneath, AI-native growth on top.

Pros:

• No cannibalization of legacy SaaS revenue. Outcome-based pricing creates a net new revenue line on top of the SoR.

• AI-native multiple stacked on legacy substrate. Dual narrative: durable ARR underneath, AI-native growth on top.

• Upsell into the SoR without a rip and replace. Land with AI on a competitor’s stack, expand into your own SoR.

• New buyer, new budget. You are not fighting for share of the same line item.

• Fresh pricing model. Per-outcome pricing is nearly impossible inside a mature per-seat contract.

• Cleaner category story. “Fin,” not “Intercom with AI.”

• Dramatically better NRR mechanics.

• Optional standalone exit path.

Cons: if the new product is a wrapper without deep hooks into proprietary workflow data, it collapses into a thin skin. Requires a second product team and a second GTM motion.

Path #2: Make AI the UI, not a feature inside the old UI

The most ambitious path. Not autofill. Not a copilot popover. Not AI-flavored buttons inside a legacy grid view. The actual bet is that the primary way customers interact with your product stops being forms, tabs, and menus, and starts being a generative surface driven by intent.

The 1,000-button screen is dying whether SaaS vendors like it or not. The question is whether the incumbent leads that transition or gets replaced.

The path plays out in three phases:

Procore is directionally trying this with Procore AI. Whether they re-architect the primary UI or just add smart panels around existing screens is the open question. Q1 2026 was 16% revenue growth with 95% GRR. That is not proof of a UI transformation. That is proof they still have time to run the play.

The candid read is that no pre-AI vertical incumbent has convincingly landed Phase 3 yet. Toast IQ is smart but SMB restaurant operators run a shift, not a chat surface. The winners will be incumbents who recognize the old UI is a liability and cannibalize their own screens before someone else does.

Pros: if you execute Phase 3, you become the AI-native version of yourself and neutralize the entire class of AI-native attackers. Highest defensive value of any path. Cons: massive product risk. Requires deprecating UX that customers are trained on and CS teams are certified on. Nobody has fully pulled it off yet.

Path #3: Side chatbot / copilot as a permanent bolt-on

Different from a copilot as Phase 1 of Path #2. Path #3 is when the chatbot is the entire strategy and just lives forever on the side of the screen.

Fast to ship. Easy to demo. Rarely rewarded.

Clio Manage AI is the honest example. Summarizes matters, creates tasks, drafts messages. Useful. But structurally an assistant. The reason Clio got a $500M Series G at a $5B valuation was not the sidecar. It was the $1B vLex acquisition happening in the same breath. The sidecar was table stakes. The re-rating came from the platform move.

AppFolio’s own 2026 benchmark: 78% of surveyed property managers say they cannot rely on AI features in their legacy PMS.

Pros: low risk, useful transition state. Cons: rarely changes budget, market does not re-rate you for it. Stop here permanently and an AI-native competitor eats your lunch.

Path #4: Build an AI agent platform (sell agents individually, be the toll road)

Highest conviction longer term. Least public proof today. Nobody has evolved this into a full ecosystem yet, but the mechanics are unusually attractive.

You sell agent by agent, priced per outcome, outside the cost band of the underlying SoR. Receptionist agent, leasing agent, AP agent, claims agent. Each on its own budget line. Each measured against a KPI. Over time, third-party developers build on your rails and you take a rev share. That last part is what nobody has cracked.

ServiceTitan is the best directional proof point. AI Voice Agent and Contact Center Pro book jobs and handle capacity. Bonney Plumbing saw 11% more bookings and 60% fewer missed calls with Contact Center Pro. ServiceTitan crossed $1B ARR with 31% growth to $614.3M in a recent quarter. The agents work. The ecosystem is not there yet.

The pricing structure is why I like this path. Agent-by-agent sales bypass the pain of expanding an MSA. You are selling into a labor line, not a software line. That budget is 5 to 10x the size.

Pros: highest ceiling. Sells into labor budget, not software budget. Sits outside the SoR cost band. Optionality to become a real vertical agent marketplace. Cons: no proven ecosystem yet. Vertical buyers usually want painful jobs automated, not a platform. Winners probably look like Path #1 first.

Path #5: M&A

Not an AI strategy. An accelerant. Only works if you know which path you are accelerating.

Clio buying vLex for $1B is Path #1 wearing Path #5 clothing. Schneider buying Cognite for $3.1B is Path #1 for Schneider, buying a pre-built industrial data foundation for its AVEVA unit.

Pros: compresses time, buys missing workflow, data, or distribution. Cons: M&A without a chosen path is balance sheet gymnastics.

What is actually happening in the market

• Working today: Path #1. Fin, Realm-X, Lumina, Clio, Cognite. This is where the tape has moved.

• Directionally right, nobody has landed it yet: Path #2. Most incumbents ship copilots and call it transformation. Whoever gets to Phase 3 first defends against AI-native attackers.

• Not working today: Path #3 as a standalone strategy. If your entire AI strategy is a chat panel, the market has already told you what it thinks.

• Too early, high conviction on the ceiling: Path #4. ServiceTitan is proving the agents work. Nobody has proven the platform-of-agents thesis. Next cycle, not this one.

• Leverage, not a path: Path #5.

The market is paying for AI that captures work while getting priced like it was built yesterday. It is not paying for chat panels. And it is waiting to see who has the courage to blow up their own UI before someone else does it for them.

How to decide which path and why. Six questions.

1. Is your data actually proprietary and asymmetric? The gate for Path #1. If your data is generic, you will build a wrapper and the market will price it as one.

2. Can your current pricing model absorb outcome-based AI monetization? If per-seat SaaS pricing cannot flex to per-outcome economics, that is your Path #1 signal. Fin at $0.99 per outcome could not have existed inside Intercom’s per-seat structure.

3. Are you willing to blow up your own UI? The Path #2 gate, and a leadership question, not a product question. If you cannot commit to killing the 1,000-button grid view, you are doing Path #3, not Path #2.

4. Who is the new buyer? Same MSA signer → Path #2. New operational buyer (COO, VP Ops, RevOps) → Path #1. Labor line owner → Path #4.

5. What can your customer actually absorb? Regulated or fragile customers do not want a UI rewrite next quarter. Give them a Path #1 product they can pilot without touching the SoR.

6. One budget line or trying to be a platform? Path #4 only works if you have real evidence customers want multiple agents. Otherwise land one outcome, then expand. Earn the right to be a platform before you claim it.

Honest cross-tabulation:

• Proprietary data, rigid legacy pricing, new buyer → Path #1

• Legacy UI is a liability, buyer lives inside the software, CEO has the stomach → Path #2

• Weak data → do not ship AI yet

• Multiple workflows, proven single-agent adoption, labor budget → Path #4

• Missing workflow, data, or distribution → Path #5 in service of #1, #2, or #4

• Everyone else → Path #3 as a transition, not a destination

If your first instinct is a chatbot, wrong path. If your first instinct is to figure out what asymmetric data you own, or what your product looks like when the grid view is gone, right path.

Cognite (Vertical AI For Manufacturing): How a Norwegian oil company accidentally built one of the best AI platforms in the world

Most people first heard of Cognite in June 2026 when Schneider agreed to buy it for $3.1B.
Almost nobody knows how it got built…

The problem that could not be bought (2014-2016). In 2014, Karl Johnny Hersvik became CEO of what would become Aker BP. Mathematician. Software background. He mapped every piece of software the company used and found the same ugly answer everyone in industry finds: less than 1% of sensor and operational data was actually being touched by a decision-maker. He scanned the market for a solution. There wasn’t one. Every vendor solved a slice. Nobody solved the substrate.

Meanwhile, Aker ASA (Aker BP’s parent conglomerate) approached John Markus Lervik. Lervik had co-founded FAST Search & Transfer in 1997, sold it to Microsoft in 2008, then founded Cxense and taken it public. Two software companies, two exits, deep in enterprise search and data infrastructure.

If you made it thus far, go check out the 2026 Vertical Software Summit. We’ll have 400+ vertical founders/operators/investors in Miami in November. Two days,

6+ billion dollar vertical Founders/CEOs.
The Vertical AI event of the year.

Check Out The Summit

Aker asked Lervik to help with their strategy for the “fourth industrial revolution.” He sat down with the CEO of every Aker portfolio company plus 50 other industrial CEOs and came back with a blunt answer: this is not an application problem, not an AI problem, a data problem. IT and OT are behind a wall. Fix the wall, everything else becomes possible.

Aker didn’t take the memo. They asked him to run it. Cognite was incorporated December 7, 2016, with Lervik as CEO, alongside co-founders Geir Engdahl and Stein Kristian Foss. Aker BP was customer zero before there was really a product.

The pre-AI product (2017-2020). Cognite Data Fusion was ugly, boring, and important. Industrial DataOps. Connecting OT and IT data. Contextualizing sensor streams, engineering docs, and time-series data into a unified model. Building a living industrial knowledge graph. Then digital twins on top.

Lervik was blunt about what it required: if you took the 1,000 best engineers in Silicon Valley and dropped them on this problem without industrial domain expertise, they would fail. The team pulled from 61 nationalities, splitting roughly between software engineers and industrial domain experts hired directly out of operations.

The GTM most people missed. No SDR army. No outbound sequences. No PLG hooks. Captive-anchor land-and-expand.

Aker BP customer zero. Then Equinor, Statoil, Statnett, Aker Solutions across the Norwegian industrial ecosystem. Then bp, OMV, Aramco. Then Mitsubishi and Idemitsu. Every deployment co-designed against real assets. Aker ownership gave Cognite standing invitations into customer conversations across an entire ecosystem.

The CS motion was Palantir-style. Forward deployed engineers, not classic CSMs. Cognite still hires Senior Field Engineers with 3+ years hands-on in oil, gas, or process manufacturing. Not a call center. A services team disguised as engineering.

The capital story before anyone cared about AI. Accel came in early. TCV led a $150M round at a $1.6B valuation in May 2021, one of the largest SaaS rounds in Europe. Read that 2021 press release carefully. The word “AI” barely appears. The pitch is “modern industrial data management.” Cognite was raising unicorn money for infrastructure work most people found deeply uninteresting.

The Robot Dog moment tells you where the product was really pointed. November 2020: Aker BP and Cognite deployed Boston Dynamics’ Spot on the Skarv installation in the North Sea. Cognite Data Fusion made the images, scans, and sensor data actionable in real time. Public reaction: “cool robot.” Real signal: Cognite was quietly building the operating layer for autonomous industrial work.

The CEO handoff (2022). In April 2022, Lervik stepped aside. This is where founder-led companies can break. Cognite handled it well.

Girish Rishi came in from JDA Software, which he had spent five years converting from a legacy on-prem supply chain vendor into Blue Yonder, later sold to Panasonic for ~$7.1B. Rishi’s specialty: taking a deeply technical industrial software company and turning it into a commercial machine. Lervik stayed on as Chief Strategy & Development Officer. Founder didn’t leave. Founder made room.

Seven months later, ChatGPT dropped.

The cutover (2023-2025): the Path #1 move. Most industrial software companies froze. Some bolted on RAG chatbots (Path #3). Cognite did neither. They ran Path #1.

June 15, 2023: Cognite launched Cognite AI inside Cognite Data Fusion. A generative AI accelerator layered on top of six years of industrial data infrastructure. RAG over private industrial data. Copilots inside Industrial Canvas. Low-code app creation from natural language. The launch explicitly named the problems most AI vendors were failing on: hallucinations, data leakage, access control. The substrate was reframed as the reason the AI actually worked.

Then came the real Path #1 move. Cognite Atlas AI launched as a separately branded, low-code industrial agent workbench. New name. New product surface. New pricing narrative. Own landing page. Own positioning as “the only low-code industrial AI agent workbench.” All of it sitting on top of Cognite Data Fusion, which stayed the durable legacy substrate.

Just like Fin sits on top of Intercom’s data. Just like Realm-X sits on top of AppFolio’s. Just like Lumina sits on top of RealPage’s. The pattern is exact.

By September 2025, Cognite disclosed production usage of Atlas AI across Aker BP, ADNOC Offshore, Celanese, Hess, HMH, Idemitsu, and Tokuyama, with SLB and Radix building on top. Aker BP publicly reported Atlas AI agents reducing engineer time on root cause analysis by over 70%. Real operating outcome on real North Sea assets. That is the Path #4 flavor showing up inside a Path #1 architecture, which is the exact evolutionary arc I described earlier.

The payoff (2024-2026). 2024: nearly $100M ARR, up 38% YoY. 29 new logos. $23M in expansion, 41% outside oil and gas. MAU up 98%. HQ moved to Phoenix, domicile to the Netherlands.

2025: revenue crossed $170M with 36% ARR bookings growth.

June 30, 2026: Schneider Electric announced it would acquire Cognite for $3.1B in cash, folding it into AVEVA. Schneider’s CEO Olivier Blum said it directly: they were buying a “truly industrial-grade AI platform” built on “a unified industrial data model.” Aker shares jumped nearly 7%.

$1.6B in May 2021. $3.1B in June 2026. Roughly a doubling through a period where most late-stage software valuations were cut in half. That is the Path #1 re-rating in numbers.

Let’s assume they were at ~$200M of ARR today, they got bought for 15X revenue when only growing ~30%. If if they would have not embraced AI they’d probably be worth 2-3x ARR. Think about that for a second…

What Cognite teaches the framework.

Cognite is the cleanest end-to-end proof of the whole newsletter.

One, they had asymmetric proprietary data before they had “AI.” Six years of OT/IT contextualization, knowledge graph, and digital twin work. That is a strong foundation for Path #1, and Cognite is the industrial answer to what “asymmetric data” actually looks like in practice.

Two, when the moment came, they did not bolt AI onto Cognite Data Fusion and call it done. They spun up Atlas AI as its own product with its own brand and its own pricing narrative. Textbook Path #1.

Three, they used Path #4 mechanics inside Path #1 architecture. Preconfigured agent templates. Third-party partners building on top. The earliest hint of a real vertical agent ecosystem, but hosted inside a coherent AI platform.

Four, they exited via Path #5. Schneider bought Path #1 because they could not have built the substrate in time. Every serious industrial software vendor now has to react to that transaction.

AI does not rescue weak substrates.

It amplifies whoever already did the boring, hard work.

Cognite did the hard work first.

Then they ran the exact play the market is now paying for.

The market caught up second.

Do me a solid and forward to a friend :-)

SaaS has no inherent value in and of itself. It’s only valuable insofar as it helps a human get a job done. With AI, we now have technology that can do a lot of the job. So we’re moving to a world where the technology vendor isn’t selling a tool, they’re selling an outcome. Or as we say, you’re not selling the pole, you’re selling the fish.

Emergence’s Jake Saper on the token-maxing hangover, why AI-Native Services own the deepest data moat in the economy, and the uncomfortable question every SaaS founder should be asking this quarter…

The token-maxing hangover just kicked in.

Three months ago it was a badge of honor to brag about your monthly token bill. The bigger the number, the more “AI-native” you sounded on the earnings call. It turns out that token spend is a spectacularly bad proxy for value created. Boards figured that out roughly in unison, and the hangover is now shaping every meaningful vertical AI decision I’m watching this quarter: pricing, model choice, org design, even who you put on your board.

Sitting underneath the hangover are three forces colliding at the same time. Open source model quality has crept to within a handful of months of the frontier. Inference got so cheap that the same job now costs roughly a tenth of what it did a year ago. And services; 80% of GDP, the part of the economy the Valley loves to ignore, are quietly being rebuilt as software delivered by humans-in-the-loop. Any one of those would matter. Together they rearrange the vertical AI board.

To sort through it Nic spent an hour with Jake Saper, GP at Emergence Capital. Jake and the Emergence team called SaaS early, called vertical software early, bet on open source back in 2023 (Together AI, now at an $8.3B valuation), and coined the phrase “AI-Native Services.” He is the calmest voice I know on the topic and, refreshingly, the least invested in his own thesis being right in the shortest possible timeframe. What follows is his read on where value pools next, and the playbooks I’d steal as a founder, operator, and investor.

Watch the Episode

This Weeks Vertical Titan:
Jake Saper (Partner @ Emergence)

Jake has been at Emergence for twelve years. Board seats include Ironclad. Early checks into Zoom, Veeva, and a portfolio that reads like a decade of correct calls on business-model change. Emergence has a habit of naming the model before the market does — “SaaS” (originally “technology-enabled BPO,” which mercifully did not stick), then vertical software, and now AI-Native Services. When Jake tells you the frontier gap is closing and the money is moving downstream, it is worth pausing.

Start with the shape of the model layer, because everything else is downstream of it. In 2023 the closed labs looked untouchable. By mid-2026 the picture is very different — the most recent open-source releases are a handful of months behind the frontier, they are free to fork, and the weights are open, which means you can actually fine-tune them on your proprietary data. That last point is doing quiet violence to the moat story the frontier labs sold their investors.

The behavioral shift Jake keeps seeing inside enterprise buyers is a multi-model posture. Frontier labs get reserved for the highest-intensity, judgment-heavy calls. The majority of runs, the boring 80%, get routed to a fine-tuned open-source model at a fraction of the cost. Harvey, Legora, and the other well-funded agentic companies are pouring engineering into fine-tuning their own open-source weights precisely so they can separate their fate from any one lab’s pricing page.

“When intelligence gets this cheap, the model becomes less of a moat. The interesting question is where value actually goes.”

Which is where the model labs’ recent behavior starts to look revealing. Anthropic and OpenAI are hiring vertical leaders iw. Ironclad’s Jason Boehmig went to run legal at OpenAI and is quietly shipping vertical-flavored products. Some will land. Most, Jake suspects, will feel like straw-grasping. Building “Claude for Science” is not the same as spending five years learning why a pharma researcher won’t switch off her current workflow. The lab has near-infinite capital and can absolutely hire the specialists. But specialization is a different company, with a different cap table and a different product culture, and history has not been kind to the mothership that tries to run it in-house.

The more interesting move is DeployCo — OpenAI’s spinout that uses FTEs plus AI to help enterprises actually deploy the models. Emergence invested. The read is that deployment work is directly adjacent to selling tokens, so the strategic logic holds. On a meta level, DeployCo may itself become one of the largest AI-Native Services businesses in the world, using AI to deploy AI. There is a joke in there about recursion and a serious point about where the durable margin lives.

The reason enterprise buyers are wary of handing everything to the labs is not paranoia, it is math. If you’re Pfizer and you give OpenAI all your molecular data, you have handed the party that is now shipping vertical products your defensibility. Fine-tuning an open-source model on your own infrastructure looks a lot more sensible. The frontier labs, in trying to expand their surface area, are the reason their best customers now want an exit.

Agentic Software vs. AI-Native Services

Jake thinks an AI-Native Services business (one that owns the full outcome end-to-end), has a better data moat than an equivalent agentic software company. The logic is if you sell software into a fund administrator, your view of the transaction is truncated. You see the piece your product touched. You do not see how it was reviewed, whether the customer accepted it, or what happened three months later in the audit. If you are the fund administrator, every delivered output generates an evaluation event, a human sign-off, and eventually a defended audit. That is a labeled dataset money cannot easily buy.

The framework Jake and his partners use inside Emergence to pick which services to chase is a simple two-by-two. Core versus non-core to the customer. Critical versus non-critical. Pfizer will never let you own drug discovery, it is both core and critical, and they will use open source to keep it in-house. Pfizer will absolutely let you own accounting, it is critical but not core, and nobody at Pfizer wakes up wanting to run an accounting department. That single lower-right box is where the most durable AI-Native Services businesses get built.

Quick PSA -
90 Days Until The Vertical Software Summit

DON’T WAIT. We are already 75% sold out.

If you want to be at the Vertical AI Event of the year where ~400 vertical software founders, operators, and investors come to gather, snag your ticket now.

The speaker line-up is a 10/10 and the entire investment community shows up.

Grab Your Ticket Here

We had to turn away 100+ folks last year who waited until the last few weeks to grab their ticket.
Don’t be that person who misses out!

Jake’s moves for building an AI-native business the labs can’t out-scale.

#01. Diversify the model spend — frontier for the 5%, open source for the rest

Every serious deployment Jake sees now — enterprise or AI-native startup — is running a multi-model architecture. The closed frontier labs handle the small share of jobs that genuinely require the top of the intelligence curve. The vast majority of runs go to cheaper, open-source models. The gap between the frontier and the leading open weights has collapsed from twelve-plus months to a handful of months, and it keeps shrinking. That changes the math on every token budget in the industry.

The infrastructure winners of this shift are already visible: Together AI, Base 10, the neoclouds serving fine-tuned open weights. The losers are anyone whose business model requires the frontier lead to stay wide. Enterprise buyers are also newly wary of pouring proprietary IP into a single lab — especially now that those labs are hiring vertical leads and launching competing apps.

#02. The frontier labs are moving into the app layer: plan for it, don’t pretend it away

Both OpenAI and Anthropic are hiring vertical leads. Jason Boehmig, Ironclad’s co-founder, just took over legal at OpenAI. Anthropic is shipping Claude for Science. The labs are also spinning up services arms — DeployCo at OpenAI, an Anthropic solutions group — to help enterprises adopt. Jake’s read: some of these vertical bets will land, most will feel like straw-grasping because the labs don’t have the workflow depth vertical incumbents spent a decade building. But you cannot assume they will fail.

#03. Beware mirage product-market fit: humans dressed up as AI is still a services shop

The single biggest risk Jake sees in the current AI-native services wave is what he calls mirage PMF. Revenue is growing fast, retention is great, customers love you — but the service is still being delivered primarily by humans. Congratulations: you built a services business and raised venture capital against it. The only real PMF signal for an AI-native services company is that the AI is performing an increasing share of the work over time, with margins to match.

A quick word from our sponsor on the Verticals podcast, Parafin. Fresh off the 2026 Forbes Fintech 50 and a new credit facility led by Goldman Sachs, they’ve now extended over $35B in offers to small businesses through platforms like DoorDash, Gusto, and TikTok Shop.

Purpose-built embedded capital for vertical platforms.

Check Out Parafin

#04. Borrow credibility

Selling into a judgment-heavy service — legal, financial, medical — is a brand game as much as a product game. Nobody got fired for hiring Goldman. Nobody got fired for hiring McKinsey. The AI-native founder rarely walks in with that pedigree, which is why Jake talks about borrowing credibility. The cheapest form is hiring — put a grey-haired industry veteran in a customer-facing role, not on an advisory board where the customer knows they’re a logo. The most interesting form is frenemy partnerships with legacy vendors: they have brand and manpower, you have AI and speed, you rev-share until the relationship outgrows itself.

#05. Own the full lifecycle: the eval data is the moat

The reason Jake now argues an AI-native services company can out-moat a pure agentic software vendor: view of the transaction. A software vendor selling into a fund administrator sees a slice — the piece of the workflow their product touches. The services vendor owns the whole lifecycle. Every deliverable gets checked by a human, sent to the customer, defended in an audit. Every one of those checkpoints is eval data on the AI you built.

#06. Your competitive set is sleepier than you think

The best story from the interview was the founder who pivoted from selling agentic software into investment banks to becoming a mid-market investment bank. His two reflections were priceless. First, once he started using his own product to deliver real client work, his immediate emotion was embarrassment. You never understand a workflow until you’re the one being paid to run it

Software founders overweight the AI arms race and underweight how slowly the actual industries they’re attacking move.

Sell the fish.
Diversify the stack. Own the outcome nobody else will warranty.

If you’re building in the AI decade. and especially if you’ve raised venture capital against an AI-native services thesis — a few key areas separate the founders who compound from the ones who become case studies:

1. Sell the outcome, not the tool. Warranty the job end to end. The eval loop that comes back from owning delivery is the moat a pure software vendor never sees.

2. Run a multi-model stack from day one. Frontier for the small share of runs that require it. Fine-tuned open weights for the rest. Independence from any single lab is a feature you actively sell to enterprise buyers.

3. Hunt critical-but-non-core. Never the customer’s identity, always their cost center. Accounting at Pfizer, not the drug pipeline. MSPs, customs brokerage, paralegal work, licensed niches with real regulatory hurdles.

4. Refuse the mirage. Track percent of work performed by AI as a first-class metric. If the line isn’t trending up each quarter, you didn’t find PMF, you found a services shop with the wrong cap table.

The founders who quietly own the next decade won’t be the ones with the loudest frontier-lab partnership. They’ll be the ones who built McKinsey and Stripe out of the same team, warrantied the outcome, and turned every audit into a labeled data row on a model they own. Be that founder, and the frontier lead compressing works for you, not against you.

— Luke Sophinos

See you next Wednesday. If this one landed, forward it to someone still bragging about their token bill :-)

Watch the Episode

Thanks for reading LINEAR. I reply to every email…
Have any questions, feedback, comments? Let us know, we work for you!

Built from firsthand experience growing and scaling more than 20 vertical SaaS platforms, this article explores what high-performing software companies do differently to increase merchant adoption, convert existing customers, and create long-term enterprise value. Read the article here.

AI Roll Ups, Two Years Later

Two years ago, this model didn’t have a name.

I wrote about it as “venture buyouts.” Slow Ventures called them “GBOs.” Matt Brown had a taxonomy. Nobody knew what to call the thing.

Now the market has settled the naming question. They’re called AI Roll-Ups. And in the last twelve months they’ve gone from a fringe experiment to arguably the most-funded NEW category in venture (outside of pure AI plays of course).

The original thesis was simple: buy operating businesses in fragmented, labor-heavy industries, layer proprietary AI on top, re-rate the margin profile from services (5-10% EBITDA) to software (30-40% EBITDA), and compound at venture scale.

That’s still the thesis.
What’s changed is that it’s now real.
The names on the board are no longer hypothetical.

The current landscape:

• Long Lake (HOA management, General Catalyst): $670M raised, 18 acquisitions, reportedly ~$100M EBITDA in under two years.

• Crescendo (contact centers, General Catalyst): $500M valuation after acquiring PartnerHero. Charging per resolved ticket, not per hour. 4x margin advantage over traditional BPO.

• Titan MSP (IT services, General Catalyst): $74M and the acquisition of RFA, an MSP with 400+ financial services clients.

• Dwelly (UK letting agencies, General Catalyst): $93M. Maintenance resolution times cut from 50 days to 20. EBITDA margins doubled where fully deployed.

• Accrual (accounting, General Catalyst): $75M to automate tax prep.

• Shield (IT services, Thrive Holdings): $100M from Thrive, led by ex-Palantir CIO Jim Siders.

• Metropolis (parking, Slow / 3L / Eldridge): $1.7B Series D, $1.5B acquisition of SP+. The original proof point.

• Cabana (pool services, Asymmetric): The capital-efficient model. ~$10M raised, $12-15M revenue, $4M+ EBITDA. Buy first, build second.

• Sequence Holdings (cross-industry, 8VC): Stealth. Scale, Cognition, and Lone Pine alumni.

• And more!

And now AGI in insurance distribution. More on that below.

What the last twelve months actually taught us:

1. The math is not universal. Long Lake and Crescendo are the poster children. But you don’t get to reference-check a category on its two best names. Renovo Home Partners collapsed. Multiple other bets are quietly stalling. The winners are winning very publicly. The losers are being repackaged as “portfolio adjustments.”

2. Distribution is the moat, not the model. Anyone with capital can copy the playbook. What they can’t copy is a founder who spent 15 years in pool services, an operator who scaled Keystone to $5B, or a client roster that took decades to build. The AI isn’t the moat. The base is.

3. Speed is the killer. Every single failure mode I’ve seen in this category traces back to acquisition pace outrunning integration capacity. Buying is easy. Integrating six businesses in twelve months while also building an AI platform while also running an OpCo is where companies die.

4. The PE-VC handshake is happening. This is the most underrated shift. Two years ago, PE and VC firms only interfaced when one was selling to the other. Now Atomic is co-investing with Rockbridge on AGI. General Catalyst is running a private equity motion inside a venture fund. The wall is coming down.

5. Not every “AI roll-up” is one. A lot of what gets labeled AI roll-up today is a traditional PE roll-up with a ChatGPT wrapper on the pitch deck. The real ones share three traits: proprietary AI built before/during the first acquisition, a services vertical where AI can do 30-70% of the work, and an operator who has actually run something in the industry.

The uncomfortable observation:

The category is now crowded enough that pattern-matching alone won’t work. General Catalyst’s $1.5B allocation, Thrive’s $1B vehicle, Khosla, Lightspeed, 8VC, Slow, Atomic, and a dozen more are all deploying against variations of the same idea. The alpha is no longer in seeing the model. It’s in picking the right vertical, the right operator, and the right structural setup.

The best of these will look like software companies in a decade. Most will look like PE-owned services businesses with better dashboards.

The interesting question is not whether AI roll-ups work. They clearly do. The question is whether they can become venture-scale. A lot of them will end up as PE exits. But a few will compound into public company machines.

That’s the actual bet.

Excited to unveil AGI today…

American Growth Insurance that is 😃

After stepping away from Portico (FKA known as CourseKey) after our sale to Private Equity, I became obsessed with AI-Powered Roll Ups.

Jack Abraham, Chester Ng, Michael Stenlick, and Brandon Zhang of Atomic all shared the excitement and we began a journey of analyzing industries where we felt AI could transform the underlying business and create the opportunity for a venture-scale outcome.

We spent nearly a year analyzing 1,000+ industries and settled on one…

Insurance Distribution.

The surprising thing about insurance brokerages was that, unlike other industries, for every single front-office person (the ones actually interfacing with the customers) there was an army of back-office individuals doing the dirty work to support them. I had never seen the level of ratios we saw here. We asked the question, “What if AI could do the dirty work? What if AI could move the back office headcount into the front office so that nearly EVERY employee was spending time with customers?” That seemed like an important and exciting opportunity.

We rolled up our sleeves and built an AI platform to prove the thesis. We studied every workflow, we automated as much as we could, and we launched tooling at ~10 different agencies across the United States. The results were incredible. Folks were spending SO MUCH more time with the customer and all the fundamental business metrics were substantially improving as a result.

We did not know insurance and we did not know M&A. We knew technology. So we had to go out and find the perfect partners to pull this off.

I feel so fortunate to say that we did that.

After talking to hundreds of prospective Founder/CEO’s we met Brian Morgan. Brian has spent his entire career building and scaling companies in the insurance distribution space. Most recently, he saw an incredible ramp up at Keystone Agency Partners as the Chief Revenue Officer. This business went from $0 to $5B in enterprise value in <5 years. It’s a beautiful case study in the classic roll-up playbook in the insurance market. Brian saw the opportunity for a new type of Roll Up. One powered by AI.

Brian connected us with Rockbridge Growth Equity, a Private Equity firm founded by Dan Gilbert of Rocket Mortgage. Tony Pulice, Partner at RGE, blew us away. His humility, his acumen, his maniacal focus on strong values, and to boot he had already scaled and exited an incredible insurance brokerage, HighStreet, and was the key board and investment partner there as it got off the ground and scaled. Tony was focused on finding a company that met the same thesis, something that was different from the typical insurance agency roll-up. It was a match made in heaven!

So today, we’re so excited to come out of stealth, after ~2 years of hard work, with a $70M round backed by Rockbridge Growth Equity and Atomic. I believe this is a first-of-its-kind partnership between leading Private Equity and Venture Capital firms. These two groups typically don’t interface unless they are selling a company to one other. But combining PE’s M&A acumen with VC’s AI focus, and most importantly, a dream team of operators and executives, is so damn exciting…

AGI is acquiring agencies across the country and deploying incredible AI technology to substantially improve the customer experience, which in turn greatly improves the underlying business. We’re hiring across a multitude of roles.

I’m honored to serve as a Co-Founder, Advisor, and Board Member of AGI and look forward to supporting the team build the future of this industry.

Thanks to Axios for dropping the news.

Can read the full announcement at the link here.

Onwards!

How To Found an AI Roll-Up

A ton of folks want to run this playbook right now. Most people are going to run it wrong. Here’s the actual sequence, based on two years of doing it.

1. Find the right industry.

This is 80% of the early work. Get it wrong and nothing downstream matters.

You are looking for the intersection of three things:

• AI can significantly move the needle across the entire business, not one or two functions. If AI only helps the sales team, that’s a feature, not a company. You need a vertical where automation can compress cost structure across the whole P&L. Front office, back office, ops, finance, everywhere.

• Classic PE roll-ups already work here. If PE has been quietly making money in the space for a decade, the fragmentation is real, the acquisition math is proven, and the debt markets understand the assets. If PE hasn’t touched it, ask yourself why.

• No venture-backed AI roll-up exists in the vertical yet. You want to be early. There are so many of these in IT. Tough to raise capital when so many are going after the same treasure.

Also test for venture scale. Can this actually get to $1B+ in enterprise value in a reasonable timeframe? A lot of these industries can produce beautiful PE-style outcomes but not venture outcomes. Be honest with yourself about the ceiling before you start.

We looked at 1,000+ industries before we picked insurance distribution. That was not a marketing number. That was the work. And it took a year.

If you made it thus far, go check out the 2026 Vertical Software Summit. We’ll have 400+ vertical founders/operators/investors in Miami in November. Two days,

6+ billion dollar vertical Founders/CEOs.
The Vertical AI event of the year.

Snag Your Ticket Here

2. Build the product first.

Slow Ventures has been the loudest on this, and they’re right. Prove the thesis with technology before you deploy a dollar into M&A.

This does two things. It forces you to actually understand the workflow, not just theorize about it. And it gives you something concrete to show investors, operators, and eventually acquisition targets.

The trap here is building the product in isolation. You need real operators giving you real feedback on real workflows. Find 3-5 friendly companies, cozy up to them, throw them some equity or free tooling, and embed. Watch the actual work. Do not build from a whiteboard.

3. Deploy it in the wild and prove it moves the numbers.

The product working in a demo means nothing. The product working inside 5-10 live operators with measurable improvements to fundamental business metrics is the entire ballgame.

We deployed at ~10 different agencies before we bought a single one. The results told us the thesis was real. If the results had been mediocre, we would have gone home. That’s the whole point of this step.

Pick 2-3 metrics that actually matter to the business (revenue per employee, time-with-customer, retention, close rate, whatever the vertical cares about). If AI doesn’t materially move those, you don’t have a roll-up. You have a services company waiting to happen.

4. Buy your first company and prove the thesis on your own P&L.

There’s an infinite difference between “the software works at a customer” and “we own the P&L and the numbers moved.” VCs know this. Operators know this. You need to know this.

The first acquisition is a proof point, not a scaling event. Pick something manageable. Deploy your tech. Show that on your own books, the thesis holds. This is the artifact you use to raise the real capital.

A few things I’d think about as well:

• Get as much seller financing as humanly possible. Debt is dangerous. I’ve watched companies collapse on this exact model because they leveraged up too fast.

• Structure the deal so the seller stays involved. Local relationships and institutional knowledge don’t transfer through a data room.

• Don’t overpay. You are trying to prove the model, not win a bidding war.

5. Recruit team and investors from inside the industry.

This is what I’m seeing most tech founders ignore. You will not out-execute a category without operators who have already scaled in it.

For AGI, we knew tech but we did not know insurance and we did not know M&A. So we went and found Brian Morgan, who had scaled Keystone as CRO (They went from from $0 to $5B in enterprise value in ~5 years). We went and found Tony Pulice at Rockbridge, who had scaled and exited HighStreet.

Same rule applies to your investors. You need long-term patient capital. You need partners who have run M&A at velocity. You need people who understand both the venture side and the buyout side.

The magic in AGI was pairing Atomic (venture, AI expertise) with Rockbridge (PE, insurance M&A acumen). That combination is very hard to fake. It also barely existed as a structure two years ago. It’s a very unique model that was complex to put together.

6. Scale like crazy.

Now you run the playbook. This is where operational discipline separates the winners from the ones you’ll never hear about again.

Things to watch:

• Integration capacity is the ceiling on acquisition pace. Every failure I’ve seen in this model comes back to this. Do not buy faster than you can integrate. If integration takes 6 months and you’re closing every 3, you’re building a landmine.

• Standardize the operating system before you scale the M&A. Common playbook, common tech, common metrics across every acquisition. Otherwise you own a portfolio, not a platform.

• Keep local brand equity where it matters. Especially in relationship-driven industries like insurance. The customer doesn’t want to know they got acquired.

• Track the AI margin impact obsessively. Your entire venture investment thesis rests on the re-rate. If margins aren’t moving with deployment, something is broken and you need to know immediately.

• Stay disciplined on debt. Just be careful with debt. It can take out your business. Don’t over-lever!

One more thing worth saying out loud. This is not a playbook for the faint of heart. You are running three companies at once: an operating business, an M&A shop, and a vertical AI startup. Each of those is a full-time job. Most people cannot do it. That’s exactly why the returns are available.

If you can do it, and you picked the right vertical, the outcome is enormous.

If you can’t, the wreckage is painful…

Tread wisely !

Do me a solid and forward to a friend :-)

Compliance has always been one of the most misunderstood wedges in vertical software. Everyone talks about it. Almost no one does it well. And in healthcare, the graveyard of AI point solutions that tried to bolt compliance onto a scribe or an RCM tool is already pretty crowded.

Which is why Zach Rosen and Brellium are worth paying close attention to.

Brellium sits in the exact middle of the healthcare revenue cycle — between the scribe and the biller — and quietly does something almost no one else is doing: turning payer compliance from a reactive, audit-driven fire drill into a real-time training loop that upgrades the entire clinical workforce. It’s a business built on a subtle but important insight: compliance in ambulatory healthcare isn’t a checklist. It’s a fractal. And the deeper you zoom in, the more non-deterministic it becomes — which is exactly where AI, network effects, and vertical depth start to compound into a real moat.

This week’s conversation with Zach is a masterclass in how to think about wedges, network effects, two-sided go-to-market, outcome-based pricing (and why it’s often a bad idea), and what it actually takes to build a category-defining vertical AI business inside one of the most closed ecosystems in tech.

Let’s get into it…

Watch Episode Here

This Weeks Vertical Titan:
Zach Rosen (Co-Founder & CEO @ Brellium)

Zach Rosen is the co-founder and CEO of Brellium, an AI-powered clinical compliance platform serving ambulatory healthcare providers across all 50 states and multiple lines of care.

Zach didn’t grow up dreaming about compliance. He grew up dreaming about sports, and later, about investment banking — a career path he says he pursued “for reasons I’m not sure I fully understood at the time.” He figured out pretty quickly that finance wasn’t for him, and while at school in the Bay Area, he watched peers his age build wildly impactful companies. That environment reframed what was possible, and he took what he calls “an exact 50/50” — if finance was the wrong path, building a business was the opposite. That was the extent of the strategic reasoning at the time.

His first company was Sapient, which he built with a few classmates and ran for a little over two years. Sapient was, by his own account, a good business. But the real inflection point wasn’t professional — it was personal.

A routine primary care visit produced a core medical misdiagnosis. A few tests should have been run that weren’t. What followed were “a couple of really painful years.” The core thesis of Brellium was born directly out of that experience: build a safety net on patient care so what happened to Zach doesn’t happen to anyone else, ever again.

That’s founder-market fit you can’t fake.

Today, Brellium screens patient visits in real time against 1,000+ clinical, payer, and OIG standards — making sure care is rendered medically necessary and documented compliantly — across specialties ranging from ABA therapy to mental health to home health to hospice. They sit in the middle of the revenue cycle, integrated with scribes and intake tools on the front end and RCM/billing on the back end. They serve providers on one side and payers on the other. And they’ve built the business over the last five years, raising $16.7M to scale into what is quietly becoming one of the more strategically well-positioned vertical AI platforms in healthcare.

To see why the wedge lands where it does, start with the shape of the market. Ambulatory care is roughly half of US healthcare spend, which is itself roughly a fifth of GDP. Inside that, Brellium concentrated on the specialties where the rules are most subjective — mental health, ABA, home health, hospice. That is where words like “individualized” and “medically necessary” become matters of interpretation, and where payers spend the most energy building private enforcement patterns that public healthcare bulletins never mention.

What Brellium sells is compliance. What Brellium does is training. A mental-health note has a laundry list of requirements — individualized language, history of present illness, medical necessity — and any miss can trigger a payer months later to pull the money back. Brellium catches those misses before the claim is filed, and that is the sale. But the reason customers renew is the Friday summary. It tells each clinician what they did well and where they need to improve, with trend lines that feed bonus tiers and promotion tracks. The product is a training system dressed as a compliance product, which is exactly why it sticks.

“At the surface we’re a compliance product. That is why customers buy. But the first-order value is that we’re a training product.”

Skeptics push back on compliance as a moat with one line — regulations are public. If the rule is a checklist of five things, a decent AI can score against it and so can everyone else.

The market Brellium is walking toward is two-sided. Providers on one side, payers on the other, and the interesting question was always which side to start with.

They picked providers, because mid-market providers move faster and payers need proof points before they will engage.

Once a payer looked at its ABA book, saw dollars leaking, and then noticed a clean correlation between Brellium-covered providers and lower risk, the incentive to lean in became rational and immediate.

The loop now closes on itself ; every enforcement event Brellium sees on one side of the network compounds into every other customer’s protection.

None of this arrived on schedule. Brellium is five years old, and the first two and a half were pivots. A lot of smart people gave a lot of confident advice, and almost none of it moved the business.

What eventually moved it was a very specific loop; build, put in front of a customer, ask them to pay, iterate on their answer, ignore everyone else. That is a boring answer. It is also the answer.

Watch Episode Here

What to steal from the Brellium story

The Brellium arc is specific to ambulatory healthcare, but the operating principles underneath it are not. Strip out the specialty codes and payer names and four transferable ideas sit there, ready for a founder in legal, construction, insurance, logistics, manufacturing, or any other vertical to pick up.

#01. Pick the path before you pick the wedge

There are only two ways to build a durable vertical software company, and AI has not changed the fork. It has only sped up each side.

Path A is the classic — compete head-to-head with the system of record, ship a wedge, bolt on adjacent modules, race to become the SoR. Ninety percent of vertical SaaS runs it.

Path B is harder — build a wedge so sharp the incumbent SoR cannot plausibly compete, and pair it with a network built from data the SoR simply cannot see. Gong did it. Brellium is doing it. The honest note on Path B in 2026 is that the wedge is easier than ever, but the network effect is marginally harder — more competition for the same data-in-motion, more founders defaulting to the easier path.

If you sign up for B, you also sign up to be a better leadership team than the A operator down the street. The Path B playbook is not clear-cut.

#02. Run the wedge through the quadrant

Before committing a team to Path B, run the quadrant. Deterministic, public rules make a wedge, not a network. Subjective rules plus private enforcement produce the data asset the system of record cannot see.

If your category lands anywhere except the bottom-right corner, Path A is the honest choice and platform expansion — not moat construction — is the whole strategy. Founders lie to themselves about this constantly and it costs them two years.

#03. Sequence the flywheel from the shorter cycle

Two-sided networks fail when founders chase both sides in parallel. Start with the side that has a shorter cycle and lower burden of proof. Stack proof points there. Then walk into the other side’s boardroom with data, not a pitch.

The same logic applies to integrations — the 800-pound gorilla platform has no rational reason to partner, so spend integration energy on the long tail where the customers actually are. Brellium is integrated with essentially every EMR in ABA, including Central Reach, precisely because it did not wait for permission from the biggest one.

A quick word from our sponsor on the Verticals podcast, Parafin. Fresh off the 2026 Forbes Fintech 50 and a new credit facility led by Goldman Sachs, they’ve now extended over $35B in offers to small businesses through platforms like DoorDash, Gusto, and TikTok Shop.

Purpose-built embedded capital for vertical platforms.

Check Out Parafin

#04. Price for what the customer actually feels

Outcome-based pricing is the phrase of the year. For risk-mitigation businesses it is usually the wrong hammer. The “outcome” is the absence of a rare, delayed black swan, which makes the incentives conflict with the customer’s interest.

Brellium sells usage instead — a per-patient-visit rate rolled into a recurring subscription, with a protection premium baked in.

Predictable for the buyer, aligned with the work, and honest about what the product does. A quiet contrarian take in a market that keeps repeating “outcome-based” as if it were always the answer.

One more thing —
Where Zach is bullish, bearish, and contrarian

We closed the conversation with a rapid-fire scorecard. Most of it will not surprise you. The last row will.

AI-enabled American manufacturing was fashionable two years ago, then quietly went cold as capital rotated into humanoids.

Zach’s read is that this is precisely the moment to look at it — a large market, strategic importance to the West, and very few of the best builders paying attention.

If you have been reading vertical AI takes for the last six months and wondering where the uncrowded lane is, that is a serious pointer.

The wedge that becomes a network wins.
The chat panel bolted onto a legacy SoR does not. Everything else is variations on that theme.

Three lines to save before you close the tab.

Compliance is a moat only when it is blurry; deterministic, public rules produce a wedge, not a network.

Two-sided flywheels start on the side with the shorter sales cycle, then let the data pull the other side in.

Outcome-based pricing is the wrong instrument for risk-mitigation businesses; usage with a protection premium is more honest, more aligned, and more predictable than the phrase everyone is repeating on LinkedIn.

See you next week. If this one landed, forward it to the founder in your inbox who keeps calling their checklist a moat.

Watch Episode Here

Thanks for reading LINEAR. I reply to every email…
Have any questions, feedback, comments? Let us know, we work for you!

Today’s newsletter is sponsored by Xplor Pay. Launching embedded payments is a major milestone, but it’s only the beginning of the opportunity. Many software companies see merchant adoption plateau after launch, not because of their technology, but because they stop investing in their payments business.

Built from firsthand experience growing and scaling more than 20 vertical SaaS platforms, this article explores what high-performing software companies do differently to increase merchant adoption, convert existing customers, and create long-term enterprise value. Read the article here.

Alright, let’s get to it…

The Rule of 40

I’ve become more of a fan of the rule of 40 recently.
I just think it’s a good north star metric for most companies.

It was popularized by Brad Feld around 2015 and quickly became the single most-cited operating benchmark in public SaaS. The appeal is its honesty: it forces a company to declare a balance between growth and efficiency rather than hide behind either one.

If your not familiar with the rule of 40, it’s pretty simple.

Annual ARR growth %
+
Ebitda margin %

=

Rule of 40

So if you grow revenue 20% YoY and you have a 10% Ebitda margin,
you’re rule of 40 would be 30. The general premise is that a healthy software business’s revenue growth rate plus profit margin should equal or exceed 40%.

Grow 40% at break-even? You pass.
Grow 10% at 30% EBITDA margins? You pass.
Grow 25% at 15% margins? You pass.

It also constrains a tempting failure mode: buying growth that doesn’t survive contact with a cost-of-capital regime. If you can’t get to 40 by combining the two, you’re either not growing fast enough to justify the burn, or not efficient enough to justify the slowdown.

For boards, it’s a clean conversation. For founders, it’s a forcing function:
How much margin am I willing to trade for this next dollar of growth?

The Rule of 40 has fallen out of love recently due to the explosive growth of so many of the AI companies, who are blitzscaling at the moment and not giving a sh*t about EBITDA margin. But all things eventually revert :-)

Why it still matters for the rest of us

Here’s the catch: very few companies are actually growing 200%+ on a meaningful base. The median vertical software company is growing 15–25% into a defined TAM with a known sales motion. For that company — the modal company — the Rule of 40 is still the cleanest way to ask “are we worth funding?”

Vertical software is different from horizontal in a few structural ways that make R40 more relevant, not less:

• TAMs are bounded. You can’t out-grow gravity for long. Eventually penetration math catches you.

• Customers are sticky and slow. Net retention is high but new logo growth is bounded by the pace of an industry’s willingness to adopt.

• Capital efficiency compounds. Because growth is steadier, every point of margin you protect early gets re-invested at known unit economics.

Translation: if you’re a vertical software founder, the Rule of 40 is still the benchmark you’ll be measured against — on the way in by growth investors, and on the way out by strategics and PE.

Public Vertical SaaS Analysis: The Rule of 40 Tells Us A Lot Of Biz Stories About Efficiency and Growth

I analyzed public vertical software companies to get a sense of where there Rule of 40 stands. The results, were, well, suprising…

Below are 20 public vertical software companies ranked by Rule of 40, using each company’s preferred margin metric (adj. EBITDA or non-GAAP operating margin).

Waystar, Veeva, and Privia Health sit at the top — all healthcare-adjacent, all proving that deep vertical workflow lock-in supports both growth and margin. Note the dashed 40% line: nine of these twenty companies clear it, which is better than many horizontal SaaS baskets. The bottom of the list (Definitive Healthcare, Health Catalyst) shows what happens when a vertical data or analytics company’s growth stalls before costs normalize: you’re left with decent margins but a shrinking top line.

This is the most honest view. The dashed diagonal is the Rule of 40 frontier — anything above it passes. Notice the two tribes: margin engines in the upper-left (Veeva at 45% margin, 16% growth; MeridianLink at 45% margin, 8% growth) and growth engines in the lower-right (Shopify at 34% growth, 12% margin; Alkami at 29% growth, 18% margin). Bubble size maps to revenue scale, which is why Shopify's dot is enormous despite modest margins. The danger zone is the muddy middle near the origin: companies that are neither growing fast nor printing cash. That is where multiples get compressed.

Shopify and Alkami are the only companies on this list still growing near 30% year-over-year at public scale. That is hard to do in vertical software. Most of the group is compounding at 10–25%, which is perfectly healthy for a business with 110%+ net revenue retention. What should worry founders: if your vertical SaaS is growing sub-15% and you’re not yet at $100M ARR, you may be under-penetrating the market rather than maturing into it. Growth is a choice until it isn’t.

Doximity, Veeva, and MeridianLink all post 45%+ margins. The common thread? Low customer acquisition cost, high net retention, and pricing power inside a regulated vertical. If you’re building in healthcare, financial services, or another compliance-heavy industry, your margin ceiling is structurally higher than consumer-adjacent verticals like restaurants or hospitality. Toast and Shopify sit at 11–12% margins not because they are poorly run, but because their markets are volume-driven and competitive. Know which game you’re playing.

Revenue scale is not destiny, but it changes the conversation. Shopify ($11.6B), Toast ($6.2B), and Veeva ($3.2B) operate at a scale where even modest margin improvement moves hundreds of millions of dollars in free cash flow. For founders under $100M ARR, the lesson is different: don’t optimize for revenue scale until you have durable unit economics. A $300M vertical SaaS business with 45% margins (MeridianLink) is arguably a better PE asset than a $6B one with 11% margins (Toast) — depending on the buyer’s thesis and leverage capacity.

Across all five views, the pattern is consistent: the best vertical software companies find a way to be good at one axis and at least acceptable on the other. Mediocre on both is what kills valuations.

How To Internalize Rule of 40 Based On Where You/Your Company Is Actually At

The Rule of 40 is not a constant. It’s a target that should evolve with your ARR and the kind of capital you’re raising. Below is a practical guide.

$0 – $1M ARR: Don’t even compute it.

Venture track: Growth should be above 3x year-over-year. Margin is basically irrelevant at this stage. You’re validating product–market fit and proving that someone will pay.

Bootstrap track: Get to break-even as fast as possible. You don’t have a venture cushion, so every dollar of burn comes out of your own pocket or a small friends-and-family raise. Target 2x+ growth, but the real goal is proving customers stay and pay you profitably.

PE track: This is not a PE-fundable stage. Get to $5M+ ARR before you even think about it.

Watch out: Founders who “optimize for profitability” here are usually rationalizing a slow sales motion. Burn is fine; aimless burn is not.

$1M – $5M ARR: Prove the motion, not the margin.

Venture track: Target growth of 150% or more. Your Rule of 40 will be deeply negative — something like 150% growth and -100% margin — and that is exactly what investors expect.

Bootstrap track: Your R40 should be turning positive. Aim for 50%+ growth and at least break-even margins — ideally 20% growth plus 10% margin for an R40 of 30. If you’re still burning at this stage, you need a very clear path to profitability because there’s no outside capital to refill the tank.

PE track: Probably still too early for traditional PE. Growth equity may start engaging if growth is above 80% and your burn multiple is under 2.

Watch out: Beware of product-market-fit theater. High growth on a tiny base often collapses the moment you turn on a real sales team. Make sure the motion is repeatable. Not to investors, to yourself!!!

$5M – $10M ARR: First real R40 conversation.

Venture track: Aim for a Rule of 40 between 60 and 80, with growth dominant — for example, 80% growth and -20% margin.

Bootstrap track: This is where bootstrapped companies can really shine. You should ideally be at or above R40 with both growth and margin contributing — think 30% growth and 15% margin for an R40 of 45. You have no board demanding hypergrowth, so compound steadily and reinvest what you earn. The discipline of self-funding usually produces better unit economics than venture-backed peers at this scale.

PE track: This is growth-equity territory. Investors want to see R40 at or above 40 with a believable path to breakeven within 24 months.

Watch out: Unit economics become non-negotiable here. It becomes a lot harder to sell the dream if it took you a ~5 or more years to get here.

$10M – $100M ARR: The R40 sweet spot — and where most companies fail it.

Venture track: You need R40 at or above 40 with growth still above 40%. Below $25M ARR, growth should be 60% or higher; above $50M, margin in most cases should start to be flipping positive.

Bootstrap track: You’ve built something durable. Target R40 above 40 with margins driving an increasing share of the score — 25% growth and 20% margin for an R40 of 55 is a beautiful place to be. The advantage of bootstrap at this stage is optionality: you can raise venture, sell to PE, or just keep compounding. Don’t give that up by reaching for growth you can’t fund internally. You are about to become very rich :-)

PE track: Prime PE zone. Target R40 above 40 with at least 15% adjusted EBITDA margin. Rule of 40 plus 20% EBITDA is what drives the highest multiples.

Watch out: The muddy middle — 20% growth and 0% margin for an R40 of 20 — gets re-rated hard. Either re-accelerate growth or take a knife to costs.

$100M+ ARR: Durability beats heroics.

Venture track: R40 should be 40 or higher with margin at 20% or more. You may be in public-company territory but regardless you will be benchmarked against names like public vSaaS comps like a Veeva or ServiceTitan.

Bootstrap track: You’re now one of the best-owned assets in software. Target R40 of 50+ with 25%+ margins and 25%+ growth. The playbook is simple: protect net revenue retention above 110%, expand existing accounts, and acquire only when you can pay cash. You built this without venture — don’t dilute it now unless the strategic logic is overwhelming.

PE track: Target R40 above 40 with EBITDA margin of 25% or more. Multi-year visibility on net revenue retention above 110% is what unlocks premium take-private multiples.

Watch out: Pure growth narratives stop working. The market pays for compounding free cash flow, not TAM slides.

To Wrap This Up…

The Rule of 40 hasn’t been disproven; it’s been temporarily drowned out by a category of companies whose growth math makes the rule trivially true.

For the other 99% of software founders — and most vertical software founders — it remains the most honest single number to manage to. The trick is knowing which side of the equation you should be solving for, and when.

Pick a side.
Defend it with numbers.
And don’t find yourself apologizing for margin in a year or two when capital is harder to come by than it is now.

If you made it thus far, go check out the 2026 Vertical Software Summit. We’ll have 400+ vertical founders/operators/investors in Miami in November. Two days, 6+ billion dollar vertical founders. The Vertical AI event of the year.

Check Out The Summit

Do me a solid and forward to a friend :-)

In 2011 Adam Harris and his co-founder Rich Castle looked at hospitality and saw nine fragmented systems where there should have been one. Fourteen years later, CloudBeds runs 20,000+ hotels in 157 markets, processes more than $20B a year, and will check in 57 million humans in the next six months. In an industry consumed by AI narrative, Adam has an unfashionable read on where the moat actually sits. “SaaS is not dead,” he says. “If anything our customers are spending more money on SaaS right now.”

The regime change he does see is system of record turning into system of action. Record-keeping was the price of admission; furnishing the next best decision — pricing, packaging, guest personalization — is the product. Studies show hoteliers know the rate change they should make 100% of the time and act on it too late 80% of the time. Awareness isn’t the bottleneck. Execution is. Software that only shows the data, is now, just a big dumb database.

The second shift is where defensibility actually lives. Frontier LLMs are giant custodial corpuses of the public internet, and Adam loves how practical they are — he built his own home-lab Jarvis on top of one. But they do not have his 4 billion data points of workflow-labeled hospitality behavior. That dataset, fine-tuned into a domain model and wrapped in a deterministic harness with confidence scores and source pointers, is the moat.

The third is pirate season. The seas are full of buccaneers — self-declared vertical AI experts who have never worked a night in hospitality, prompt-wrappers hallucinating cancellation policies into real legal exposure, cut-corner boarding parties flying a flag they haven’t earned…

Watch Episode Here

Listen To Episode Here

This Weeks Vertical Titan:
Adam Harris (Founder & CEO @ CloudBeds)

Adam grew up in San Diego, studied economics at Berkeley, did a stint on Wall Street, and then left to build. Fourteen years ago he and his co-founder Rich started CloudBeds with a bet that a hotelier shouldn’t need nine disparate systems to run a property. Today it’s the category leader — 20,000+ hotels across 157 markets, 550,000 people logging in daily, 57 million check-ins expected in the next six months, and more than $20 billion in transactions running through the platform every year.

The current chapter is turning CloudBeds from a system of record into a system of action. Signals, the company’s AI forecasting and revenue engine, sits on 4 billion data points and runs millions of permutations every hour to produce deterministic, source-linked recommendations. They’ve also built three of their own foundation data models: 1) the guest, 2) rate and availability, and 3) everything else.

Adam’s frame: the corpus of proprietary, longitudinal, labeled hospitality data is the treasure. Frontier LLMs are extraordinary general brains but they can’t tell a boutique property in Lisbon why to raise a Thursday rate for a specific event window. Two years reinventing CloudBeds’ data architecture and DevSecOps was the price of admission for everything they are starting to ship.

The line every incumbent operator should write on a wall: hospitality is one of two industries where you can piss off a guest at 11pm and they’re still with you at breakfast.

Watch Episode Here

Listen To Episode Here

The captain’s playbook — ten moves for guarding the gold in pirate season.

Operator-grade lessons from Adam Harris on why the dataset is the treasure, how to move from system of record to system of action, why deterministic harnesses beat probabilistic vibes when a boarding party is on the horizon.

#01. The treasure is the dataset, everyone else is a pirate plundering for it

Adam is blunt about the reason CloudBeds isn’t losing sleep over AI upstarts: he sits on 4 billion data points, runs millions of permutations on them every hour, will check in 57 million humans in the next six months and processes more than $20 billion in transactions a year across 157 markets. OpenAI and Anthropic have the world’s corpus of text. They do not have the corpus of what a hotelier actually did last Thursday at 6pm when a booking pattern shifted. That data — longitudinal, labeled by the workflow itself, jurisdictionally scoped — is not something a horizontal model can synthesize.

His framing of the current moment: pirate season. Everyone is racing to plunder the gold. LinkedIn is full of self-declared vertical AI experts who have never worked a night in the industry they claim to disrupt. That noise will burn itself out. The companies that emerge are the ones who spent the last decade quietly building a proprietary dataset — and the last two years re-architecting it for AI-era access patterns.

Action item: Audit your dataset the way an acquirer would. Is it longitudinal, labeled, jurisdiction-aware, and tied to real transactions? If not, that’s what you’re building for the next 24 months — before you touch a model.

#02. Stop being a system of record, become a system of action

Adam keeps returning to the same phrase: record-keeping is not the value anymore, the work you should be doing with the record is. He calls Salesforce &ldquo;a big dumb database&rdquo; in its default shape, useful only once you strap Apex and harnesses on top. CloudBeds spent years consolidating nine hospitality systems into one system of record, and the entire product strategy now is to convert that record into action — Signals forecasts demand, prices rooms, packages offers, and furnishes the next best move before the hotelier is late to it.

The evidence he cites: studies show hoteliers are not dumb — they see the rate change they should make — but 80% of the time they act too late. Awareness is not the bottleneck. Execution is. So the software job description flips from show me my data to & do the work on my behalf, with a confidence score and a source I can audit.

Action item: For each dashboard in your product, ask: what is the one action this data implies, and could we take it on the customer’s behalf with their approval? That’s your Q1 roadmap.

#03. Deterministic harness beats probabilistic vibes, build the guardrails first

Hospitality is one of two industries where you can wrong a customer at 11pm and they are still with you the next morning (the other is medicine). Probabilistic outputs are not acceptable when the answer is this room $100 or not&; needs to be $100 every time, and &ldquo;is availability real or not&rdquo; needs to prevent an overbooking. CloudBeds wraps its agents in a deterministic harness — three data models (people, rate/availability, and everything else) exposed via MCP servers and APIs, with every answer carrying a confidence score and a pointer to the source row that produced it.

Adam’s LinkedIn here-say: connect your hotel to this Claude skill and it’ll write god-tier forecasts is a lie. Claude will be right 96 times out of 100, and the other four will lose real money. Context windows expose the drift; the industry ignores the disclosure. The unlock isn’t more agents — it’s the harness underneath them.

Action item: Ship a confidence score and a source-of-truth link with every AI-generated answer in your product. If a customer can’t audit the reasoning, they can’t trust the action.

#04. Own three data models, then let everything else be a marketplace

CloudBeds refuses to build everything. Adam draws the moat around three foundation data models — the guest (a human), rate and availability, and the operational data around both — and lets a massive partner marketplace fill in the long tail. That focus is why they can talk about being the orchestration layer with a straight face: the three models are theirs, the workflows on top can be anyone’s, and the partners are motivated because the data they get back is worth more than the fee they pay to plug in.

This is also why Adam picks his integrations from a paranoid place. A single AI upstart’s cancellation-policy chatbot has already generated real legal case law by hallucinating refunds. When you’re the system of record for 20,000+ hotels, one careless partner can drag your brand into a lawsuit. He says this isn’t protectionism?but instead adult supervision of a category that hasn’t grown into its power yet.

Action item: Draw a Venn diagram of what you must own, what you should partner on, and what you should refuse to do. If the first circle has more than three things in it, you are trying to build too much.

#05. Kill the seat license before somebody else does, price on the key, the token, or the lift

Adam is candid: nobody wants to log into Salesforce anymore, they want the data-and-action point. UI is dying. He watched customers cut Salesforce from 50 seats to one, piping the data into a vibe-coded internal tool that runs the actual playbook. Horizontal SaaS is going headless because it has to. CloudBeds got lucky here — they always charged by the key or the bed, not by the human, so the AI regime change doesn’t force them to unwind a per-seat contract.

The new pricing surface is a portfolio: transactional fees per token consumed when an LLM crawls their dataset, percentage-of-lift when their AI LLM demonstrably improves a property’s RevPAR, and traditional SaaS (but on a per room basis) for the system of record platform. His finding after running both experiments: the number came out roughly same. So the choice to his company has become what is cleanest to explain and easiest to sell.

Action item: Model the same customer-seat, per-transaction, and share-of-lift pricing this quarter. Pick the one that maps to how the buyer measures value, not the one your CFO’s spreadsheet is easiest to reforecast.

#06. Hybrid models: big model for language, your model for the specifics

Adam is emphatic: CloudBeds has no intention to reinvent math or visualization or general reasoning, those are commodity capabilities in the frontier models. What they provide is a vertical-specific model that knows why a rate goes up on a Thursday and not on a Friday for a boutique property in Lisbon during a specific event window. That is where CloudBeds spends its training budget: a fine-tuned corpus on very specific hospitality use cases and roles, orchestrated with other models that are best at the neighboring tasks.

He also believes the edge is coming, iterally. Phones will run trillion-parameter models. Nvidia’s roadmap makes it plausible. Cloud model choice will become a utility decision (Groq for sentiment because it’s on X, a Chinese or Japanese model for a specific reasoning benchmark, and so on). The architecture that survives is one where any model is swappable and the domain harness is yours.

Action item: Design your inference layer so the frontier model is a plug — swap Claude for Sonnet or an in-house fine-tune without touching the workflow. If you can’t do that today, that’s the refactor that saves you when the token bill flips.

A quick word from our incredible partner,
Parafin - the leading embedded capital partner for vertical software.

• 4.8 star rating on Trustpilot, with over 500 reviews [source]

• 71% CSAT rating across all partners

• 84 NPS

• Serving top Fortune 500 companies, including Amazon, Walmart, and DoorDash

Check Out Parafin

#07. Agents are only in phase one self-driving, swarm them, watch them, don’t trust them yet

The best analogy in the episode: self-driving cars are on phase four of five today— humans get into a Waymo without thinking about it. It’s been a bout a decade since they launched to get to this.

Agents are on phase one. They are cool, they daisy-chain, they fail often, they produce false flags and silent errors, and they burn tokens in loops nobody notices. Adam confesses he burned $3,200 in a month on a research harness he left running by accident. He tells that story publicly so smaller operators don’t repeat it.

The right posture is to swarm agents inside a governed sandbox — persona agents, department plus-ones, coding agents running 24/7 against the codebase — while your security and eval teams pressure-test what they leak. His EA has calendar access, so his EA-shaped agent does too.

Action item: Set a per-agent token budget and a red-team cadence this month. If neither exists, you are one accidental research loop away from a five-figure invoice.

#08. Over-index on planning, permitting takes longer than construction

Adam’s favorite operator metaphor: building AI features is like building a home. Permitting and planning eats 80% of the timelien. Foundation and walls go up fast once you have the right permits. Most companies invert this — they start construction, then have to tear it down when GDPR or California privacy law or brand-safety guardrails catch up. CloudBeds runs every AI initiative through a permit stage: what are the governing laws, are we tolerant of any AI slop in this output, where are the brand guardrails, etc.

The counterintuitive result: many initiatives never leave the permitting phase — because the honest answer is “we can’t make lemonade out of this one.” That’s not failure; that’s the discipline that keeps CloudBeds shipping products that actually move the needle. There is no silver bullet, and the humility to say so publicly is a real competitive advantage in a category that is currently drunk on hype.

Action item: For every AI feature request, run a two-page permitting doc — jurisdictions, guardrails, tolerance for weird output, failure playbook, before a single line of code is written. Kill the ones where the answer is “we don’t have the ingredients.”

#09. Direct booking wins — LLMs are the new discovery layer, and the inventory owner is king

Adam’s rapid-fire answer to whether AI benefits OTAs or hotels long-term: direct booking, one hundred percent. The reason is simple. Consumers already ask ChatGPT, Anthropic and Perplexity to plan trips to Montana. The current OTA advantage — being one of the sources of record the LLMs cite — is a rented advantage that expires the moment LLMs learn to talk directly to inventory owners. CloudBeds already does that distribution on behalf of its hotels, and sees those same LLM-originated requests hitting the platform.

The person who is king and queen in the long story is the inventory owner. LLMs collapse the layers of discovery and reservation. Anyone whose entire business model is being a middleman in that collapsed chain is watching their moat drain in public.

Action Item: If your customer sits on unique inventory or a unique dataset, expose it directly to LLMs via MCP or a well-documented API this quarter. Do not wait for an OTA-style intermediary to broker the relationship.

#10. AI can help A LOT with the little things

I asked Adam for the coolest vertical AI use case he’s seen, the answer is disarmingly small…

A high-touch luxury property programmatically prints a personalized welcome note at check-in and a personalized thank-you at checkout — remembers your favorite cocktail from three years ago at a sibling property in the chain — and the guest feels seen. That is the whole game. Not a robotic vacuum. Not a chatbot. A machine-generated moment of loyalty that a human team would have a hard time scaling to every guest.

The lesson generalizes: AI in a vertical wins when it lets the operator do something they always wanted to do for every customer, but couldn’t afford to do at scale. Not when it removes humans. Not when it slaps a chat window on the login screen. When it hyper-personalizes what used to be generic. that’s the compounding moat, because it’s the reason the guest comes back.

Action item: Pick one moment in your customer’s journey that used to require a human to remember them. Make your product do it automatically, at 100% coverage, this quarter. That is the AI feature that earns loyalty.

The operator takeaway from a 14-year incumbent still at the wheel: the treasure you spent a decade quietly burying is the only reason the pirates can’t catch you.

So guard the treasure…
Then sail the action no other crew can.

If you’re running a vertical software business in 2026 — especially one with a decade of customer workflow buried in the hold — four moves separate the captains that compound from the crews the pirates eventually catch:

1. Fortify the vault before you touch a model. Two years of DevSecOps and data-model work isn’t overhead; it’s the vault. Longitudinal, labeled, jurisdiction-aware transaction data is the gold a frontier LLM cannot mint on its own.

2. Forge the harness before you let the crew swing. Confidence score. Source-of-truth link. MCP-exposed data models. Trust is engineered, not promised and probabilistic answers to deterministic problems can show up as lawsuits, not blog posts.

3. Turn the system of record into a fleet. Dashboards were the price of admission at the port. The product is the next best decision, taken on the customer’s behalf, before they’re late to it.

4. Split the booty on the outcome, not the seat. Per key, per transaction, per share of lift. When SaaS and outcome-based experiments produce the same revenue, pick the one that matches how the buyer measures the plunder.

The crews that win this decade won’t be the loudest ships on the sea. They’ll be the ones who spent a decade quietly burying a dataset nobody else could copy, and had the discipline to lock it in a harness before anyone let the guns loose on a customer. Be that captain, and pirate season becomes free R&D…

Watch Episode Here

Listen To Episode Here

Thanks for reading LINEAR. I reply to every email…
Have any questions, feedback, comments? Let us know, we work for you!

Today’s newsletter is sponsored by Xplor Pay. Launching embedded payments is a major milestone, but it’s only the beginning of the opportunity. Many software companies see merchant adoption plateau after launch, not because of their technology, but because they stop investing in their payments business.

Built from firsthand experience growing and scaling more than 20 vertical SaaS platforms, this article explores what high-performing software companies do differently to increase merchant adoption, convert existing customers, and create long-term enterprise value. Read the article here.

Alright, let’s get to it…

Eighteen years bootstrapped, then $507M in one round

Dave Bateman didn’t raise venture capital because he thought bootstrapping was noble. He did it because no one would give him money.

The founding: 2003, paper ledgers everywhere

Entrata was incorporated in July 2003 as Property Solutions International. The multifamily property management industry was running on paper ledgers and fragmented on-premise software held together with duct tape. Property managers tracked leases on spreadsheets. Rent collection happened by check. Maintenance requests lived on Post-it notes.

Bateman’s first company was DearElder.com, a service that automated letter delivery for missionaries. It worked. It made money. But it wasn’t big. In 2002, he stumbled into property management through a friend who managed apartments. The workflows were medieval. Bateman started building.

He taught himself to code because his developers couldn’t solve the database bottleneck fast enough. He was obsessive about architecture. Not features. Not speed. Architecture. He believed that if you built the foundation right, everything else would compound.

Most founders optimize for shipping fast. Bateman optimized for building once.

The evolution: 2015, launch of the Operating System

For the first twelve years, Entrata sold property management software in a fragmented market. In 2015, the company made a strategic shift that would define the next decade. They changed their name from Property Solutions International to Entrata. And they launched the Entrata Operating System.

Not a point solution. A unified, cloud-native platform with a single login and a Unified Data Layer for each customer. CRM, ERP, property operations, resident experience, payment processing, maintenance workflows. All in one system. All on one database.

Legacy providers had grown through acquisition. They’d buy competitors, bolt the products together, and call it a suite. The result was five logins, seven databases, and zero integration. Entrata built from the ground up. One login. One platform. One system of record.’

The business model: subscription plus transactions

Entrata didn’t survive on subscription revenue alone. Bateman embedded payment processing early. Operators paid a monthly subscription to use the software. But the real revenue came from processing billions of dollars in rent payments every year. By 2021, Entrata was processing over $20 billion annually.

All subscribers of the Operating System are required to use Entrata’s payment solution. That’s not optional. It’s the deal. Because rent is a recurring monthly obligation, payment volume is durable, predictable, and immune to economic cycles.

That transactional revenue created a cash flow engine that let the company grow without dilution. Subscription revenue funded operations. Payment revenue funded R&D. The dual model made bootstrapping possible.

Scaling offshore before it was cool

Bateman built a 1,200-person engineering team in Pune, India. Not because labor was cheap. Because he couldn’t hire fast enough in Utah. But offshore development doesn’t work unless someone is awake when the offshore team is working. So Bateman hired technical managers in Utah to work overnight shifts. They’d code alongside the Pune team in real time. No handoffs. No email threads. Just live collaboration across time zones. It was brutal. It was expensive. It worked.

The litigation years

For years, Entrata fought legal battles with Yardi Systems, the legacy incumbent in property management software. The details don’t matter. What matters is that Bateman was running a high-growth software company while defending lawsuits that could have killed the business. He didn’t have the luxury of raising $50M to hire lawyers. He had to win on product. So he did.

The 2021 round: $507M, first institutional dollar

By 2021, Entrata had over $200M in ARR. More than 20,000 apartment communities on the platform. Over 2,100 employees. And zero venture capital. Silver Lake led a $507M round in July 2021. Ryan Smith (Qualtrics), Todd Pedersen (Vivint), Josh James (Domo), and Dragoneer joined. It was the largest private investment round in Utah history.

Dave Bateman retained majority ownership even after the round. Kyle Paster from Silver Lake and Todd Pedersen joined the board.

Most founders raise capital because they’re out of cash. Bateman raised because the opportunity was bigger than the balance sheet.

The business didn’t need capital to survive. It needed capital to dominate. Adam Edmunds, a serial SaaS founder who had built and sold Allegiance to MaritzCX in 2014 and served as president at Podium, joined as CEO. Chase Harrington became president and COO. Bateman stayed on as chairman. The company shifted from founder-mode to scale-mode.

The leadership transition: 2022

In March 2022, Silver Lake acquired Bateman’s majority stake and took control of the company. Bateman stepped down as chairman following a public controversy in early 2022. The company moved forward under Edmunds’ leadership.

The growth phase: acquisitions and AI

In July 2023, Entrata acquired Rent Dynamics, a provider of rent credit reporting services, for up to $50M in contingent consideration. In June 2024, Entrata acquired Colleen AI, a Tel Aviv-based AI company, for $53.2M. The company wasn’t buying competitors. They were buying capabilities: credit reporting, AI infrastructure, products that deepened the platform and expanded ARPU.

The 2025 round, the dividend, and the IPO

In May 2025, Entrata raised $200M from Blackstone at a $4.3B valuation. Six months later, in November 2025, the company paid a $356.3M special dividend to shareholders, funded by cash and a $400M term loan from JPMorgan.

This is the type of activity that get a lot of PE firms in a hot water. Taking out a massive loan to pay off shareholders is typically not a loved act. The business can support it, and you know those PE investors were in year 4 of their hold and getting nervous!

Then, in May 2026, Entrata filed to go public on the NYSE under the ticker ENT. The S-1 shows $509M in revenue, 24% growth, 60% gross margins, $50.7M in GAAP net income, and a business that powers ~2.5 million residential units — roughly 10% of the entire U.S. multifamily market.

What Entrata actually does

Entrata is a unified operating system for property management. It connects owners, operators, residents, and vendors in a single platform. The product has three layers, plus an AI engine sitting on top:

• OXP (Operations Experience Platform): guided workflows for operators — CRM, ERP, property ops, leasing, accounting, procurement, vendor management.

• RXP (Resident Experience Platform): the consumer-grade Homebody app. Residents pay rent, submit maintenance, manage their lease.

• Payments Infrastructure: the financial layer. Rent collection, procure-to-pay, refunds, reconciliation — unified rails, settlement logic, and accounting data.

• ELI (Entrata Layered Intelligence): an agentic AI engine. Essentials is free; ELI+ is the premium tier covering Leasing, Payments, Renewals, and Maintenance AI.

All three layers run on the Unified Data Layer — a proprietary database that captures real-time profile attributes, property data, behavioral signals, transactions, and workflow activity.

Go-to-market: enterprise land-and-expand

Entrata doesn’t sell to mom-and-pop landlords. They sell to enterprise operators managing thousands or tens of thousands of units — the top 50 property management companies, the REITs, the institutional owners. As of December 2025, Entrata served 4 of the top 10 NMHC operators and 10 of the top 50.

The sales motion is land-and-expand. A customer comes in on a subset of their portfolio — maybe 5,000 units — validates the product, then migrates the rest. 20,000. 50,000. 100,000. Of the 15% unit growth in 2025, 60% came from existing customer expansion and 40% from new logos. Net retention is 117%. Customers don’t just stay. They grow.

Eighteen years bootstrapped doesn’t mean eighteen years small. It means eighteen years of compounding without dilution.

By the time Entrata raised, the founder had built a business so valuable that he could take $500M+ in financing and still own the majority.

Entrata’s S-1: The Numbers That Matters

The S-1 is 230 pages. Here’s the highlights.

The ACV nobody talks about: ~$1.8M

The S-1 discloses 233 customers with $500K+ ARR driving 84% of total ARR. Do the math and the implied average contract value is ~$1.8M. That’s not a SMB SaaS number — it’s a mega-enterprise number, in the same zip code as Veeva and Guidewire, the only publicly listed vertical software companies with comparable ACVs.

Every second-order effect of a mega-enterprise motion shows up in the S-1: capital efficiency (S&M only ~13–17% of revenue), expansion (117% NRR, 97% GRR), and switching cost (3–5 year contracts with mandatory payments baked in).

Revenue: $509M and accelerating

Five-year CAGR sits around 24%. 2024 came in at $412M (~30% YoY), 2025 at $509.3M (24% YoY). Q1 2026 is $143.5M vs. $116.6M in Q1 2025 — 23% growth and accelerating margin. Worth noting: ARR growth actually re-accelerated from 22.5% in Q4’25 to 23.1% in Q1’26. Small move, big signal — most companies decelerate into an IPO.

The mix is two buckets:

• Subscription-related: $437.7M in 2025 (86%). Includes monthly OS subscriptions, payment processing fees, rent credit reporting, utilities, and embedded products like insurance and resident screening.

• Professional services and other: $71.6M in 2025 (14%). Implementation, training, consulting.

Payments are bundled into subscription-related. The S-1 doesn’t break it out cleanly, but they were processing $20B+ in rent annually as of 2021 — and that number has only grown.

Gross margins: Not great but expanding fast

GAAP gross margin moved 56% (2024) → 60% (2025) → 63% (Q1’26). Non-GAAP gross margin: 61% in 2025, 64% in Q1’26. Almost ten points of expansion in two years. It’s still below pure-play SaaS (80%+) — the gap is the ~15% of revenue that’s transactional (payments, insurance, screening) carrying a lower margin profile. That’s the trade-off for owning the rail.

Operating income: 26% margin in Q1 2026

Operating margin: 13% (2024) → 16% (2025) → 26% (Q1’26). Non-GAAP op margin: 23% in 2025, 28% in Q1’26. S&M dropped from 20% to 17% of revenue YoY. R&D held at 14%. G&A fell from 15% to 13%. The S-1 also shows headcount essentially flat versus 2022 — Entrata cut ~5% in 2024. Operating leverage isn’t a slide. It’s payroll.

Net income: GAAP profitable

$21.8M in 2024 (5% margin), $50.7M in 2025 (10% margin), $23.3M in Q1’26 (16% margin). GAAP. Not adjusted. Most high-growth SaaS is burning cash. Entrata is printing it.

Cash flow: strong but lumpy

Operating cash flow was $161.9M in 2024 (37% FCF margin), $100.1M in 2025 (17% FCF margin), then -$34.8M in Q1’25 and a hard rebound to $56.6M in Q1’26 (38% FCF margin). The 2025 dip was driven by changes in customer deposits, not operations. Adjusted FCF: $82.5M in 2025 (16%), $26.8M in Q1’26 (19%). CapEx of $13M in 2025. Capital-light by design — spend is people and cloud.

Units and ARPU: the two-variable growth engine

The S-1 lays out the unit count: 2.13M units (Dec ‘24) → 2.44M (Dec ‘25, +15%) → 2.49M (Mar ‘26, +12% YoY). At ~10% U.S. multifamily share (23.4M units), the TAM is ~90% unpenetrated. ARPU: $194 (2024) → $209 (2025, +8%) → $216 (Q1’26, +9%). Units growing ~13%, ARPU growing ~10% — that’s the ARR equation. Per the S-1, 60% of new units come from existing customers expanding their portfolios on the platform.

The top cohort tells the future

The S-1’s cohort disclosure is the most interesting page in the filing: the 1,000+ unit cohort hits $580 per unit annually. That’s 3x the company average and a window into where every customer ends up after a few years of attach. Entrata’s own filing says the platform replaces seven systems on average at onboarding — CRM, ERP, payments, screening, accounting, marketing, resident apps — all collapsed into one bill. That’s what’s doing the work in the ARPU number.

Customer concentration: 233 customers, 84% of ARR

Per the S-1: 233 customers with $500K+ ARR (up from 183 in 2024, +27%). Those 233 are 84% of total ARR. This is enterprise vertical SaaS — the top 50 PMCs in the U.S. are the target market. Some operators run multiple customer accounts (different portfolios, different legal entities from M&A), so the literal customer count is fuzzy on purpose.

Retention: 97% gross, 117% net

The S-1 prints gross retention at 99% (Dec ‘24) → 97% (Dec ‘25). The 2-point dip came from legacy point solutions Entrata acquired — they’re <3% of ARR but generated about half of all churn/downgrade events. Net retention held flat at 117% — the cohort of customers from 12 months ago now generates 17% more revenue. Every year, the base grows on its own.

Rule of 40: 41 trailing, ~49 forward

On a trailing basis (23.1% ARR growth + 18.3% GAAP operating margin), Entrata prints a 41.2. On the Q1’26 run-rate (23% growth + 26% GAAP op margin), you get to ~49. Both clear 40. Both are best-in-class. That’s the new minimum bar to IPO in 2026 — $500M+ ARR, accelerating, Rule of 40 compliant. The S-1 clears all three.

The competitive frame: real estate doesn’t have one winner

Here’s the uncomfortable part. The best vertical SaaS companies become the de-facto standard their industry consolidates on. In multifamily real estate, that may not be possible. RealPage ($10B+ valuation), AppFolio ($5.6B market cap), and Yardi ($15–25B private valuation) are all real. No one is getting to 30%+ share. Entrata is at ~10% and crowded at the top. The bull case is that the category is big enough that three or four winners all compound. The bear case is that nobody gets the monopoly multiple.

Debt and capital structure

As of March 31, 2026: a $400M JPMorgan term loan at 6.4%, a $75M undrawn revolver, and ~$389.6M net long-term debt. The term loan funded the $356.3M special dividend in November 2025. The IPO will partially de-lever the balance sheet but this is a recap story, not just a growth story.

Headcount, RPO, SBC

2,198 employees (including Pune) as of March 31, 2026 — that’s roughly $232K of revenue per employee. RPO of $279M (up from $272.5M at year-end), most of which recognizes within 12 months. SBC was $15.5M in 2024 and $35.0M in 2025 (7% of revenue) — the wedge between GAAP and non-GAAP op margin.

Valuation context

Entrata is targeting a $4.3B valuation at IPO. At $509M trailing, that’s ~8.4x — and ~7x forward at consensus growth. High-growth vSaaS (30%+ / 70%+ GM) trades 10-15x; profitable vSaaS (20%+ / GAAP positive) trades 6-10x. Entrata splits the middle: strong growth, expanding margin, GAAP profitable, embedded payments, AI layer, enterprise base. 8.4x is reasonable. Some of us think it’s cheap.

Entrata is not a growth-at-all-costs story. It’s a profitable, efficient, capital-light vSaaS with strong unit economics and a decade of TAM left.

Embedded payments when your customer uses ACH, not cards

Most embedded payment strategies are modeled on credit card interchange. Rent collection runs on ACH. That changes everything.

The problem

Credit cards generate 2-3% interchange. High volume, high margin, easy math. Most rent payments happen via ACH, electronic checks, or debit. Interchange is effectively zero. The traditional embedded payments playbook doesn’t work. If you’re building vertical SaaS in a category where the core transaction is ACH-heavy, you need a different model.

Entrata’s approach: fixed fees, not percentage-based

For ACH, electronic checks, and debit cards, Entrata charges a fixed per-transaction fee. Not a percentage. A flat dollar amount per payment. The only exception is credit cards, which are priced as a percentage. This does two things:

• It protects margin. ACH has negligible processing cost. A fixed fee creates predictable margin per transaction regardless of payment size.

• It scales with volume, not size. Rent is high-dollar but low frequency (monthly). A percentage fee would look expensive to the end user. A fixed fee feels like infrastructure.

The distribution advantage: mandatory adoption

All subscribers of Entrata’s Operating System are required to use the payment solution. This isn’t optional. If you’re on the platform, you process payments through Entrata. That means 100% attach. Zero incremental CAC to distribute the payments product. Every new customer is a payments customer. Every unit added is incremental payment volume.

If payments are optional, attach is 30–50%. If payments are required infrastructure, attach is 100%. That’s the entire game.

Revenue model: subscription plus usage

Entrata charges operators a monthly subscription fee to access the payment solution, then a per-transaction fee on every payment processed. That dual structure creates a revenue floor (subscription) and a ceiling that scales with unit count and transaction volume (usage). The subscription alone doesn’t cover cost-to-serve — but it signals payments are a core product, not a bolt-on.

Why ACH-heavy models are better than card models

• Lower churn risk. Rent is a recurring human need. Payment volume is durable. Card spend is discretionary; rent isn’t.

• Predictable volume. Rent is due monthly. Same cadence. Same behavior. Easy to forecast.

• Higher LTV. Property managers don’t churn. Residents turn over every 12-18 months, but the building doesn’t move.

The real value isn’t the payment fee — it’s the data

Entrata records payment revenue gross of interchange and third-party fees, so the $509M includes payments to processors like Worldpay. Net take rate isn’t disclosed, but the 56% → 60% → 63% gross margin march tells you the blended economics work.

The real value isn’t the payment fee. It’s the Unified Data Layer. Every rent payment feeds Entrata’s system of record: payment history, resident behavior, lease terms, delinquency risk. That data powers the AI agents, underwriting models, the credit reporting product, and insurance upsells. Payments aren’t the end state. They’re the data-capture mechanism.

Three rules for embedded payments in ACH-heavy verticals

1. Make payments mandatory. Optional products have 30% attach. Required infrastructure has 100%.

2. Price by transaction, not percentage. Fixed fees protect margin when interchange is low.

3. Monetize the data, not just the fee. Payments create the system of record. The system of record creates the platform.

If you’re building vertical SaaS in logistics, healthcare billing, supplier payments, or any category where the core transaction is ACH-based — this is the playbook. Entrata just published a 230-page proof of it.

The story is that a Utah founder bootstrapped through paper ledgers, lawsuits, and an entire generation of VC fashion — and built one of the cleanest vertical SaaS businesses going public this decade.

If you made it thus far, go check out the 2026 Vertical Software Summit. We’ll have 400+ vertical founders/operators/investors in Miami in November. Two days, 6+ billion dollar vertical founders. The Vertical AI event of the year.

Check Out The Summit

Do me a solid and forward to a friend :-)

In 2014, Zain Jaffer was running Vungle from a London bedroom on a few thousand dollars. By 2019 it had sold to Blackstone for $780M. By 2021 it was public at $4B.

The playbook that got him there — managerial layers, seat-based pricing, predictable MRR, hire fast, scale the org chart — is the same playbook he now walks into Blazel’s all-hands meetings to systematically unlearn. “My brain was wired still for a very SaaS-based world,” he says. “What worked previously does not work anymore.”

The first thing he had to unlearn was the org shape…

Look at Anthropic — founders and CTOs joining as individual contributors on a deliberately flat technical bench. Three-person companies running real revenue. The instinct to add headcount, layer in middle management and create org charts that look like a public company is a 2018 reflex that now actively suffocates the speed AI-native execution requires.

The second was where defensibility actually lives. The product is the easy part. The token bill, the latency stack, the frontier-lab dependency, the seat-based price floor — all of that gets compressed in public over the next 24 months. What doesn’t compress: owned audience, owned distribution, a dataset of human-labeled edits you can fine-tune into a model the labs can’t replicate without your customer relationships. Brand, voice and proprietary data are the moats. The model is rented from whoever is cheapest this quarter.

The third was the trap of easy revenue. Zain ran Blazel as an AI-native marketing agency for long enough to know how seductive it is — replacing existing budgets, no procurement friction, no UI required, a clean shot to $10M ARR in a year. And how it kills you. Every AI feature cannibalizes your own team. Every new entrant races you to the bottom on price. Your board freaks out the moment you propose taking revenue down to switch to outcome-based pricing. He pivoted Blazel into a platform before the trap closed — and the rest of this issue is the operating system underneath the move.

Watch Episode Here

Listen To Episode Here

This Weeks Vertical Titan:
Zain Jaffer

Zain has started multiple companies. The most successful was Vungle — the mobile video advertising business he scaled from a London bedroom into one of the defining ad-tech outcomes of the last decade, acquired by Blackstone for $780M in 2019 and taken public at a $4B valuation. After Vungle he moved to the other side of the table for several years as an active operator, investor and board director across venture-backed companies.

Now he’s running Blazel, an AI-native marketing platform that helps founders and executives publish high-signal content on LinkedIn and convert it into pipeline. Blazel started life as an AI-native agency, harvested the workflow data from a team of human content operators, fine-tuned its own model on the labeled edits, and pivoted into a self-serve platform priced from hundreds of dollars a month — replacing what used to cost enterprises millions in in-house production.

The frame Zain keeps returning to: a second-time founder’s playbook is baggage, not an asset, in an AI-native company. The reflexes that made Vungle scale — bigger orgs, MRR forecasting, seat-based pricing, agency cost structures — actively slow down the kind of company Blazel needs to be. The hard work isn’t building the product; it’s unlearning the playbook that built the last one.

The line every AI-native founder should write on a wall: history is buried with the corpses of great products that lacked distribution. In a world where 99.6% of LinkedIn is AI-generated, owned audience and authentic voice are the only moats the model companies can’t ship.

Watch Episode Here

Listen To Episode Here

The post-SaaS founder playbook — ten moves from a $780M operator starting over.

Operator-grade lessons from Zain Jaffer on what to unlearn after a $780M exit — the AI-native agency trap, outcome pricing, building distribution before product, and why every SaaS reflex from 2018 is now a liability.

#01. Burn the SaaS instincts — your last playbook is now your biggest tax

Zain came back into the game with what looked like an unfair advantage — a second-time founder who’d built Vungle from a London bedroom to a $780M sale to Blackstone and a $4B public listing. Instead he calls it baggage. “My brain was wired still for a very SaaS-based world. What worked previously does not work anymore.” The reflex to add managerial layers, the obsession with predictable MRR, the discipline of seat-based pricing — every instinct that made Vungle scale is now actively slowing Blazel down.

The Anthropic shape is the new template: leading founders and CTOs joining as individual contributors on a deliberately flat technical org. Three-person companies running real revenue. The founders who refuse to let go of 2018 SaaS muscle memory are the ones writing severance checks in 2026.

Action item: List the five operating habits you copy-pasted from your last company. Pick the most expensive one — managerial layer, MRR forecast, seat-based pricing — and kill it this quarter before AI does it for you.

#02. The AI-native agency is a trap dressed up as a wedge

Blazel started as an AI-native marketing agency. It worked. Zain says you can get to $10M ARR almost casually right now — you’re replacing budgets that already exist, there’s no procurement, no vendor approvals, no UI required. Then the trap closes. Every AI feature you ship cannibalizes your own headcount. Every new AI-native competitor races you to the bottom. And when you finally tell your board the right move is to take revenue down and shift to outcome-based pricing — they freak out.

His call: agency is a brilliant bootstrap and a phenomenal cash-flow business, but it is not a venture-scale company. Use it as a wedge to learn the workflow, harvest the labeling data, and earn the right to build the platform. Set a clear date by which the human-in-the-loop becomes a product feature, not the product. Zain pivoted Blazel before the agency hit critical mass on purpose — because at $10M ARR you can’t.

Action item: If you’re running an AI-native services business, write down today the platform you’re earning the right to build — and the revenue level at which you’ll force the pivot. Above that line the cultural switching cost is fatal.

#03. Distribution is the moat — own your audience before you need it

Zain’s hardest-earned lesson from two decades of building: history is buried with the corpses of great products that lacked distribution. The companies winning right now — Lovable, Cursor, the vibe-coding wave — aren’t winning on engineering. They’re winning on owned channels. Their CEO posts. Their COO posts. Their account executives post. Their AEs post. Press releases are rented megaphones. A newsletter, a podcast, a substack, an active personal brand on every executive — those are owned assets that compound for years.

There’s a downstream play, too: every piece of high-quality content you publish gets indexed by search engines and AI search agents. When a buyer asks ChatGPT or Perplexity for the best tool in your category, the answer is partly downstream of what you’ve been saying in public for the last two years. Most companies treat content as a marketing line item. Zain treats it as the most underrated permanent capital expenditure a startup can make.

Action item: Get your CEO, COO, and CTO posting two to three times a week on the channel where your buyers actually live. If you can’t sustain it manually, that’s exactly why platforms like Blazel exist — but the worst answer is doing nothing.

#04. Burn the MRR religion — outcome and usage pricing is the new contract

Zain calls predictable monthly recurring revenue “a staple of vertical SaaS” — and then names it as one of the instincts he had to unlearn. In an AI-native world the customer is paying for outcomes (leads generated, posts published, tickets resolved) and the cost structure is variable (tokens, GPU time, model calls). Forcing a flat per-seat license over that reality is a SaaS-era abstraction that’s about to crack in public.

It is uncomfortable. Boards hate it, finance hates it, and your sales team hates it because comp plans don’t work the old way. But this is exactly the discomfort that’s creating the opening. The companies that survive cannibalize their own MRR before someone else does. The ones that don’t are stuck protecting the seat license while a three-person AI-native team prices on outcomes and undercuts them by an order of magnitude.

Action item: Re-price your top-tier offering on the outcome that actually matters to the buyer this quarter. If finance pushes back, run it in parallel — but get the pricing surface that matches your economics into the market.

#05. Mandate AI fluency — every employee on Claude Code by Monday

At Blazel every employee — AE, intern, designer, marketer — has access to the codebase and is expected to ship. Zain mandates it. He’s seen the Anthropic blueprint from Thiel fellows working inside: same tools every public dev has, no proprietary harness, no review gauntlet, just ship-it culture. That’s why they outpace orgs ten times their size.

It comes with a cost. Vibe-coded passion projects pile up. People try to rebuild your CRM in a weekend. The discipline is to channel it: prototypes ship to staging, never to production, until they’ve been validated with a design partner. Done well, it functions like in-house forward-deployed engineering — your team builds for real customers, then your engineers harden what works into the core product. The cultural shift is the whole point: nobody hides behind a waterfall doc anymore.

Action item: Hand every non-engineer at your company a Claude Code seat and a staging branch this week. Set one rule: production code requires review, staging is fair game. Watch your roadmap accelerate.

#06. Pick one ICP, refuse the rest — Vungle’s $4B vs AppLovin’s $250B

The most expensive mistake of Zain’s career: at Vungle they served two markets — brand advertising and app installs — at the same time. AppLovin picked one. Today Vungle is worth $4B and AppLovin trades between $150–250B depending on the day. Same era, same TAM, same technology stack. The difference was a single product decision about who they refused to serve.

He’s applying the lesson to Blazel: head of marketing at a Series B company, post-PMC, with marketing collateral to train on, a stable website to anchor in, and a dream of activating the entire employee base on content. Yes, SMBs and creators self-serve the product — Zain treats that as a subset of the ICP feature set, never as the ICP itself. The hard thing is building for the demanding customer; the SMB market falls out of it for free.

Action item: Write your ICP in one sentence: role, company stage, and the specific outcome they’re judged on. If you can’t, you’re running two companies. Pick one and shut the other.

A quick word from our incredible partner,
Parafin - the leading embedded capital partner.

• 4.8 star rating on Trustpilot, with over 500 reviews [source]

• 71% CSAT rating across all partners

• 84 NPS

• Serving top Fortune 500 companies, including Amazon, Walmart, and DoorDash

Check Out Parafin

#07. Build your own model before token spend builds it for you

Sounds insane when OpenAI and Anthropic are trillion-dollar companies. But Zain watched Blazel’s token bill blow past $70K in a single month — Cloud Code spend, API calls, agent-on-agent orchestration trying to scrub AI slop out of generated content. At some point the math flips. Latency stacks up across agent chains. Margin compresses. The fix isn’t more agents on top of Sonnet — it’s collapsing the workflow into a fine-tuned model whose weights already know what “good” looks like for your vertical.

He’s not telling founders to train from scratch — that’s still insane. He’s telling them to assume that within a few years, a trillion-parameter model will run under their desk thanks to Nvidia, and to design the company so that bringing the model in-house is a Tuesday, not a re-platforming. Until then, harvest labeled data from every human-in-the-loop edit. That dataset is the moat.

Action item: Instrument every human edit your product touches as labeled training data. The day your token bill exceeds two engineers’ salaries, you have the data to fine-tune your way out of it.

#08. Where humans belong: labeling, evals, and forward-deployed engineers

Zain is bullish that fully-automated platforms beat human-heavy agencies — and equally clear about where humans are non-negotiable. Three places. First, data labeling: OpenAI, Anthropic, xAI and Google all quietly pay surgeons, lawyers and PhDs real money to label outputs. The dirty secret of the frontier labs is that they’re services businesses on the back end. Second, evals: an LLM-as-judge only works if its intuition was bootstrapped from world-class humans. Third, forward-deployed engineers — the Palantir move — sitting inside enterprise accounts to integrate, customize and drive adoption. For larger ticket sizes, FDEs are now table stakes.

What that means for the org chart: you don’t need a 200-person services arm. You need a small bench of the best domain experts in the world labeling the edge cases, and FDEs embedded in your top accounts. Everything in the middle — the content team, the QA team, the implementation consultants — collapses into the model over time.

Action item: Map your roles into three buckets: replaced by AI, augmented by AI, and a permanent moat for humans. If anyone is in bucket one without a transition plan, you have six months.

#09. Audience is the only counter-move to AI slop

Zain’s company analyzed LinkedIn at scale: 99.6% of posts are AI-generated, and only 0.4% qualify as acceptable training material. We are watching model collapse in real time on the world’s largest professional network. The reflex is to add more AI to the feed. The actual move is the opposite — authentic, edited, voice-driven content is now arbitrage, because everyone else is producing slop at infinite scale.

His prescription is unromantic: write your unique content into a projects folder, build a voice guide, record voice notes, use AI to draft and then spend real time editing. Or — if your time is too valuable to be on the content treadmill at 10pm on a Sunday — use a platform that does the orchestration for you. Either way, the bar for what stands out is going up, not down. Authentic voice beats infinite generation, forever.

Action item: Audit your last ten posts for AI tells — “not just X, but Y,” staccato bullets, the word “genuinely,” em dashes that don’t earn their keep. If more than half ring AI, you’re feeding the slop. Reset the voice guide.

#10. The new venture-class public stocks — and the bubble signal to watch

Zain has a sharper read on the public markets than most operators: SpaceX, Anthropic and OpenAI are functionally a new asset class — venture-shaped stocks trading on public exchanges, with ten-year hold periods baked into the price. Retail demand is so intense that SpaceX allocations are 6% of asks even for active investors with banking relationships. SPVs are layered, opaque, and in some cases illegal — people are getting wiped out.

The signal he’s tracking: when retail piles into pre-IPO vehicles at scale, that’s historically the most reliable leading indicator of a bubble cresting. Doesn’t mean the underlying companies are bad bets — Nvidia, Microsoft and Google at IPO would have made you rich. It means the next six to twelve months are going to be loud. The infrastructure providers, the GPU layer, and the application layer all look like winners in different scenarios. The model layer itself may be the most fragile of the three.

Action item: Stress-test your company against a 50% compute-cost collapse and a 50% model-pricing collapse. The strategy that survives both is the one worth executing.

The operator takeaway from a $780M founder doing it all again: your last playbook is the most expensive tax you’re paying.

If you’re building an AI-native company in 2026 — and especially if you’re a second-time founder leaning on instincts from the last cycle — four moves separate the founders who compound from the ones who get cannibalized:

1. Flatten the org and put every employee in the codebase. Anthropic’s shape isn’t a quirk — it’s the new operating template. Mandate AI fluency across every function. Kill managerial layers before they kill your shipping speed.

2. Price on outcomes, not seats. Cannibalize your own MRR before a three-person AI-native team does it for you. The pricing surface that matches your variable cost is the one that survives the next 24 months.

3. Build owned distribution before you need it. A CEO podcast, a company newsletter, every executive posting in their own voice. The 0.4% of authentic content on LinkedIn is now the rarest asset in B2B marketing — and it compounds for years.

4. Refuse the easy $10M ARR. If you’re running an AI-native services company, set the date you pivot to platform — and set it before the cultural switching cost makes the pivot impossible.

The founders who win this decade won’t be the ones with the most impressive last playbook. They’ll be the ones with the discipline to burn it — to flatten the org, to price on outcomes, to own the distribution, and to refuse the easy revenue that traps everyone else. Be that founder, and you get to do it all again.

Watch Episode Here

Listen To Episode Here

Thanks for reading LINEAR. I reply to every email…
Have any questions, feedback, comments? Let us know, we work for you!

Today’s newsletter is sponsored by Xplor Pay. In the Investor-Operator Guide to Embedded Payments, Xplor Pay examines how embedded payments increasingly impacts retention, operational scalability, monetization, and long-term enterprise value across vertical SaaS.

Built from firsthand experience building and scaling more than 20 vertical SaaS platforms, the guide covers why many payment programs underperform after launch, what sophisticated investors increasingly evaluate, and how high-performing SaaS platforms operationalize payments successfully as they scale.
Download the free guide today.

Alright, let’s get to it…

The Token Uplift: Why Vertical AI companies can get to a billion faster than the Vertical SaaS ever could.

Every legacy vertical SaaS company is priced the same way, and the ceiling of every one of them is set by the same equation: average revenue per account, multiplied by the number of logos in the category, multiplied by the slice of those logos you can realistically win. The denominators are fixed by physics. There are only so many dealerships, only so many law firms, only so many school districts. When you do the arithmetic on what it takes to build a billion-dollar SaaS company in most verticals, you discover something uncomfortable: in many of them, it can’t be done from seats alone.

The old math: seats, locations, and a thin payments cut

Most people still model vertical AI like a better SaaS company. That is too small. The real prize isn’t a nicer product or a smarter automation layer or a new seat tier — it’s a new revenue architecture. Legacy vertical SaaS sold software access. AI-native vertical software can sell software access plus completed work. That one change makes the company a lot bigger, not because the market got bigger, but because revenue per customer can expand far beyond old SaaS ceilings.

For two decades, vertical SaaS has been priced the same way regardless of industry. There’s a per-seat fee (the user), a per-location platform fee (the rooftop, the office, the school), and a payments take rate bolted on once the workflow runs through the software. Toast does it for restaurants, ServiceTitan for HVAC, Procore for construction. The model is so standardized you can sketch the cap table on a napkin.

The constraint is brutal and arithmetic. There’s only one equation that really matters: Revenue = number of customers × annual revenue per customer. Most founders obsess over the first half — logos. The better founders obsess over the second half — ARPA (Annual Revenue Per Account, the average dollar value a single customer contributes over a year). If your average customer pays you $10,000 a year, a billion-dollar revenue company requires 100,000 logos. In most verticals, 100,000 logos is more than the entire addressable market. The reason every legacy SaaS deck eventually pivots to “expansion into adjacent verticals” is not ambition — it’s that the original TAM mathematically can’t get them there.

Historically, there was only one way to raise ARPA: find more seats or build more products. Embedded payments helped — a 30 bps take on the GMV flowing through your invoicing module can meaningfully lift ARPA without selling another seat. But payments are a rate, not a workflow — they don’t actually do anything for the customer that a generic processor wouldn’t. They’re an attach, not a product. And critically, the cost center they sit on top of is transaction fees, which is a small line item compared to the one AI is about to attack: labor.

The new math: sell the work, not the chair

A seat-based SaaS company gets paid when a human logs in. An AI-native company gets paid when work gets pushed through the system. That means revenue can scale with task volume, not just headcount — and that distinction matters enormously in verticals, because verticals are full of repetitive, measurable, high-value work.

Vertical AI keeps the seat fee and the payments line, and adds a third revenue line that didn’t exist before: AI credits, charged for the work the software actually does. Calls handled. Documents reviewed. Lessons generated. Clinical consults delivered. Claims processed. Follow-ups completed. The credit line is metered, it scales with usage, and crucially, it replaces a labor budget that is orders of magnitude larger than the software budget it sits next to.

Embedded payments helped the first generation of vertical SaaS lift ARPA. AI credits do the same thing — against a much larger cost center. That’s the token uplift, and it’s why the next generation of vertical software needs far less market penetration to build billion-dollar businesses.

The pricing instinct most founders get wrong: they compare an AI product to legacy SaaS comps and end up too cheap. The right anchor isn’t CRM pricing or document-management pricing — it’s payroll, outsourced services, error reduction, and throughput. A dealership AI that recovers missed service appointments shouldn’t be priced against a CRM seat. A legal AI that absorbs chunks of junior-associate time shouldn’t be priced against a case-management subscription. Wrong comparison set, wrong company.

Below, four head-to-head examples. Same verticals. Same buyers. Different unit economics.

Case Study #1: Car Dealerships

Labor Bridge & AI Average Revenue Per Account Origin:

A typical franchise dealer staffs ~10 BDC + service advisors at a ~$60K fully-loaded comp — roughly $600K of annual labor on inbound calls and scheduling. Toma’s $240K AI ARPA represents capturing ~40% of that single labor line.

Addressable Labor: ~$600K / yr per dealer (BDC + service-advisor phone & scheduling labor)

AI Capture: ~40%

Target AI ARPA: $240K / yr (1.7× vs. legacy)

Market: ~16,800 franchise dealers in the US

Legacy path to $1B —At ~$12K/mo all-in ARPA, a billion of ARR requires ~6,950 dealers — roughly 41% of every franchise rooftop in America.

AI-native path to $1B —At a blended $8K/mo subscription + $12K/mo in AI call credits (a busy store handles 4,000+ inbound calls a month), ARPA approaches $20K/mo. A billion in ARR comes from ~4,200 dealers — about 25% of the market.

Toma is selling the BDC headcount, not the software seat. The AI credit line is bigger than the SaaS line within 12 months of go-live.

Case Study #2: Legal

Labor Bridge & AI Average Revenue Per Account Origin

A mid-level associate bills out near $400K/yr; ~$150K of that is the drafting, diligence, and research workload AI can now absorb. Harvey’s $18K/attorney/yr is ~12% of that addressable wedge — and still cheaper than the alternative.

Addressable Labor: ~$150K / yr per attorney (AI-addressable) (Junior-associate drafting, diligence & research time)

AI Capture: ~12%

Target AI ARPA: $18,000 / yr (12× vs. legacy)

Market: ~1.33M licensed attorneys in the US

Legacy path to $1B —Clio’s blended ARPU sits near $130/mo, or ~$1,560/yr per seat. A billion of ARR requires ~640,000 paying attorneys — nearly half of every lawyer in the country.

AI-native path to $1B —Harvey’s blended take is roughly $1,500/attorney/mo once credits ramp ($18K/yr). A billion of ARR comes from ~55,000 attorneys — about 4% of the market, and concentrated inside the AmLaw 200 where adoption is already mid-double-digits.

The same lawyer who paid $1.5K/yr for a practice tool now pays $18K/yr for an associate that doesn’t sleep. Same logo, 12× the revenue.

Case Study #3: Education

Labor Bridge & AI Average Revenue Per Account Origin:

Average US teacher comp is ~$66K, and ~$20K of that is the prep, grading, and IEP/comms work AI can offload. MagicSchool’s $200/teacher/yr captures only ~1% of that today — the runway up to even 5% capture is a 5× ARPA expansion before they touch a new logo.

Addressable Labor: ~$20K / yr per teacher (AI-addressable) (Teacher prep, grading & differentiation hours)

AI Capture: ~1%

Target AI ARPA: $200 / teacher (200× vs. legacy)

Market: ~13,300 US public districts · ~3.8M public school teachers

Legacy path to $1B —PowerSchool sells the district, not the educator. At ~$60K/yr per district ARPA, a billion of ARR requires ~16,700 districts — more districts than exist in the United States. Hence the international and higher-ed expansion.

AI-native path to $1B —MagicSchool can sell every teacher in a district. At $200/teacher/yr fully loaded, a billion of ARR requires ~5M teachers globally — or ~1.3M US teachers, ~35% penetration of US public school teachers. The unit is 50× bigger than ‘one district’.

Legacy SIS hit a ceiling at the number of districts. AI flips the denominator from buildings to humans, and humans are 200× more numerous.

Case Study #4: Healthcare

Labor Bridge & AI Average Revenue Per Account Origin:

Average US physician comp is ~$350K, and ~$40K is the literature search, chart synthesis, and point-of-care decision work AI can absorb. OpenEvidence’s $1,500/physician/yr is ~4% of that line — and unlocks a malpractice-avoidance multiple on top.

Addressable Labor: ~$40K / yr per physician (AI-addressable) (Clinical research, chart synthesis & decision-support time)

AI Capture: ~4%

Target AI ARPA: $1,500 / physician (1,800× logos vs. legacy)

Market: ~1.1M practicing US physicians

Legacy path to $1B —Doximity’s revenue model is concentrated: ~600 pharma brands × ~$1M ARPA = bulk of revenue. A billion requires capturing roughly the entire top-of-funnel pharma marketing wallet.

AI-native path to $1B —OpenEvidence at $1,500/physician/yr enterprise pricing needs ~660K physicians — ~60% of US doctors, achievable because the product is used several times per shift and ROI is measured in avoided malpractice. Adoption already exceeds 40% of US clinicians.

Doximity monetized the doctor’s attention once a quarter. OpenEvidence monetizes the doctor’s decisions every hour.

The Vertical Software Summit Is Back!

This is the LAST day for early bird tickets. Price goes up tomorrow.

Hear biz stories from 6+ billion dollar vertical ai Founders+CEO’s
We are already 50% booked and still ~4 months from the Summit!

Grab Your Ticket Here

The chart: ten more industries, same pattern

The four deep-dives aren’t outliers. Run the same math across any vertical with a meaningful labor line, and the legacy SaaS model either requires a mathematically impossible market share or a forced expansion into adjacent categories. The AI-credit model collapses the required penetration into the low single digits — sometimes the low double digits in tiny markets, but always a fraction of what the seat-only model would have needed.

The pattern is consistent: wherever there’s a repetitive, measurable, expensive human task — claims adjudication in insurance, intake and routing in home services, lesson generation in schools, clinical synthesis in medicine — the labor line is 10× to 50× larger than the software line that currently serves it. Whoever turns that labor line into a metered work unit captures a multiple of the legacy ARPA.

What this changes

The best AI-native companies don’t need to win the whole stack first — they just need to win one expensive work loop. Toma doesn’t need to replace CDK Global to become important; it only needs to become the default layer for one painful, recurring, high-value job: handling inbound demand that human teams consistently drop.

Once that wedge works, every adjacent workflow carries its own AI work revenue. Harvey doesn’t need to become the operating system of every law firm on day one — it only needs to become indispensable in enough substantive workflows that firms start budgeting for AI output, not just AI access. MagicSchool doesn’t need to replace PowerSchool; it needs to own a growing share of teacher and administrator workload.

The seat license was a proxy for value when the software made the human faster. AI credits are not a proxy — they’re a direct claim on the wage bill. That shift moves the ceiling of every vertical software company from “how many seats can we sell” to “how much of the work can we do.” The second question has a much larger answer.

The companies that will compound the fastest from here start with embedded payments and a tight seat — because that’s how you earn the right to be in the workflow — and then expand the credit line until it dwarfs the subscription. Legacy vertical SaaS made money by becoming the system a business used. The next generation will make far more money by becoming the system that actually gets the work done. If AI only makes software better, you get a better SaaS company. If AI makes software billable against labor, you get a much bigger one.

That’s the whole game.

It’s time to ask yourself a very important question…

Which game are you playing?

Do me a solid and forward to a friend :-)

In late 2022 the consensus take on application-layer AI was that it was a death sentence. Every startup was a thin wrapper around ChatGPT. Every wrapper would be obfuscated the moment the underlying model got smart enough to read user intent. The narrative was so loud that VCs — already allergic to a vertical with few venture-scale exits — looked at Adeel Khan, a public-school principal with no software background, and said no roughly 98 times in a row. Three years later Magic School is leading the AI race with millions of teachers and 20% of children in America using his product….

The first thing the consensus investors got wrong was who the user is…

Teachers spend ninety minutes a day at a computer and the rest of it performing in front of children. They are not going to assemble a stack of Claude projects, MCP connectors and custom skills to extract bleeding-edge value out of a horizontal model. They want one button that already knows their roster, their rubric and yesterday’s quiz. The wrapper isn’t the liability — the wrapper is the entire product.

The second was what defensibility looks like. The gap between a stock Claude answer and a Magic School answer is everything Magic School knows that Claude can’t — the kid’s IEP, the district’s curriculum, the writing rubric, the implementation context. The vertical platform doesn’t compete with the model; it composes context on top of the model. Every copycat that’s shipped — PowerSchool, curriculum companies, a hundred lookalike UIs — almost never shows up in Magic School’s deals, because brand, distribution and context are the moats. The model is rented.

The third was what an organic PLG flywheel actually feels like. Five months from launch to a million teachers, zero paid acquisition. The number-one Google result for “AI for teachers.” Eighteen-hour days from a team of heroes keeping the lights on while the order-taking calendars filled themselves. The rest of this issue is the operating system underneath the three: how Adeel hired, how he priced the pilots that underwrote a Series A, when he flipped on monetization, and what he tells the board the next inflection looks like.

Watch Episode Here

Listen To Episode Here

This Weeks Vertical Titan:
Adeel Khan, Founder & CEO
@ MagicSchool

Adeel grew up in a low-income household on free-and-reduced lunch and a Pell grant. He started his career as a public-school teacher in Atlanta, became an assistant principal at several schools across Denver, and eventually got to open his own public high school — Conservatory Green in Denver Public Schools — built on a simple promise: love the kids, hold them to high expectations, graduate 100% of seniors to four-year colleges. He did. After a year of principal coaching and a personal sabbatical, he came back from traveling in November 2022, the same month ChatGPT launched.

He had never built software. He didn’t know what a product manager was, what retention metrics were, what a Series A meant. A close friend who had just raised a seed told him: I’ve watched you do hard things, I bet you could do this. He hired a contractor, recruited a couple of engineer friends and a former student as an intern, and shipped the MVP of Magic School in May 2023. By October it had crossed a million teachers, all organic. Today it partners with districts representing roughly one in five U.S. children, has a teacher in nearly every American school building, and runs in 160 countries.

The frame Adeel keeps coming back to: his lived experience as a teacher and principal wasn’t a soft asset — it was the entire unfair advantage. He couldn’t have built this without it, and Magic School’s go-to-market team is now 85–90% former educators by design. The product is what falls out of user-empathy at scale, not what gets bolted on after.

The line every vertical AI founder should write on a wall: in education there is no greater opportunity to make impact at scale, and most of the work the consensus market won’t fund is exactly the work that needs doing.

Watch Episode Here

Listen To Episode Here

The Magic School Vertical AI playbook — ten moves from a school principal who out-shipped the “tech people”

Operator-grade lessons from the most extreme vertical-AI PLG story to date — how Adeel hired for domain, priced pilots, built the easy button on top of horizontal models, and survived the shape change from order-taking to real enterprise sales.

#01: Domain expertise is the unfair advantage you can’t fake

Adeel had never seen a product manager, didn’t know what retention metrics were, had never heard the phrase Series A. He had run a public high school in Denver from scratch — graduated a class of first-generation college-bound seniors with 100% four-year acceptances — and he had taught English in Atlanta before that. When ChatGPT shipped in November 2022, he had a point of view on what generative AI could do for teachers that no Stanford CS grad in San Francisco could have, no matter how good the model got. The product fell out of that point of view.

His honest take: he absolutely could not have built this without the principal years. Someone who’s never taught could still win in vertical AI — but only if they treat domain depth as a constraint they actively recruit for, on the team or on the cap table. Today 85–90% of Magic School’s go-to-market team are former classroom teachers. Customers feel it in the first meeting, and they say so out loud.

Action item: Audit your team by classroom hours, not LinkedIn titles. If your AE, your CS lead and your solutions architect have never lived your customer’s job, hire one who has — even if it costs you a quarter of velocity.

#02. If your user has 90 minutes a day at a computer, build them the easy button

The fashionable take in 2023 was that application-layer AI was a death sentence — platforms would obfuscate every wrapper as the models got smarter. Adeel bet the opposite, on a very specific population: K-12 teachers spend one or two hours a day at a screen, in front of kids the rest of the time. They are not going to maintain ten Claude projects, three MCP connectors and a custom skill library. They want one button that already knows their kids, their roster, yesterday’s quiz data, and their district’s writing rubric.

Three years in, his read has only sharpened: to get bleeding-edge value out of horizontal models today you need context, connectors, skills, prompts and discipline. That’s a hobby, not a job. Vertical AI’s job is to package the entire stack into a single click for someone who would never assemble it themselves. The wrapper isn’t the liability — the wrapper is the whole thing.

Action item: Map the entire chain of context your power-user version of yourself would assemble to do your user’s job. Then ship the one button that hides all of it.

#03. Free is a feature until it isn’t, and you have to be honest about when

Magic School was free from launch in May 2023 to January 2024. Adeel’s heuristic was naive on paper and brilliant in retrospect: the biggest companies he had ever loved as a consumer were free. So get teachers in the product, figure out the rest later. In five months it went 0 to 1M users, all organic — number one Google result for “AI for teachers” — and roughly one of every three U.S. teachers had logged in.

The hard part was knowing when to turn the meter on. They flipped to paid in January 2024 at the moment the PLG halo was loudest, not the moment it had faded. The lesson isn’t “go free” — it’s that free is a deliberately temporary strategy with an end date, and the end date is determined by the customer’s willingness to pay, not by your runway anxiety.

Action item: Write down today the metric or moment that triggers your free-to-paid flip. If you can’t define it, you’re not running a PLG motion — you’re running a free product that’s going to stay free.

#04. Sign the contract for $1,000 — price the renewal at full ticket

Adeel’s most original move during the order-taking year: he offered districts spring-semester pilots at $1,000 per school site or $2,000 for a whole district — Miami-Dade with 350,000 students, same price as a single school. The trick was the contract. Pilots locked in the standard renewal ticket price for July, the moment K-12 buying cycles actually close. The pilot was cheap; the contract was a real signed agreement.

Internally the renewal rate was a wide-ranging bet — the CRO modeled 50%, Adeel modeled close to 100%. Bain Capital Ventures wrote the Series A in the middle of that uncertainty. Everyone renewed at full price. The structural insight: in long-cycle enterprise sales, the asset that matters is a signed paper agreement, not the revenue under it. Price the entry low enough that no one says no, then make the renewal economics non-negotiable.

Action item: Re-write your pilot contract this week so the entry price is whatever your buyer can sign without procurement, and the renewal is locked at your real ACV. Make it impossible to say no on the way in, and impossible to lowball on the way out.

#05. Order-taking buys you the right to build a real sales org. Don’t confuse the two

Year one, anyone on the Magic School team who wasn’t an engineer was a salesperson. Calendars were packed with inbound. People wanted to do this, what does it cost, let’s go. That worked because the market was pulling so hard. But Adeel is explicit: that wasn’t sales. That was order-taking dressed up as sales, and order-taking has a half-life.

The fall after monetization, he brought on a CRO with three previous K-12 sales orgs under his belt. That hire brought a team, a strategic-AE playbook, prospecting discipline, and the muscle to operate after the halo faded. Year two of selling looks nothing like year one — and any vertical AI company that mistakes order-taking for a sales motion will hit a brick wall the quarter the model stops generating inbound for them.

Action item: Identify the quarter your inbound stops being free. Hire the second-time CRO at least two quarters before that — and let them rebuild the whole motion, not retrofit yours.

#06. Hire technical-enough sellers — but domain-deep ones

Adeel’s hiring filter for go-to-market: domain expertise is the rare ingredient, AI literacy is the trainable one. The buyers he sells to — district CIOs and chiefs of academics — are geeking out on the new Claude release on X. So sellers have to be technical enough to answer a hard question about how the model works. But that’s the trainable half. The other half — having stood at the front of a classroom, having sat through a school-board meeting, having lived a district implementation — is what closes the deal.

Magic School’s solutions architects skew toward people who used to run district technology rollouts themselves. When they walk into a CIO’s office and tell the story of a deployment they actually owned, the conversation stops being about features and starts being about partnership. Customers buy the partnership; the product is the receipt.

Action item: Pull your last five lost deals. Was the gap a feature, or was it that nobody on your side could speak your buyer’s professional language? If it’s the second, change who you’re hiring before you change what you’re building.

A quick word from our incredible partner,
Parafin - the leading embedded capital partner.

• 4.8 star rating on Trustpilot, with over 500 reviews [source]

• 71% CSAT rating across all partners

• 84 NPS

• Serving top Fortune 500 companies, including Amazon, Walmart, and DoorDash

Check Out Parafin

#07. Defensibility is the context you assemble that horizontal models can’t

The pessimistic 2023 thesis was that horizontal models would absorb every workflow as their reasoning got better. That bet hasn’t paid out. Adeel’s frame three years in: the gap between a stock Claude answer and a Magic School answer is everything Magic School knows that Claude can’t — the teacher’s grade, the district’s curriculum, the writing rubric, yesterday’s quiz results, the kid’s IEP. The vertical platform doesn’t compete with the model. It composes context on top of the model.

That’s also the answer to the wave of copycats. PowerSchool, curriculum companies and every legacy K-12 vendor has launched a Magic-School-shaped product — same UI, same labels on the toolbar. They almost never show up in Magic School’s deals. Distribution is owned, brand is owned, context is owned. The model is rented from whoever’s cheapest this quarter.

Action item: List every piece of context your product packages that the underlying model doesn’t have. If the list is short, you’re a wrapper. If it’s long and growing, you’re a vertical platform — defend it like it.

#08. Treat the legacy SoR as a partner, not a rival — until they prove otherwise

Every district has a system of record — PowerSchool and friends. They’ve all shipped copycat AI features. Adeel’s playbook isn’t war; it’s optionality. Some incumbents are going headless on purpose, telling Magic School they’d rather integrate than reinvent — the Salesforce-of-K12 frame. Others are defensive and want to win it themselves. Treat both as legitimate. The defensive ones are answering the same board question every CEO is getting: where is our generative AI revenue?

What earns the option to partner later is winning now. Brand halo, district trust, a product teachers refuse to give up — that’s the moat that eventually forces a serious integration conversation, or makes the incumbent irrelevant in the discussion. The mistake is picking a fight before you’ve earned the leverage.

Action item: Pick the top three systems-of-record in your vertical. Reach out to one this week with an integration proposal. Use the other two as the BATNA in your own product roadmap.

#09. PLG inflection isn’t forever — survive the shape change

Adeel had to do real personal work to accept that not adding 150,000 users a week wasn’t failure — it was math. There are only 3.5 million U.S. teachers. The Slack and Notion arcs are the same shape, just slower: explosive top-of-funnel, then a deliberate transition into enterprise sales, then a new chapter built on a much harder kind of growth. Compress that into AI-era timelines and the founder has to grow up faster than any normal SaaS playbook prepares you for.

The Anthropic vs OpenAI moment played the same theme on a bigger stage. When you hit an inflection, the world tells you everything is possible. The discipline is choosing the one thing — knowledge-work tools, not Disney contracts — and giving up the other ten for now. The companies that don’t make that choice spend their inflection instead of compounding it.

Action item: Write down the one growth metric that defines your next 12 months — and the three exciting things you are explicitly not chasing while you compound it.

#10. Education is hard for VCs and easy for impact — pick your reason and own it

Adeel raised the seed in the worst possible quarter for application-layer AI, in a vertical that has historically produced very few venture-scale outcomes, with a teacher-founder profile no investor had pattern-matched on. Most of the 100 people he met said no. The pattern eventually broke — but only because the company started speaking for itself in the rounds that came after.

His closing frame: literacy and math scores are at their lowest point in decades. If your North Star is impact, there is no better mission to pick. You will not be the VC darling, but the impact at scale is unlike anything else software can do. Vertical AI founders who pick a hard, unfashionable, deeply-mattering vertical are not making a worse career bet — they’re making a different one, and the people they serve will know.

Action item: Write the version of your founder bio that leads with the customer’s outcome, not your fund’s logo. If it reads compelling without a single investor name, you’re underwriting the right thing.

Most of the doomer takes on application-layer AI in 2023 were directionally wrong, and they were wrong in the same direction that the 2024 copycats are still wrong.

The founders actually winning verticals like K-12 didn’t out-engineer the platforms — they out-empathized them.

Live the user’s job.
Then build the easy button only you could build.

If you’re building vertical AI in 2026 — and especially if you’re sitting on a so-called thin wrapper that quietly works — four moves separate the founders who compound from the ones who get eaten:

1. Recruit domain depth onto the cap table. If you haven’t done the user’s job, hire someone who has and give them real equity. Magic School’s go-to-market is 85–90% former educators by design — the product is what falls out, not what gets bolted on.

2. Package every piece of context the horizontal model doesn’t have. Roster, rubric, last week’s data, the workflow your user actually runs. The wrapper isn’t the liability — the wrapper is the durable margin.

3. Use free as a deliberately temporary strategy with an end date. Adeel ran it for eight months and flipped to paid at the peak of the halo. The window is wider in AI than in old SaaS, but it still closes.

4. Price the entry impossible to refuse — and the renewal impossible to lowball. $1,000 pilots that locked districts into full-ticket July renewals underwrote a Series A. Re-write your pilot contract this week.

The vertical AI founders who win this decade will not be the ones who out-engineered the platforms. They will be the ones who out-empathized them — who lived the user’s job before the model existed, and who quietly built the relationship, the brand and the context the model can never own. Be that founder, and the wrapper outlasts the wave.

Watch Episode Here

Listen To Episode Here

Thanks for reading LINEAR. I reply to every email…
Have any questions, feedback, comments? Let us know, we work for you!

Today’s newsletter is sponsored by Xplor Pay. In the Investor-Operator Guide to Embedded Payments, Xplor Pay examines how embedded payments increasingly impacts retention, operational scalability, monetization, and long-term enterprise value across vertical SaaS.

Built from firsthand experience building and scaling more than 20 vertical SaaS platforms, the guide covers why many payment programs underperform after launch, what sophisticated investors increasingly evaluate, and how high-performing SaaS platforms operationalize payments successfully as they scale.
Download the free guide today.

Alright, let’s get to it…

The 4x4 Survival Map — Moat Depth vs. LLM Kill-Zone Risk

Most founders are asking the wrong question about AI.

They want to know if their category will survive. They’re reading think pieces about whether LLMs will replace SaaS.

But here’s what they’re missing: LLMs are not coming for software as a category. They’re coming for specific positions on a strategic map.

This week, I’m breaking down the 4x4 Survival Map, a framework that plots software companies across moat depth and LLM kill-zone risk.

There are 16 positions on this map. Only four are fortresses.

The framework is simple. Two axes. Sixteen positions. Four survival zones.

The Y-Axis: Moat Depth

This measures how defensible your business actually is. Not how defensible you think it is. Not how many features you shipped last quarter. But whether a customer can realistically rip you out and replace you.

Row 1 is Fortress. These are companies that are fortified and incredibly difficult to kill. Network effects, workflow lock-in, regulatory capture, or physical-world integration. The deepest moats in software.

Row 2 is Strong Moat. Real defensibility, but not bulletproof. These companies have switching costs and integration depth, but LLMs can still chip away at the edges.

Row 3 is Thin Moat. Defensible today, fragile tomorrow. These companies are protected by relationships, brand, or execution, but those advantages erode fast when foundation models improve.

Row 4 is Exposed. No moat. Easily displaced by better execution, better pricing, or an AI-native competitor.

The X-Axis: LLM Kill-Zone Risk

This measures how much your product overlaps with what foundation models do natively.

Column 1 is No LLM Overlap. Your business has zero overlap with what LLMs do. Think SpaceX. Think robotics. Think businesses that live in atoms, not bits.

Column 2 is Low Overlap. LLMs touch your product, but they don’t threaten the core. Your moat is the system of record, not the features around it.

Column 3 is High Overlap. LLMs can replicate significant chunks of your value prop. You’re defensible if you own proprietary data, workflow depth, or distribution. But the clock is ticking.

Column 4 is Direct Overlap. You are basically an LLM wrapper or easily replaced by one. This is the kill zone—unless you’re routing across the labs themselves.

The map color-codes risk. The legend tells you everything you need to know. Impregnable means fortified and incredibly difficult to kill. Safe zone means real moat, but not bulletproof. Risky means exposed if the model companies want it. Kill zone means the labs will eat you.

Now let’s walk through each row.

Row 1: Fortress — Impregnable and Incredibly Difficult to Kill

These are companies that LLMs fundamentally cannot displace because their moats are structural, not feature-based.

Audacious Moonshots (No LLM Overlap)

SpaceX. Anduril. Varda. Hadrian.

These businesses operate in the physical world with massive capital requirements, regulatory moats, and government contracts. SpaceX does not worry about GPT-5. Why? Because launching rockets requires billions in capital, decades of engineering iteration, regulatory approval from multiple governments, physical supply chains, and deep relationships with defense customers.

LLMs can help SpaceX engineers write better code. But they cannot replace the business. The moat is physics, capital, and regulatory capture.

If your business requires atoms, not just bits, you’re safe.

AI Roll-Ups (Low Overlap)

Longlake. Crescendo. AGI. Cabana. Thrive Holdings.

These businesses are buying services companies, then layering AI on top to improve margins. The moat is not the software. It’s ownership. You own the companies. You control the customer relationships. LLMs become a tool you deploy, not a competitor you fear.

Longlake is acquiring hoa agencies and embedding AI agents to automate as much as possible. Crescendo doing the same with call centers. OpenAI will not compete with Longlake because OpenAI does not own the agencies, the customer relationships, or the regulatory licenses.

Ownership is a great moat in the AI era.

AI Services (High Overlap)

Crosby. Camber Health. Hanover Park. EvenUp.

These companies have high LLM overlap—their product is often powered by foundation models—but they survive because they sell outcomes, not software.

Camber Health uses AI+humans to automate medical billing end-to-end. The LLM does the heavy lifting. The human gives the rubber stamp. Customers buy Camber because Camber guarantees a 95 percent first-pass claim approval rate and delivers measurable EBITDA uplift.

The moat is trust. You’re liable for the outcome. You’re embedded in the customer’s operations. You combine AI with human QA to ensure accuracy. LLMs can replicate the model. They cannot replicate the trust, the workflow lock-in, the human element, or the guarantee.

Horizontal AI Routing (Direct Overlap)

Genspark. Perplexity.

This is the one that surprises people. Direct LLM overlap, but still a strong moat. Why?

Because the routing layer is structurally defensible as long as the foundation model market stays somewhat fragmented. And the foundation model market is going to have options for a long time.

Here’s the dynamic. Most users do not want to rotate models. They do not want to remember that GPT-5 is better at code, Claude is better at writing, Gemini is better at research, and Grok is better at real-time data. They want one interface. One subscription. One product that just figures out the right model for the job.

That’s the value of horizontal AI routing. You pay a small tax to a router, and in exchange, you get access to every frontier model with the best one selected automatically based on the query.

For a lot of users, that’s worth the tax. Maybe even worth more than the tax.

The only way the routing fortress falls is if one foundation model becomes so dominant that there’s no need to route. That’s not the trajectory we’re on. OpenAI, Anthropic, Google, xAI, Meta, and emerging Chinese labs are all racing. The frontier keeps splitting across models, capabilities, and use cases.

Row 2: Strong Moat — Real Defensibility, But Not Bulletproof

Robotics (No LLM Overlap)

Figure. 1X. Physical world moat.

Robotics lives in the physical world. LLMs can help robots understand natural language and reason about tasks. But they cannot replace the hardware, the control systems, or the deployment complexity. Hardware is hard. That’s a moat.

At some point model companies will likely move into robotics, but its still far out and these companies and there investors are betting on being so far ahead, that when that time comes, it will be too late.

Vertical Systems of Record with Proprietary Data (Low Overlap)

Procore. ServiceTitan. Veeva.

This is the strongest position in SaaS. These companies own the industry-specific workflow. They capture proprietary data. They integrate deeply into business operations.

Ripping out Procore means losing construction project history, retraining every project manager, rebuilding integrations with subcontractors, and breaking reporting dashboards that run the business. The switching cost is brutal.

LLMs will make Procore better. They’ll automate RFPs and predict cost overruns. But they will likely not replace Procore.

Why? Because the moat is system-of-record status. Proprietary workflow data. Workflow lock-in. Integration depth. You are the source of truth.

It’s more likely an Vertical AI takes out procore than an LLM. They are fighting too many bigger battles to focus on anything too industry-specific.

Horizontal Finance (High Overlap)

Stripe. Adyen. Brex. Ramp.

Financial infrastructure has high LLM overlap risk because LLMs can automate reconciliation, expense categorization, and fraud detection. But these companies survive because payments create lock-in. Once you run money through Stripe, switching is painful. Once you roll out corporate cards to all your employees you don’t want to do it again.

The moat is distribution, trust, and capital relationships. Fintech survives if you own the rails, not just the workflow.

Horizontal CRMs (Direct Overlap)

Salesforce. HubSpot.

CRMs have direct LLM overlap. Foundation models can draft emails, summarize calls, update records, and predict churn. So why are they still in the strong moat category?

Data lock-in. Your entire sales history lives in Salesforce. A push to headless. Enterprise inertia means ripping out Salesforce at a 10,000-person company is not going to happen very soon. Seat licenses go down? Yes - already happening. But full churn? I dont think so.

But the moat is eroding. LLMs are making it easier to migrate CRM data. New vertical CRMs are emerging that use AI to automate data entry. The feature advantage Salesforce built over 20 years can be replicated in 6 months by an AI-native startup.

Data lock-in buys time, but not forever.

Row 3: Thin Moat — Risky and Exposed If Model Companies Want It

Legacy White-Collar Services Giants (No LLM Overlap)

McKinsey. Deloitte. EY. KPMG.

Consulting firms have no direct LLM overlap because they sell judgment and relationships. But their moat is thin. Why? Because LLMs are automating the grunt work that junior consultants used to do. Market research. Competitive analysis. Financial modeling.

The partners are safe. The $1,500 per hour strategic advice is still defensible. But the $200 per hour analyst work? That’s getting automated.

McKinsey knows this. That’s why they bought QuantumBlack and are building AI tools to augment consultants.

Horizontal Point Solutions (Low Overlap)

DocuSign. Dropbox. Box.

These companies have low LLM overlap, but thin moats. DocuSign does e-signatures. LLMs do not directly threaten that. But LLMs can automate the workflow around signatures. Drafting contracts. Extracting terms. Routing for approval.

If OpenAI builds a document workflow agent that handles the entire contract lifecycle, what is left for DocuSign? Distribution and integration depth.

Vertical-Specific GPTs (High Overlap)

Open Evidence. Harvey. MagicSchool. GC AI.

These all started as vertical wrappers around foundation models. They survive today because they tune models for specific verticals, build models that only hold industry-specific data, expand into vertical-specific workflows, and own customer relationships.

The only defensible path is to move faster than the labs in their given industry, build deeper workflow integration, expand into multi-product, and capture proprietary data. Open Evidence is doing this right. They started as a clinical decision support tool. Now they’re embedding into EHRs and capturing clinical usage data.

Vertical GPTs survive if they can continue to fine tune AI to be industry-specific, and expand their product surface area beyond GPT’s only.

Horizontal Voice (Direct Overlap)

Wispr and others like it.

Voice AI is a feature, not a business. OpenAI, Google, and Anthropic all have voice APIs. They’re near-free. If your entire product is voice interface to LLMs, you’re in the kill zone.

Row 4: Exposed — The Labs Will Eat You

Vertical Point Solutions (No LLM Overlap)

Vertical BI. Vertical marketing. Vertical Data Connectors.

These have no LLM overlap, but also no moat. They’re easily displaced by better execution or better pricing. If your only defensibility is “we got here first,” you’re exposed.

Horizontal Productivity Suites (Low Overlap)

Notion. Monday. ClickUp. Asana.

Productivity tools have low LLM overlap today, but rising risk. LLMs can generate project plans, summarize tasks, automate workflows, and draft meeting notes.

Notion is already embedding AI everywhere. That’s smart. But it also means the product layer is getting thinner. The moat becomes rip and replace. But if you can one-click move it all over to Claude that’s not a place I want to be.

Horizontal Design Tools (High Overlap)

Figma. Canva.

Generative design is real. LLMs can already generate mockups, create brand assets, and design social media graphics. Canva knows this. That’s why they bought Leonardo.ai for hundreds of millions.

The survival path is to own the creative workflow, not just the design output. Figma is trying to survive because it’s a collaboration tool. The value is the shared workspace, the version history, the comments. But what is stopping LLM’s from doing this eventually? Probably nothing.

Horizontal Content Generation (Direct Overlap)

Jasper. Copy.ai. Writer.

This is the dead zone. These companies are horizontal content wrappers. They take GPT-4, add some prompts, and charge $50 per month. OpenAI has already pretty much eaten this.

The Vertical Software Summit Is Back!

Hear biz stories from 6+ billion dollar vertical ai Founders+CEO’s
We are already 50% booked and still 4.5 months from the Summit!

Don’t miss out by grabbing tickets late :-)

Grab Your Ticket Here

How to Know If Your vSaaS Is in the Kill Zone

Most founders think they have a moat because they have customers. That’s not a moat. That’s momentum. Here’s how to assess whether your vertical software business can survive the LLM era.

Step 1: Map Your Position

Ask two questions.

First, how deep is your moat? Real moats look like network effects, proprietary workflow data, embedded fintech, regulatory compliance, physical-world integration, or system-of-record status. Fake moats are feature velocity, better UX, or being first.

Second, how much does your product overlap with core LLM capabilities? Content generation, summarization, Q&A, and basic automation are all high-overlap areas.

Plot yourself on the grid. Be honest. If you’re in the bottom-right quadrant—exposed with direct overlap—you’re in the kill zone.

Step 2: Identify Your Actual Moat

Real moats in vertical software are structural, not operational.

System-of-record status means you’re the source of truth for critical business data. Proprietary workflow data means your product gets smarter as customers use it. Embedded fintech means you control payments or financial rails. Regulatory compliance means certification or audit trails create switching costs. Network effects mean value increases with more users. Physical-world integration means hardware or supply chain dependencies. Multi-model aggregation means users pay you to abstract away a fragmented market.

Those are real moats. Feature velocity is not. Better customer support is not. Being first is not.

Step 3: Decide Your Survival Strategy

If you’re in the kill zone, you have three options.

Go vertical, fast. Horizontal content generation is dead. Vertical content generation with proprietary context and workflow depth can survive. You need to become the system of record for a specific workflow in a specific industry.

Embed into a system of record. Become a feature inside someone else’s fortress. If you can’t own the customer relationship directly, partner with someone who does.

Or sell outcomes, not software. Become an AI services business. Use the LLM as a tool, not the product. Get paid when the customer wins. Own the liability. Provide guarantees.

If you have a thin moat, deepen it. Add proprietary data capture. Build workflow lock-in. Embed payments. Get closer to the system of record. Move faster than the labs.

If you’re impregnable, stay impregnable. Do not drift toward horizontal. Do not become a feature. Protect the moat. The temptation will be to expand horizontally because the market looks bigger. But horizontal is where you die.

Step 4: Apply the “Could OpenAI Do This?” Test

Ask yourself: could OpenAI replicate 80 percent of my product with a new prompt template and a vertical fine-tune?

If yes, you’re at risk.

If OpenAI would need to acquire companies, build physical infrastructure, navigate regulation, own proprietary data, or aggregate across competing labs to compete with you, you’re safer.

The moat is not the software. The moat is everything around the software that makes it impossible to replicate. The workflow. The integrations. The trust. The liability. The outcome guarantee. The proprietary data. The multi-model abstraction layer.

That’s what survives.

What This Actually Means

The 4x4 Survival Map is not static. Companies move. Moats erode. LLMs get better. Salesforce used to be impregnable. Now it’s vulnerable. Horizontal content tools used to be venture-backable. Now they’re dead.

But the framework holds. Moat depth and LLM overlap determine survival.

Most founders are optimizing for the wrong things. They’re adding features when they should be deepening moats. They’re chasing horizontal scale when they should be going vertical. They’re building AI wrappers when they should be building systems of record or routing layers that abstract the fragmentation.

The AI era does not kill vertical software. It kills undifferentiated horizontal software with thin moats.

If you’re building in vertical software, the opportunity is bigger than ever. But you need to know where you stand. You need to map your position. You need to deepen your moat. You need to move fast.

The labs are coming. The only question is whether they eat you, ignore you, or you become their distribution.

//

So where does your company sit on the 4x4 map? Are you deepening your moat or adding features? Hit reply. I read every response.

Until next week,

Luke

Do me a solid and forward to a friend :-)

The honest version of inception-stage investing in 2026 is that the consensus market has fled to safety. The biggest funds are chasing founders who look de-risked on paper — Fortune 500 résumés, second-time CEOs, ex-FAANG operators — even though the entire historical record of venture says the largest outcomes came from young, scrappy, unfunded founders the consensus would not have touched. Gutter is built as an institutional critique of that drift. The name itself is the bet: if you’re going to call yourself Gutter, you’ve got to be good.

The first signal isn’t pedigree — it’s resourcefulness.

How much has this founder accomplished relative to the resources they actually had? The bar isn’t absolute output. It’s the ratio. A 22-year-old who bootstrapped a software business on internet-poker winnings gets credit a freshly-funded ex-VP does not. And it’s the deepest predictor in Gutter’s portfolio of who actually compounds.

The second is lifelong commitment — the founders whose attachment to the problem transcends any single company. Dan’s phrase: we couldn’t pay them not to do it. They grew up with family inside the industry, or worked in it themselves, or have been obsessed since they were a kid. That commitment is what survives a pandemic, a hyperinflation cycle, and AGI showing up mid-sprint. The cleanest pitch with the thinnest tether to the customer is the founder who quits the second the company gets actually hard.

The third is evidence density. When Gutter backtested every Elbow Grease application against which ones converted to an interview, the single dominant variable was hard evidence per word in the pitch. Named customers. Recorded calls. Specific numbers. \”We reduced X to Y for Bob, here’s his phone number, by the way Bob has a cousin Jim, and Jim uses it too.\” That is the texture of a founder who is actually customer-obsessed — and it’s almost impossible to fake against 40 hours of diligence. The rest of this issue is the operating system underneath those three.

Watch Episode Here

Listen To Episode Here

This Weeks Vertical Titan:
Dan Teran & James Gettinger
(Co-Founder & GPs @ Gutter Capital)

Dan and James met playing rugby at Johns Hopkins and spent a decade on divergent careers before founding Gutter together five years ago. Dan was a partner at Pre Hype in New York, then founded Managed by Q — a vertical SaaS and B2B marketplace for office services — which he ran as CEO for five years before selling to WeWork in 2019 for $220M. James started a software company at 22, learned every lesson the hard way, and bootstrapped it by playing internet poker, eventually becoming one of the largest winners in online poker and DFS on sites like DraftKings and FanDuel before turning angel.

Gutter just announced Fund III — a $75M vehicle focused on the earliest stages of company building — and opened applications for the second cohort of Elbow Grease, their in-house accelerator. Cohort one funded an ex-SpaceX engineer putting satellites into very low earth orbit and Keeper Systems, AI agents for 211, the non-emergency hotline for people facing eviction or food insecurity. The Gutter HQ on Canal Street in Chinatown houses 70+ operators across 15 portfolio companies on two floors — the obstacle course Elbow Grease founders are dropped into for ten weeks.

The frame they keep coming back to: Gutter is anti-prestige to the bone. The consensus market wants founders who look de-risked on paper. Gutter wants founders who have done a tremendous amount with very few resources — the ratio, not the absolute. James calls it the most underweighted variable in venture today.

The line every founder should write on a wall: Dan and James are looking for teams they “couldn’t pay not to do it.” Authenticity is at an all-time premium. Gutter does not publish requests for startups — they don’t want founders studying to the test. The aperture is wider than the vertical-AI label suggests: if you’re doing your life’s work, they want to hear about it.

Watch Episode Here

Listen To Episode Here

Ten moves for founders before the seed. How investors think about backing early entrepreneurs.

Operator-grade lessons for the things you’re actually deciding before your first real check — what signal beats pedigree, how to pitch with evidence density, what to send Gutter before they ask, and why doing your life’s work is now the most underpriced asset in venture.

#01. Refuse the pedigree shortcut — back the resourceful

Most VC money in 2026 still chases founders who look de-risked on paper: Fortune 500 résumés, ex-FAANG, second-time founders battle-tested by a tier-one logo. James calls it out plainly: that profile is in direct conflict with how most of the great companies actually got built. Jobs, Bezos, Gates — young, scrappy, no one would give them the time of day. The flight to safety at the top of the market has quietly left the highest-variance founders unfunded.

Gutter’s filter for signal-without-pedigree is resourcefulness: how much did this person accomplish relative to the resources they had? A 22-year-old who bootstrapped a software business by playing internet poker on the side gets credit a Stanford MBA with a $2M friends-and-family round does not. The point isn’t to fetishize struggle — it’s to underwrite a real input to outcomes that the consensus market has stopped pricing.

Action item: Stop putting your prestige logos at the top of the deck. Put a slide that shows what you accomplished per dollar and per month. If that ratio isn’t extraordinary, fix it before you raise.

#02. Lifelong commitment — “we couldn’t pay them not to do it”

Dan’s single most-used phrase for inception-stage founders: we couldn’t pay them not to do it. The best-performing founders in the Gutter portfolio carry a commitment to the problem that transcends any given company — they grew up with family in the industry, they worked in it themselves, or they have been obsessed since they were a kid. That commitment is what survives a pandemic, a hyperinflation cycle, and AGI arriving mid-sprint.

Charisma without that commitment has a short half-life. Founders who can tell a beautiful narrative but parachuted into the space last quarter tend to leave for the next hot thing the second the company gets genuinely hard. The companies Gutter has watched die were often the ones with the cleanest pitch and the thinnest tether to the customer.

Action item: Write the version of your origin story where you do not get funded — and still keep working on this for ten years. If you can’t write it credibly, you have an investor problem, not a market problem.

#03. Drive + pace of learning > airplane-test charisma

The market over-rewards founders who are charismatic and agreeable — the ones you’d want next to you on a long flight. James’s blunt observation: many of the people who have actually done it would not pass that test. What does correlate with outcome is drive (the initiation energy plus the refusal to quit the hundreds of times the ball is on the fence) and pace of learning. The job changes shape at every revenue inflection — $300K, $1M, $10M, $100M, $1B — and the founder has to change with it.

When Don Valentine first met Steve Jobs at Sequoia, he called the person who made the intro and said: don’t introduce me to more people. Jobs was not yet the CEO we remember; he became him on a 30-year journey. The Gutter question isn’t “are they great today” — it’s “are they on a trajectory steep enough that they will be”.

Action item: Track your own learning velocity. List the three biggest things you understand now that you didn’t 60 days ago. If the list is thin, your trajectory is too.

#04. Evidence density is the single biggest predictor of a yes

When Gutter backtested every Elbow Grease application against which founders made it to interview, one metric dominated: evidence density. Hard evidence per word in the pitch. Specific customer names, exact numbers, recorded calls, real testimonials — not a wall of text claiming “customers love it”. The founders who advance pile on primary evidence that they understand the space better than anyone in the room.

Dan’s mental model: “we reduced X to Y for Bob, here’s his phone number, by the way Bob has a cousin Jim, and Jim uses it too.” You pull the string and the founder keeps going. That is the texture of a founder who is actually customer-obsessed — and it’s almost impossible to fake for the 40 hours of diligence Gutter runs on each seed investment.

Action item: Read your last pitch deck and count the named customers, the specific numbers, and the exact quotes. If that count is under ten, your deck is hand-waving — go talk to fifty more people this month.

#05. Send three recorded customer calls — before they ask

Dan’s pro tip for any founder applying to Elbow Grease (or anywhere): in addition to a product demo, attach three recorded customer calls. A pitch to a non-customer. A QBR with a current customer. A feature walkthrough. They don’t need to be polished — they need to be live, dynamic, unscripted, and obviously real.

Most investors won’t actually listen to all three. But knowing they could is a different kind of signal. It tells the room you are so deep in the work that the calls are happening daily anyway, and you’re not hiding anything. In a world where every meeting is transcribed and recorded, the founders who don’t share customer interaction are the ones quietly admitting it isn’t happening yet.

Action item: Record your next five customer conversations (with permission). Pick three. Send them, unprompted, the next time an inception-stage investor takes a meeting.

#06. Judgment is the hardest signal — Elbow Grease is the obstacle course

Drive and pace of learning you can screen for in a couple of meetings. Judgment — do they have a history of good decisions, can they explain why a past mistake was a mistake, will they actually learn from it — is much harder. James and Dan can’t see it in a Zoom. So they built Elbow Grease, a 10-week accelerator inside Gutter HQ on Canal Street, specifically to put founders through an obstacle course of real decisions in the same room as 70+ other operators across 15 portfolio companies.

They watch how the founder connects with other founders. How they enlist later-stage CEOs to introduce them to customers and talent. How they handle a no-decision. “You would be blown away at how effective some of the Elbow Grease founders have been at just enlisting these much later-stage founders to do work for them — they are demonstrating their resourcefulness before our very eyes.” Demo day at the end? No. The next check comes from Gutter, or it doesn’t come.

Action item: Manufacture your own obstacle course. Spend the next month embedded somewhere that forces you into uncomfortable judgment calls weekly — a customer’s office, a peer’s company, a co-working space full of operators ahead of you.

A quick word from our incredible partner,
Parafin - the leading embedded capital partner.

• 4.8 star rating on Trustpilot, with over 500 reviews [source]

• 71% CSAT rating across all partners

• 84 NPS

• Serving top Fortune 500 companies, including Amazon, Walmart, and DoorDash

Check Out Parafin

#07. Reference founders precisely — “would I bet on them, not rehire them”

Gutter does heavy reference work — talking to dozens of people who actually know the founder. But the lesson is in how the references get interpreted. The wrong question is “would you hire this person back into the job they had?” The right question is “would you bet on them as a founder?” Dan’s own references in 2014 — pre-Managed by Q, fresh out of a paralegal job — would not have read as exceptional. They didn’t need to. Scott Belsky, Hunter Walk, and Satya Patel were betting on the founder, not the employee.

When a reference does come back mixed, the question is never “is this disqualifying” — it’s “what are the specific circumstances in which the next 18 months would be different?” References have killed plenty of Gutter processes. But they’ve killed the right ones, by being specific about context and what was actually being measured.

Action item: Pre-write the reference questions you want investors to ask about you. Make them specific: founding context, judgment under uncertainty, what you learned from the worst quarter. Send them to your references, not the investor.

#08. Hunt the weird verticals where AI has dropped the cost of being there

Gutter’s first Elbow Grease cohort included a former SpaceX / Northrop Grumman engineer putting cheap satellites into very low earth orbit, and Keeper Systems — AI agents for 211, the non-emergency hotline for people facing eviction or food insecurity. Neither was in Gutter’s stated thesis. Both got funded because the application quality forced the fund to broaden its aperture, and because the founders had domain depth no Stanford generalist could fake.

James’s frame: AI is dropping transaction costs across the entire economy, and “it feels like the entire economy is up for grabs.” The old rule that you need a big proven TAM dissolves when you can credibly displace a meaningful chunk of human labor inside a small market. Niche is no longer a death sentence — it’s increasingly where the most interesting Elbow Grease companies are showing up.

Action item: Make a list of verticals you actually understand that no Sand Hill partner would ever fund. Rank by how much human labor is currently doing work AI could now do credibly. Go where the score is highest.

#09. Don’t fetishize youth — sometimes the mid-career parent is the unlock

The fund is skewing younger; that’s real. Cheap software development means a 22-year-old operations person can ship product without engineers. But Dan is explicit about the trap: don’t fetishize youth. Some of the most prolific vibe-coders in the Elbow Grease cohort were not 19-year-olds — they were mid-career operators who had been thinking about a problem their entire careers, and the AI toolkit finally cut the dependency on engineers they’d been chafed by for two decades.

The archetype: the founder of Punch, a father of five (sixth on the way) in Atlanta who got permission from his family to spend ten weeks in the Canal Street office. He shipped product faster than almost anyone in the Gutter portfolio because his attitude was “I know this is my shot.” The right founder profile is not an age — it’s the person for whom this set of tools, right now, is the difference between never building and finally building.

Action item: Audit the founders you admire from the last 18 months by life stage, not by Twitter age. Identify the underrated profile in your network — and either back them or recruit them.

#10. Do your life’s work — authenticity is at an all-time premium

Gutter does not publish requests for startups. The reason is deliberate: they don’t want founders studying to the test. Dan’s closing advice — to Elbow Grease applicants and to everyone else — is that authenticity is at an all-time premium. Founders who are doing the thing only they can do get further, faster, against a backdrop where everyone else is pattern-matching to whatever a16z tweeted last week.

The aperture is wider than the AI label suggests. Gutter does the vertical AI work, but funded a satellites company and a 211 agents company in the same cohort because both founders were obviously on their life’s work. The question isn’t “is this a thesis area” — it’s “is this person doing the thing that, if you took venture capital off the board entirely, they would still be doing.”

Action item: Write the one-page version of why you are the only person who could be building this. If it reads generic, change the company, not the deck.

The pedigreed founders Gutter passes on,
and the ones they back the check on.

The Gutter founder matrix maps the two variables that drive inception-stage outcomes in their portfolio: relentless drive on the vertical axis, customer obsession and domain depth on the horizontal. The top-right corner is where the check gets written. The bottom-left is where references and evidence density tank the process. Most of the consensus market is busy underwriting the top-left.

We closed the episode by putting Nikhil — and Nic — in the seat and pricing the map in real time: which of the most highly valued vertical AI bets are sitting inside the kill zone, and which still have a clean moat, a real network, or an unloved vertical underneath them. Below: where on the map each lands, and the operator-grade reasoning underneath.

The pattern is uncomfortable but consistent. The hype-cycle category of any given year almost never produces the biggest outcome of that year. The biggest outcome comes from the category everyone agreed was dead, niche, or already won. If your whole thesis is in the consensus column, you’re not actually playing for power-law returns — you’re playing for second place.

The patterns that show up in the pitches Gutter passes on.

• Walls of text without names or numbers

  If the pitch can’t point to specific customers, exact metrics, or a phone number Gutter could actually call, it’s hand-waving. Evidence density was the single biggest predictor of who got an interview.

• RFS chasers

  Founders building whatever VCs are tweeting about this month. Gutter deliberately doesn’t publish requests for startups — they don’t want anyone studying to the test.

• Charisma without commitment

  Beautiful narrative, three-month tether to the space. The half-life of caring predicts the half-life of running. The pitch survives a meeting; the company doesn’t survive the next downturn.

• Pedigreed but unscarred

  Fortune 500 résumés where the line on the deck is the credential, not the accomplishment-per-dollar. The consensus market overpays for this; Gutter is built to pass on it.

None of these are character verdicts. They’re all repairable — but the founders who don’t notice them in their own pitch are the ones who don’t make it through the funnel.

The texture Dan & James see in the founders they actually back.

• Resourcefulness-per-dollar

  Founders who accomplished something extraordinary with almost nothing — and can prove it. The ratio, not the absolute. The most underweighted variable in the consensus market.

• Lifelong commitment to the problem

  Family in the industry, years in the back office, or an obsession since childhood. The kind of founder you couldn’t pay not to do this — even with venture off the board entirely.

• Evidence-rich, name-game pitches

  Specific customers, real numbers, three recorded calls attached unprompted. The founder pulls the string and just keeps going.

• Demonstrated judgment in the wild

  Enlists later-stage CEOs to do work for them. Tells a slow-moving customer she’s flying in Friday for a decision — and they send the address. Drive plus pace of learning, in real time.

The common thread: the founder has clearly been doing the work long before the deck got written, and the evidence is everywhere in the pitch. Gutter doesn’t have to imagine the company — they can see it operating.

The operator takeaway from a 40-hour Gutter diligence: pedigree is the cheapest input you can fake — evidence, drive and judgment are the ones you can’t. The check goes to whoever stacks the latter three on the same page.

The most uncomfortable line of the conversation: most consensus VC funds in 2026 are pattern-matching to founders who look de-risked on paper. The historical record says that’s the opposite of where the biggest outcomes come from. Gutter is a deliberate bet on the founders the consensus market is quietly leaving on the table.

Watch Episode Here

Listen To Episode Here

Thanks for reading LINEAR. I reply to every email…
Have any questions, feedback, comments? Let us know, we work for you!

Today’s newsletter is sponsored by Xplor Pay. In the Investor-Operator Guide to Embedded Payments, Xplor Pay examines how embedded payments increasingly impacts retention, operational scalability, monetization, and long-term enterprise value across vertical SaaS.

Built from firsthand experience building and scaling more than 20 vertical SaaS platforms, the guide covers why many payment programs underperform after launch, what sophisticated investors increasingly evaluate, and how high-performing SaaS platforms operationalize payments successfully as they scale.
Download the free guide today.

Alright, let’s get to it…

Everyone Needs To Have A Quota

I’ve been digging deep into all sorts of vertical businesses lately with their management teams.

I thought it was pretty standard for sales people to be have quotas, and for non-sales people to have bonuses based on their performance / the companies performance.

It’s interesting to see the amount of early stage companies, I’m talking anyone sub $10M that don’t have this or a similar framework.

If that’s you, it is literally the easiest / most important lever you can quickly plug into your business.

If you already have some resemblance of it, probably still worth reading as I’m going to go into what I did after 10 years of trials and tribulations that worked.

IT’S SO SIMPLE.
Highly recommend NOT making it complicated.

#1. Everyone has a quantitative number they are gunning for EVERY QUARTER
That natural push by employees is to have qualitative goals, and it’s also to have a bunch of them. This just allows for them to hide behind something. Don’t let this happen. REGARDLESS of their department, make it quantitative. Make it black and white. Even if they are in HR, you can give them something quantitative IE an eNPS score. 30 days into working with me everyone knew I hated/didn’t believe in qualitative goals.

#2. 50% of everyones bonus is tied to the attainment of this number

You can’t just give people a goal. You have to incentivize them. You have to reward them. Even if every non sales person is comp’d 90% base, 10% bonus, it’s better than nothing. Humans are funny, they will fight hard for that extra 10%. Do SOMETHING.

#3. Your company needs to have a quantitiatve number they are gunning for EVERY QUARTER

For most companies, this number is usually ARR. But whatever you choose, it’s important that you show it EVERYWHERE. That folks can always see how they are doing.

#4. The remaining 50% of everyones bonus is tied to the team achieving said goal.
If the company doesn’t attain AT LEAST 70% of their quarterly plan, no bonus for anyone. It makes no sense to pay people out when the company isn’t performing. This ties in the team dynamic you don’t want to lose in these types of plans.

#5. Sales people are comp’d differently, ideally 50% base, 50% commission. Commission plan should be scaling IE if their quota is $1M of ARR per year, they get 6% for the first $250K of ARR signed, 7% for the next $250K of ARR signed, 8% for the next $250K of ARR signed, 9% for the next $250K of ARR signed, 12% for everything signed thereafter.

Overtime, you can adjust this. For example if your laser focused on growth and profitability, you can bonus people 50% their own goal, 25% company ARR goal, and 25% company EBITDA goal. We definetly did this based on the stage of the business.

So to wrap it up

People will give you a bunch of reasons why it won’t work, why it doesn’t make sense for X department or Y person. It’s all bullshit. In the rare events that actually happens just correct it. Get the baseline data and work from there.

NO MATTER WHAT do something.
Just make sure it’s quantitative.
Make sure there is a cash component tied to it.
And make sure there is a clear company goal for every quarter.

This alone will work wonders for you company.

Quick PSA —
Have the GOAT of customer success breaking down how to get your company to 120% net dollar retention on a live webinar this Tuesday. This is especially important in the AI age as retention is getting harder and harder.

Limited to first 25 founder sign-ups.

Register here

Adaptive Innovations:
A New AI Roll Up For Home HealthCare

If you want to understand where Vertical AI gets really interesting, don’t look at another copiloting layer slapped onto legacy software.

Look at a company that said: we are not going to sell software to the incumbents — we are going to become the operator.

That’s what Adaptive Innovations is doing in home health.

And I think it’s one of the clearest examples yet of what an AI-native roll-up can look like in the real world. Adaptive isn’t pitching itself as a better dashboard, a clever note-taking assistant, or a workflow add-on. It is pitching itself as the AI-native healthcare provider rebuilding the way care is delivered in America, starting with home health. That distinction matters a lot.

The big idea

Most software companies try to improve a broken workflow.

Adaptive is trying to own the workflow, automate the bureaucracy around it, and then use those savings to serve more patients, pay clinicians better, and compound operational advantage.

That is a very different ambition.

In home health, the market is massive, the demand is growing, and the administrative drag is absurd. Adaptive says home health is a $140B industry, that 40% of home health patients never receive care, and that around 40% of industry spend goes to admin. Felicis frames the problem even more sharply: over $100B is spent annually on home health, 40%–50% goes to coordination overhead, and roughly $40B in referrals get rejected every year because the administrative math doesn’t work.

That is why this story matters.

This is not “AI for paperwork.”

This is AI as the basis for a new operating model in a market where the incumbents were built for a different era.

Why home health is such a compelling market

A lot of founders look at home health and see a messy, regulated category.

That’s true.

But the best markets often look messy from the outside.

Home-based care sits at the center of multiple long-term tailwinds:

• the aging of the U.S. population,

• staffing shortages,

• rising costs in institutional care,

• payer pressure for better outcomes,

• and strong patient preference to stay at home.

The demographic case alone is enormous. Americans age 65+ are projected to grow from 58 million in 2022 to 82 million by 2050, and that cohort’s share of the population is expected to rise from 17% to 23%. On top of that, Alzheimer’s prevalence could more than double to 13 million by 2050. This is exactly the kind of demand curve that makes “last-mile care delivery” a generational business opportunity.

And unlike many sexy AI categories, this isn’t a market that depends on creating new demand.

The demand is already here. The system just can’t serve it efficiently.

The real bottleneck is not care. It’s coordination.

This is the most important insight in the entire story.

Home health doesn’t fundamentally break because clinicians don’t exist or because patients don’t need care.

It breaks because the administrative stack surrounding care is so heavy that providers reject patients who are clinically eligible.

Felicis lays it out plainly: two out of every five patients referred to home health never get treated not because they shouldn’t be treated, but because intake, scheduling, prior auth, documentation, and billing make the economics unattractive. In other words, the problem is not clinical demand. The problem is delivery infrastructure.

That is a huge unlock.

Because once you identify that the bottleneck is administrative, not medical, the solution changes.

You stop asking, “How do I sell software to providers?”

And you start asking, “How do I build a provider that is structurally better because AI is embedded into every non-clinical workflow?”

Adaptive’s core strategic move: become the provider

This is the move.

Adaptive’s founders reportedly first looked at home health through an AI lens and initially tried to improve the back office from the outside. But they kept hitting the same wall: the agencies were too calcified to absorb the redesign. So they changed the thesis.

Instead of selling tools into the system, they decided to become the system.

That is why I think “AI-native roll-up” is the right framing here.

Adaptive is not just software. Adaptive is not just services. Adaptive is not just an AI assistant.

It is an AI-native operator in a fragmented, high-friction, regulated market.

Home Health Care News says Adaptive uses AI across intake, eligibility, scheduling, charting, coding, QA, and other back-office functions. D Magazine says the company acts as the provider while using proprietary AI to run intake, scheduling, documentation, and billing. That gives Adaptive end-to-end control over both the care-delivery layer and the tech layer.

That is the key strategic difference between vendor AI and operator AI.

Vendor AI sells efficiency.

Operator AI captures the economics of efficiency.

This is what the best AI-native companies are starting to realize

If you only sell software into a broken system, you often inherit the adoption bottlenecks of that system.

The provider is busy. The workflows are entrenched. The incentives are messy. The procurement cycle is slow. And even if your tool works, the operator may not change behavior enough to unlock the full value.

Adaptive’s founders seem to have recognized that the highest leverage point was not selling intelligence to providers but embedding intelligence inside a provider they controlled.

That’s a much harder company to build.

But it’s also a much more ambitious one.

Because if you control the workflow, you get:

• the data,

• the margin,

• the speed of iteration,

• the customer experience,

• and the operational flywheel.

That is how you go from “AI feature” to “category-defining business.”

The founding team is what this model requires

This is not a business you can fake your way into.

Felicis says the founding team includes:

• Alex Wendland and Ryan Tolsma, who came at the market from an AI lens,

• Logan Stinson, who spent years scaling home health operations across Texas,

• and Hunter Stinson, a former nurse and U.S. Army Ranger officer who had worked directly within these broken workflows.

That blend matters.

Because this is not a “two cracked engineers and a prompt” company.

This is a business that requires:

• clinical credibility,

• operational density,

• reimbursement fluency,

• local-market execution,

• and technical talent strong enough to automate ugly real-world workflows.

Adaptive also says it has assembled talent from places like Citadel, Jane Street, Scale AI, and Palantir, while Home Health Care News reports a 30+ person engineering team building the tooling in-house.

That is the right recipe for an AI-native operator: domain operators + clinical talent + serious engineers.

The traction is what makes this story go from interesting to important

A lot of AI-native care companies sound compelling in theory.

Adaptive is more interesting because there are already real operating metrics behind the pitch.

Across its own site, Felicis, Home Health Care News, and D Magazine, the company is reported to have:

• delivered 100,000+ patient visits,

• built a network of 500+ referring healthcare organizations,

• employed 200+ clinicians,

• operated with 24/7 clinical coverage across five Texas regions,

• reduced clinician documentation time by around 80%,

• achieved a 4.9% rehospitalization rate,

• and admitted 99% of patients within 48 hours of referral.

Felicis also cites:

• 4.9% rehospitalization vs 12.9% Texas average and 10.2% national average,

• 95%+ improvement in ambulation vs 89.5% national average,

• and a 4.5-star CMS quality rating against an industry average of roughly 3.

Those are not “we think AI could help” metrics.

Those are operating metrics.

And that is the whole point.

The economic flywheel is the real product

The best part of this business is not the AI layer itself.

It is the flywheel the AI layer makes possible.

Home Health Care News quotes Alex Wendland saying the company’s goal is to drive admin costs to zero, then recycle those savings into higher clinician pay and a compounding flywheel that lets it serve the full market. Adaptive also describes itself as payer-agnostic, saying its lower admin overhead allows it to accept patients regardless of insurer and work across Medicare, Medicare Advantage, commercial insurance, fee-for-service, episodic, and value-based models.

This is why the model is so attractive.

If you reduce admin friction enough, you can:

1. accept patients others reject,

2. improve response time,

3. improve clinician economics,

4. attract better labor,

5. generate better outcomes,

6. deepen payer and referral relationships,

7. and compound scale.

That is a serious operating flywheel.

And unlike many AI businesses, it is anchored in something very concrete: the removal of bureaucratic labor from a system drowning in it.

This market is structurally built for roll-ups

Another reason I like this story: the market structure fits the strategy.

The home health sector is fragmented enough that both PE and strategic buyers have been consolidating it for years. The PMC study identified 749 unique home health agencies involved in PE transactions from 2006 to 2024, with acquisitions clustering in waves and heavily concentrated in the South. Capstone also says private company buyers represented 67.2% of strategic volume in 2025 because of the sector’s fragmented composition, while overall home care M&A activity rose 40.5% YoY to 104 transactions.

That matters because it tells you two things:

1. The market is fragmented enough to consolidate.
2. Buyers already understand the density game.

What Adaptive appears to be doing is not the old PE playbook of buying local operators and layering in modest cost discipline.

It is a new version:

• build the AI-native operating system,

• use it inside owned care delivery,

• improve economics and outcomes,

• then scale across regions and potentially service lines.

That is a much more interesting form of roll-up than “buy 20 branches and centralize accounting.”

Funding: the capital markets are starting to recognize the difference

Adaptive emerged from stealth with $60 million total funding, including a $50 million Series A led by Felicis and a previously undisclosed $10 million seed. Other investors include Bain Capital Ventures, Optum Ventures, Sunflower Capital, Conviction, BoxGroup, Dorm Room Fund, Constellation, and angel investors from OpenAI.

That syndicate tells you what the market sees here.

This is not just a software play. This is not just a healthcare services play. And it is not just an AI infra play.

It sits in the middle of all three.

That’s exactly why it can become important.

The broader lesson: in AI, the biggest opportunities may belong to operators, not vendors

This is the lesson I keep coming back to.

A lot of AI founders still think the best move is to sit one layer above the incumbent, sell productivity, and hope adoption follows.

But in markets like home health, the real prize may go to the company that says:

“We are going to own the workflow, own the economics, and own the customer experience.”

That is much harder.

But the reward can also be much bigger.

Now let’s talk about Special

Because Adaptive is not the only company seeing the opportunity.

On June 2, former DOGE operatives Nate Cavanaugh and Justin Fox announced Special, an AI company they describe as building an operating system to transform critical American industries through AI. Their financing was led by Andreessen Horowitz. The round size was not disclosed in the sources I reviewed.

Their thesis sounds very familiar:

• massive inefficiency in essential service businesses,

• AI as the engine for productivity,

• and vertical integration through acquisition as the way to capture value.

Their first vertical is Figure Health, focused on the aging population. Special says it is already under contract for its first acquisition in Texas, serving 1,400+ patients and employing hundreds of nurses. It plans to use SpecialOS to improve efficiency and increase nurse pay.

So yes — there are now multiple serious teams converging on roughly the same insight:

Critical service industries with labor constraints + regulatory complexity + old workflows are perfect terrain for AI-native operators.

But here’s the key difference from traditional vertical software

Adaptive is starting with home health. Special is starting with elder care via Figure Health. Those are adjacent, not identical, bets.

And even within home-based care, the sector is large enough to support multiple scaled businesses. Capstone highlights the entire home care continuum — personal home care, home health, hospice — as a major and growing part of U.S. healthcare infrastructure, while Mordor estimates the home healthcare software market alone at $5.08B in 2026, growing to $9.27B by 2031. The care-delivery market itself is obviously much larger than the software layer.

That means this is not a “winner-take-all app category” like we typically see in vertical software. When you actually own the service business themselves, theres a lot more room…

My take on Adaptive vs. Special

I think these companies are reading the same macro correctly.

The next wave of great AI businesses will not just be better SaaS tools.

They will be AI-native operators in sectors where:

• labor is expensive,

• admin is suffocating,

• outcomes matter,

• compliance matters,

• and local execution matters.

Adaptive seems more clinically and operationally specific today. It already has meaningful reported traction in Texas, clear patient-outcome metrics, and a tightly defined wedge in home health.

Special feels broader and earlier — more of a platform thesis around “Main Street services” with elder care as its first proving ground. It is more explicitly ideological and acquisition-first in how it frames the opportunity.

But I don’t think the right takeaway is “pick one.”

I think the takeaway is the category is real.

And once multiple smart, well-capitalized teams start attacking the same ugly operating problem from different angles, that is usually a sign the market is bigger than people think.

The real opportunity

The best AI companies in healthcare may not look like software companies at all.

They may look like:

• providers,

• care networks,

• staffing systems,

• revenue engines,

• and compliance machines

That just happen to be built on top of deeply integrated AI.

That’s why Adaptive is so interesting.

It is not selling “AI for home health.”

It is trying to build the AI-native home health provider.

And if that works, the upside is not a slightly better software multiple.

The upside is building a new kind of healthcare company altogether.

Final thought

I think a lot of people still underestimate how many categories will be rebuilt not by software vendors, but by AI-native operators who own the workflow end-to-end.

Adaptive is one of the best examples I’ve seen so far.

And Special’s raise is further proof that serious capital is now chasing the same idea.

Home health, elder care, hospice, and adjacent categories are not too small. They are not too messy. They are not too regulated.

They are exactly the kind of markets where deep operational pain + demographic demand + AI automation can create very large companies.

That’s what makes this time so exciting.

Industry expertise is becoming MORE important in the AI era, not less

A lot of people think the AI era means the winning founders will just be the best engineers.

I think the opposite.

The software layer is getting easier to build every quarter. Models are better. coding is faster. infrastructure is more accessible. You can compress months of product work into weeks now.

But the industry layer?
That is still brutally hard.

You can learn Cursor.
You can learn Claude.
You can learn how to ship product faster.

It is much harder to learn:

• how claims actually move through a healthcare workflow

• how reimbursements break economics

• which compliance steps are “must have” vs “nice to have”

• who really makes the buying decision

• where the workflow actually breaks at 4:47 PM on a Tuesday

That stuff is usually in the founder’s DNA, or it isn’t. Bessemer makes this point really clearly: vertical AI rewards insider expertise more than horizontal SaaS did because you are reimagining nuanced, regulated workflows, not just digitizing generic tasks. Greylock says something similar: deep domain-specific focus is one of the best ways to build a durable moat at the AI application layer.

So I think the advantage in this era shifts toward teams with industry-native pattern recognition.

Why?

Because when software gets cheaper, the scarce asset becomes:

• trust

• workflow understanding

• proprietary data

• edge-case intuition

• speed of customer empathy

That’s why pure technologists are often at a disadvantage in regulated verticals. The bottleneck is no longer “can you build the tool?” The bottleneck is “do you understand the work deeply enough to know what should be built?” Greylock explicitly argues that founding teams with both domain experience and technology background are advantaged, especially in regulated industries.

My simple takeaway:

In 2018, the hard part was building software.
In 2026, the hard part is building the right software for a very specific workflow.

And the founders who move fastest will usually be the ones who already know:

• the language of the customer

• the pain that actually matters

• the ugly workaround everyone pretends is fine

• the metric that gets the buyer promoted

That’s why I’d rather back a great operator from the industry who learns AI than a great AI builder who has to spend 24 months learning the industry from scratch.

Because in vertical AI, context compounds faster than code.

If you made it thus far, go check out the 2026 Vertical Software Summit. We’ll have 400+ vertical founders/operators/investors in Miami in November. Two days, 6+ billion dollar vertical founders. The Vertical AI event of the year.

Check Out The Summit

Do me a solid and forward to a friend :-)

Nic and I sat down with Nikhil Basu Trivedi, co-founder and GP of Footwork VC and a 16-year VC veteran (early Canva, Lattice, Frame.io), on the question every operator is quietly asking: in a market where the model companies eat horizontal categories overnight and 20 funded competitors crowd every obvious idea, where are the parts of the map you can actually survive on? We map the kill zones to avoid, the moats that still compound (brand, networks, workflow trust), the unloved low-NPS verticals quietly ripe for AI-native challengers, and the consumer-meets-enterprise zone where the biggest companies of the cycle are already being built.

Most of the map is a kill zone.
Find the opportunity.

The honest version of building in vertical AI today is that most of the obvious map is hostile territory. The model companies eat horizontal workflows on a quarterly cadence. The consensus categories are auctions with 20 funded competitors before the round even closes. The old SaaS moats — features, integrations, a tidy ICP — get re-priced overnight. Nikhil’s frame for cutting through it isn’t a thesis, it’s a survival map: a small set of zones on the board where a vertical AI company can actually compound, and a much larger set you should refuse to play on at all.

The first survival question is the kill zone: how badly does your product overlap with what Anthropic or OpenAI is going to ship horizontally next year? Pure document review, generic Q&A, white-collar workflows that live happily in a Claude tab — the labs will eat. The survivors are products that need proprietary data, complicated integrations, or workflows that simply do not fit a chat box. If a one-pager titled “why a Claude tab can’t do this in 12 months” writes itself, you’re on the safe side of the map. If it doesn’t, change the product before you raise.

The second is the moats that still compound: in this era, very few do. Brand is the one technologists chronically underrate and the one Nikhil is most bullish on — the iPhone effect, where users simply do not switch off the products they trust, even when something technically better ships. Real network effects are the rarest and most valuable; the next LinkedIn, the next ACV Auctions. And there is workflow trust in traditional industries — the quiet truth that, once a credit union or a pharma company actually likes you, they do not want to re-evaluate every six months. Build one of these or you’re racing on execution speed alone, which is a brutal place to live.

The third is where on the map to actually build: the unloved low-NPS verticals everyone else ignored. CPG (a $3T+ category with 10,000 brands doing $100M+, almost no software at scale). Credit unions. Blood testing. Pharma research. Accounting back-offices. Antifragile demand, terrible incumbents, no real software heritage — and the buyer is finally ready, because their staff is using ChatGPT at home and asking why their forty-year-old EHR can’t do the same. And quietly underneath all of it, the founders Nikhil keeps backing share one trait: years of having lived in or studied the buyer. Domain depth, not pedigree, is the unlock. The rest of this issue is the full map.

Watch Episode Here

Listen To Episode Here

This Weeks Vertical Titan:
Nikhil Basu Trivedi
(Co-Founder & GP @ Footwork VC)

Nikhil started his VC career nearly two decades ago. He co-founded Footwork five years ago with one partner and built it deliberately small: just the two GPs as the entire investment team, leading the seed or Series A in roughly five companies a year. Footwork is now investing its second fund — $225M, following a $175M first fund — and has made 25 investments to date.

Before Footwork, Nikhil was an early backer of Canva (led the seed in 2014), Lattice, and Frame.io. The Footwork portfolio today reads as a deliberate refusal of the consumer/enterprise binary — Elicit (AI for scientific research, co-led with Spot), Fuse (AI-native loan origination for credit unions, banks, auto lenders), Confido (the source of truth for ops, sales and finance at $100M+ CPG brands), plus stealth investments in blood testing and AI-native services in pharma and financial services.

The frame he keeps coming back to: roughly three-quarters of Footwork’s portfolio self-identifies as both consumer and enterprise. That isn’t a market-sizing trick. It’s a bet that the very biggest companies of this decade will be the ones with the audacity to conquer multiple markets — and that picking a side too early is quietly the most expensive mistake a founder can make.

And the line every founder should write on a wall: Nikhil is looking for teams with “years of having studied the buyer.” In the unsexy verticals where the customer is skeptical and the AI thesis isn’t pre-sold, that domain depth — not pedigree — is the unlock. Several of his best founders are first-time entrepreneurs in their 20s who simply know the back office cold

Watch Episode Here

Listen To Episode Here

Ten coordinates on the survival map.

Operator-grade lessons for the things you’re actually deciding this week — which part of the map is safe to build on, which moats actually compound, which “stupid” vertical to attack, and which categories to walk away from before the model companies eat them — written for founders shipping on Monday morning, not the memo author a year later.

#01: Step one on the map: refuse the consumer/enterprise box

For twenty years the industry trained founders and investors to pick a side. Consumer team or enterprise team. PLG funnel or top-down sales motion. Nikhil’s contrarian read: the biggest companies in the world serve both, and almost always became both over time. Canva did $4B+ of ARR organically spreading from consumers to prosumers to the enterprise — $600M+ of that ARR now sits in the enterprise. SpaceX is a consumer subscription business (Starlink) renting B2B capacity to Colossus. OpenAI, Anthropic, Mercor, Open Evidence — every defining company of this cycle quietly carries both DNAs.

On the survival map, the binary itself is a trap. Picking one side caps your ceiling and quietly hands the other side to a competitor who didn’t. Three-quarters of Footwork’s portfolio refuses the box on purpose — not as a market-size argument but as a product-quality one. The bar set by ChatGPT and Claude at home is now the bar inside the enterprise. Build for one audience only and you’ll be out-magicked by someone who built for both.

Action item: Stop fighting over the consumer vs. enterprise label in your seed deck. Map both wedges. Pick the one you ship first — but design product, brand and pricing as if the other is two years away, not never.

#02: Earn the end user — it’s the cheapest path through the map

Elicit started as a self-serve AI product for individual researchers. Today it has enterprise-wide licenses with multiple top-20 pharma companies — none of which were sold top-down. Open Evidence didn’t pitch hospital systems; it earned physicians one at a time and let the bottoms-up demand pull the contract through. Both quietly went around the procurement playbook every legacy vertical software company is still running.

On the survival map, end-user love is the cheapest distribution you can buy. When the software is doing the work alongside a human — not just recording it — it has to feel personal, immediate, and trustworthy in the first 90 seconds. That’s a consumer bar, not a buying-committee one. Founders who internalize that ship a fundamentally different product, and quietly route around the incumbent sales motion entirely.

Action item: Make the first-time experience self-serve and unforgettable, even for an enterprise-bound product. Track activated end-users per company before you track logos.

#03: Plan for the B2B2C act — it’s a free moat on the map

Fuse is building an AI-native loan origination system for credit unions, banks, auto lenders. It is not, on the tin, a consumer product. But the byproduct of doing the job well is a dramatically better experience for the consumer applying for the loan — and that consumer is otherwise being courted by a neobank with a much sharper UI. Toast is the cautionary tale: a vertical software darling that never built a real consumer brand and is now vulnerable to DoorDash owning the moment of intent.

Almost every vertical SaaS company at scale ends up B2B2C — through embedded payments, financing, marketplaces, or just by being the touchpoint where the end customer experiences your buyer’s business. The founders who plan for that arc early get an extra moat for free. The ones who ignore it leave it for someone else to take.

Action item: Audit where the end consumer touches your product even when you sell B2B. Build for that surface as if it’s a standalone consumer brand — because eventually it will be one.

#04. Brand is the most underrated AI moat — start compounding it now

Nikhil’s most contrarian take: when the post-mortem of this AI era is written, brand will be the moat that quietly held everything together. People don’t switch off the products they trust with their data and their workflow, even when a technically superior competitor launches. iPhone is the case study — the flip phones and folding phones are objectively cool, and almost no one moves. The second Apple ships something, the entire market shifts back. That’s brand doing the work.

In vertical AI this is enormous, because the incumbents you’re attacking — your LabCorps, your 40-year-old EHRs, your legacy ERPs — have terrible NPS but have brand inertia. The opening is real, but only if you build a brand operators love. Capital can help; what compounds faster is a small set of early customers who are extraordinarily happy and tell their friends. That word-of-mouth in a small, tight-knit vertical is the hardest thing for a better-funded latecomer to replicate.

Action item: Treat early NPS, customer love and word-of-mouth in your vertical as a balance-sheet asset. Measure them. Resource them. Don’t outsource them.

#05. Stay out of the LLM kill zone — pick the workflows chat can’t do

Nobody can prognosticate where the model companies are in five years, but you can absolutely choose not to stand in their path. Footwork’s underwriting filter: is this product fundamentally different from what Anthropic or OpenAI will ship horizontally? Is there a real reason for stickiness — proprietary data, complicated integrations, workflows that simply do not fit a chat interface? If not, it’s a thin bet no matter how strong the team.

It’s even worse if your core value prop overlaps with what the LLM already does well. Pure document review, generic Q&A, anything that lives comfortably in a Claude tab — the labs will eat. The work for founders is to find the parts of the job where a chat box is structurally the wrong UI, where the data is hard to get, and where the integration cost is the moat. That’s where defensibility quietly lives.

Action item: Write a one-pager titled ‘Why a Claude tab can’t do this in 12 months.’ If you can’t fill the page, change the product before you raise.

#06. Hunt the legacy verticals with terrible NPS and no software heritage

Ask a room of founders to name a venture-backed software company at scale selling into the CPG industry. Long pause. CPG is a top-six GDP category — $3T+ of spend — with 10,000 brands in the US doing more than $100M of revenue each, almost none of them well-served by software. Footwork’s portfolio company Confido is building the source of truth across ops, sales and finance for those brands, integrating with every retailer, doing trade promotion management and forecasting that has been manual until literally now.

These categories are ‘unloved’ for a reason: they’re harder to sell into, the buyer is more skeptical, the AI tailwind isn’t obvious to the customer yet. But they are also antifragile — people will keep eating, drinking, getting blood drawn, applying for credit-union loans, taking GLP-1s — and they are wildly underpenetrated by software. The boring vertical with low NPS incumbents and a real labor problem is, in 2026, often a far better venture bet than the obvious one.

Action item: Make a list of the ‘no one would build a SaaS company there’ verticals you actually understand. Rank them by NPS of the incumbent and willingness to pay against labor. Go where the score is worst.

A quick word from our incredible partner,
Parafin - the leading embedded capital partner.

• 4.8 star rating on Trustpilot, with over 500 reviews [source]

• 71% CSAT rating across all partners

• 84 NPS

• Serving top Fortune 500 companies, including Amazon, Walmart, and DoorDash

Check Out Parafin

#07. Network effects are still rare, still real, and worth raising your hand for

Nikhil’s 16-year dream company is the one that finally unseats LinkedIn — because LinkedIn is a brutal case study in just how durable a real network effect is. The Mercor / Handshake / data-labeling space is the closest thing this cycle has to that opening: experts and talent on one side, businesses on the other. Whoever actually builds the next professional network — the one where people show their skills, not just their resumes — is the real long-term threat in that category.

The broader point: network effects in this AI era are quietly some of the most powerful moats still available. ACV Auctions in car auctions, a category nobody would have predicted to be network-effect driven, is the archetype. If you have a real one — or sit on a latent network like Handshake did — it is one of the very few assets that compounds against falling model prices and accelerating competition.

Action item: Map the two-sided dynamics inside your vertical. If you can credibly seed both sides with the same product, you have a moat the model companies cannot replicate by shipping a better LLM.

#08. Earn the right to do more than one thing — but only when the team can

Focus still works. Anthropic’s relentless focus on coding agents is what’s powering its run inside the enterprise; OpenAI’s split attention across consumer and enterprise arguably cost them the developer mindshare they’re now chasing back. So the playbook isn’t ‘do everything from day one.’ It’s that at some point — sooner than the SaaS playbook would suggest — the very best companies are forced to do multiple things really, really well, and the founders who built a team capable of that compound faster than the ones who didn’t.

Nikhil isn’t the board member walking in saying ‘focus less.’ He’s the one watching whether the founder is openly thinking about the second and third act, hiring leaders who can shoulder them when the time comes, and staying flexible in a market that re-prices itself every quarter. That mindset is the real differentiator between a good company and a generational one.

Action item: Keep this quarter’s focus brutal. But sketch the second act on a single page and revisit it monthly. Hire one leader this year who could credibly run it.

#09. Back domain depth — especially in the unsexy verticals

The pattern across Confido (CPG), Fuse (credit unions), Elicit (pharma research): the founders are not Silicon Valley generalists parachuting in. Several are first-time entrepreneurs in their 20s. What they share is years of having lived in or studied their buyer. In categories where the customer is more skeptical and the AI thesis isn’t pre-sold, that domain depth is the unlock. You cannot fake your way into the trust of a CPG ops leader or a credit union CIO.

The corollary for operators: if you’re attacking a hairy vertical with no software heritage, your sharpest weapon is the depth of your relationships and the specificity of your insight into how the work actually gets done. The flashy go-to-market deck loses to the founder who can sit in the back office for a week and come back with a roadmap.

Action item: If you don’t have years of domain depth, hire it into the founding team — and spend a month embedded with your buyer before you ship v2.

The map, drawn live — which obvious bets are kill zones, and which are quietly the safe ones.

We closed the episode by putting Nikhil — and Nic — in the seat and pricing the map in real time: which of the most highly valued vertical AI bets are sitting inside the kill zone, and which still have a clean moat, a real network, or an unloved vertical underneath them. Below: where on the map each lands, and the operator-grade reasoning underneath.

The pattern is uncomfortable but consistent. The hype-cycle category of any given year almost never produces the biggest outcome of that year. The biggest outcome comes from the category everyone agreed was dead, niche, or already won. If your whole thesis is in the consensus column, you’re not actually playing for power-law returns — you’re playing for second place.

The operator takeaway from the rapid-fire round: don’t build inside the kill zone, no matter how loud the category is. Spend your time where the moat is brand, network, or workflow depth the LLM cannot replicate — that’s the only stretch of the map that actually compounds.

The most uncomfortable line of the conversation: selling AI into white-collar industries where the user already has Claude open is structurally harder than selling AI into a back office that has nothing today. The unloved legacy vertical with a 40-year-old EHR and no business network is, in 2026, often the better square on the survival map than the keynote-friendly one.

Survive the map first.
Then build something the labs can’t follow.

If you’re a founder or operator shipping vertical AI right now, four moves separate the companies that survive the map from the ones that don’t:

1. Map the kill zone before you map the market. If a Claude tab can credibly do your core value prop inside 12 months, you don’t have a company — you have a feature waiting to be eaten. Pick workflows that need proprietary data, hairy integrations, or a UI that isn’t chat.

2. Compound a real moat from day one. In this era only three actually hold: brand (the iPhone effect), real network effects (the next LinkedIn, the next ACV), and workflow trust earned inside a traditional industry. If you can’t name which one you’re building, execution speed is your only moat — and that’s a brutal place to live.

3. Build in the unloved, antifragile verticals. CPG, credit unions, blood testing, pharma research, accounting back-offices. Terrible incumbents, no real software heritage, demand that isn’t going anywhere. The unsexy squares on the map are quietly the biggest ones — and the consensus auction never shows up.

4. Refuse the binaries that cap your ceiling. Consumer or enterprise. Focused or expansive. Software or services. The defining companies of this cycle — Canva, SpaceX, Anthropic, Open Evidence — refused all three. Plan for the second and third acts early, even if you only ship the first one this year.

The operators who win this cycle are the ones who treat vertical AI like a map, not a thesis. Most of the board is hostile. A small set of squares — unloved verticals, real moats, workflows the labs can’t reach — is where the next generational companies are quietly being built. Find those squares, plant your flag, and compound for a decade.

Watch Episode Here

Listen To Episode Here

Thanks for reading LINEAR. I reply to every email…
Have any questions, feedback, comments? Let us know, we work for you!

MaintainX Acquired By AutoDesk for $3.6B (!!) Let’s dig into the company…

Most people will look at MaintainX and say: “It’s just maintenance software.”

That is technically correct. And strategically useless. Because the real story is not “maintenance software.”

The real story is:

MaintainX built a mobile-first system of record and action for the people who keep the physical world running — then layered data, AI, and enterprise rollout on top of it.

That is a much more interesting company.

And it helps explain why Autodesk agreed to acquire it for approximately $3.6B in an all-cash deal. Autodesk said MaintainX is expected to exceed $135M of ARR in calendar 2026 while growing more than 50%. They paid 26x EOY ARR. Holy sh*t…

The founding story

MaintainX started in 2018 with a very simple observation:

The software given to frontline and industrial workers was stuck in the past.

MaintainX’s own framing is that the tools keeping the world moving were outdated, and that maintenance teams were still relying on antiquated systems for critical workflows. Chris Turlica and his cofounders had already built and sold a company together, and Turlica later said the team wanted to build something as easy to understand as Slack or Asana, but designed specifically for frontline maintenance, safety, and operations.

BCV tells the founding story even better.

Their version is that Turlica’s earlier consumer chat app was being used by factory workers to photograph clipboards and discuss work orders. That is one of those great startup moments where the market tells you what the actual product should be. Not a generic workflow app. Not another dashboard. A purpose-built operating layer for physical operations.

That wedge mattered.

Because if you start with the technician, the supervisor, the plant manager, and the actual job to be done, you don’t build “software for maintenance.” You build software that becomes embedded in the daily motion of keeping equipment online.

The early journey and fundraising history

The company moved quickly.

In March 2019, MaintainX announced a $3.8M seed round led by August Capital and Amity Ventures. At that point the pitch was straightforward: digitize procedures, inspections, audits, checklists, and work orders for deskless teams who were still operating on paper, radios, spreadsheets, and email. The company said more than 1,600 businesses had joined the platform and were ready to start using it daily.

In June 2021, MaintainX announced a $39M Series B, and disclosed a previously unannounced $11M Series A from 2019. That brought total funding at the time to $54M. The company said revenue had grown 12x from early 2020, retention was 98%, and customers across 20+ countries were using the platform to manage hundreds of thousands of physical assets.

By December 2023, MaintainX had raised a $50M Series C at a $1B valuation. Turlica wrote that the company had grown to 6,500+ customers, processed 15M work orders, and managed 2.5M assets.

Then in July 2025, MaintainX raised a $150M Series D at a $2.5B valuation, bringing total funding to $254M. The company said it served 11,000+ companies, managed 11M+ assets, and was using the new capital to expand AI, machine health monitoring, predictive maintenance, and broader enterprise asset management capabilities.

Then came the Autodesk deal.

From seed in 2019 to a $3.6B exit agreement in 2026.

That is a serious run.

What the product actually became

This is the part I think people miss. MaintainX did not stay a simple CMMS.

It kept widening the surface area.

Today the core stack includes:

• work order creation, assignment, and tracking

• preventive maintenance scheduling

• condition-based work triggers

• asset health and reliability insights

• parts and inventory management

• inspections, safety procedures, and compliance workflows

• analytics and reporting

• AI copilots and natural-language querying for maintenance teams

That is how vertical software gets more valuable.

You start with the obvious pain and you solve it with a STUPIDLY simple capability.
Then you absorb adjacent workflow.
Then you absorb data.
Then you absorb decisioning.

At some point, you stop being a tool.

You become the operating layer.

MaintainX’s own homepage now frames the company around work intelligence, asset intelligence, AI-powered reporting, predictive parts needs, and compliance digitization. That is no longer “just maintenance.” That is workflow ownership around uptime.

The GTM motion was sneakily powerful

This may be my favorite part of the story.

BCV describes MaintainX as a rare industrial software company that behaved almost like a PLG wedge inside a traditionally top-down market.

Frontline workers liked the app.
Factories adopted it because it was easy to use.
Regional managers noticed that certain sites were outperforming.
COOs saw the pattern and standardized around it.

That is an elite motion.

You get bottoms-up love and top-down expansion.

You get usability as the wedge.
You get operational data as the moat.
You get enterprise standardization as the scale unlock.

Autodesk explicitly called out MaintainX’s pre-built integrations and scalable go-to-market growth motion in operations as part of the rationale for the acquisition.

That line is doing a lot of work.

Because it tells you Autodesk was not just buying a product.

It was buying a repeatable distribution engine into operations.

The post-sales motion mattered too

A lot of software companies win the deal and then lose the deployment.

MaintainX seems to have taken implementation seriously.

Its implementation page says the company has completed 3,000+ implementations with a stated 98% adoption rate, and offers tailored onboarding led by specialists with engineering and operational backgrounds. The messaging is very clear: get sites live in weeks, not years; standardize, optimize, and build a center of excellence.

For enterprise customers, MaintainX highlights a structured three-week onboarding process, a dedicated implementation specialist, support for multi-site rollouts, ERP/IoT/BI integrations, SSO, MFA, custom permissions, unlimited custom dashboards, and dedicated account management.

And the case studies back it up.

At US LBM, the rollout happened in waves of 3-5 sites at a time, with early focus on the 20% of the tool that would drive 80% of the impact. The company standardized asset naming and reporting, but still allowed sites enough flexibility to adapt local procedures. That is smart implementation. Not rigid. Not chaotic. Structured enough to scale, loose enough to spread best practices.

The customer success motion also looks mature.

MaintainX’s enterprise CSM role includes onboarding, product demos, training sessions, webinars, quarterly business reviews, adoption monitoring, churn mitigation at lagging facilities, and upsell/cross-sell identification with the sales team. That tells you a lot about how the company expanded inside large accounts.

This is what good vertical software looks like.

Not just software people love.

Software that gets implemented, adopted, expanded, and embedded.

Why Autodesk paid up

This is where the story gets really interesting.

Autodesk is trying to connect design, make, and operate.

MaintainX gives them a much stronger position in the last part.

Autodesk said the acquisition strengthens its ability to connect operations workflows with the broader lifecycle, brings richer data on asset history and maintenance patterns, and helps extend Autodesk’s relationship with assets from years to decades.

The investor presentation sharpens that further.

Autodesk frames the deal as a way to add roughly $40B of Operations TAM to its existing $78B Design & Make TAM, while extending its duration with assets and systems from a few years to 50–60+ years.

The idea is simple: design software knows how an asset was supposed to work; MaintainX knows how it actually performs in the field. Put those together, and you get a much more valuable loop.

BCV put it well too.

For decades, design tools captured how an asset should work, while operations platforms captured how it actually worked, and those datasets rarely talked to each other. Autodesk + MaintainX closes that loop.

That is what Autodesk is buying:

Not just CMMS revenue.

A frontline data asset.
A workflow foothold in operations.
A long-duration relationship with the physical world.
And eventually, a better foundation for system-level AI.

Big kudos to the team

This is a huge win.

Chris Turlica deserves a lot of credit for seeing a very unsexy problem clearly and then building with unusual intensity around it.

Same goes for Hugo Dozois-Caouette and the broader team.

They picked a market where the pain was real, the buyer need was persistent, the workflow was mission-critical, and the software incumbents were ancient.

Then they did the hard part:

They actually executed.

They built something workers used.
They made the data useful.
They made implementation real.
They expanded into enterprise.
And they gave a strategic buyer a very clean reason to pay a premium.

That’s the playbook.

Find a workflow the world cannot live without.
Build the system of action around it.
Turn usage into data.
Turn data into intelligence.
Turn intelligence into strategic value.

MaintainX did that.

And a $3.6B outcome is what that can look like when you do it really, really well.

The Hidden Ceiling in Embedded Payments

A lot of vertical SaaS founders think embedded payments is about choosing the “right” model. Referral, hybrid, or PayFac. But I’m starting to think that framing misses the bigger issue entirely.

Because the real mistake usually isn’t choosing the wrong model. It’s locking into one too early.

Early-stage SaaS companies naturally optimize for speed. They want the fastest path to launch, the simplest onboarding experience, and the least operational friction possible. Totally reasonable. Nobody wants payments turning into a six-month operational project while they’re still trying to grow the core platform.

The problem is that what works early often becomes limiting later, because SaaS businesses don’t stay static for very long.

As companies scale:

• Customer segments become more diverse

• Enterprise deals introduce new requirements

• Sales teams need more flexibility

• Payments revenue becomes materially important to company economics

That’s where rigid payments strategies start creating problems. Not obvious failures, but hidden ceilings.

At first, everything still looks healthy on the surface. Merchants onboard. Transactions process. Revenue grows. But underneath, the same payments structure is getting applied across completely different customer types. SMBs get treated like enterprise accounts. Enterprise accounts get forced into SMB workflows. Sales teams lose flexibility, and adoption slows because the process no longer fits every segment equally well.

Most founders don’t notice this immediately because nothing technically “breaks.” Growth just becomes harder than it should be.

That’s why the strongest SaaS platforms don’t treat payments like a fixed infrastructure decision. They treat it like an evolving revenue strategy that matures alongside the business itself.

The priorities change over time:

• Early on, the goal is speed and adoption

• Later, the focus shifts toward improving revenue per customer

• Eventually, the focus becomes margin optimization and enterprise monetization

Those are completely different operational priorities, and they rarely fit neatly into one permanent model.

The best operators understand this intuitively. What works at 100 customers usually breaks at 1,000. What works for SMBs often doesn’t work for enterprise. What feels operationally simple early on can quietly suppress long-term revenue later.

That’s why more sophisticated SaaS platforms are starting to structure payments differently based on the customer segment itself. SMBs might prioritize simplicity and fast onboarding. Mid-market customers may require a more balanced approach. Enterprise accounts can often support higher-control, higher-margin structures.

Not because complexity is the goal, but because maximizing long-term revenue usually requires flexibility.

Honestly, I think this is where the embedded payments conversation is heading. The future probably isn’t about choosing between referral, hybrid, or PayFac. It’s about building the ability to move between them as the business evolves.

That’s part of why I’ve been paying attention to what Xplor Pay is doing lately. They seem to think about embedded payments more as an evolving growth system than a static infrastructure decision, which honestly feels much closer to how SaaS businesses actually operate.

Because at a certain stage, payments stops being just infrastructure. It becomes part of the growth strategy itself.

Your strongest AI Users are probably worth 10x your other employees

One strong AI-native Employee can now do the work that used to require a researcher, a designer, a coordinator, a writer, and a developer.

That changes everything, and we’re really not talking about it enough.

The right question is:

Who in our org has rebuilt their entire workflow around AI?

That’s who matters.

Because once someone figures this out, they stop behaving like a normal employee inside a normal function.

They become a throughput machine.

They learn faster.
They iterate faster.
They ship faster.
They document better.
They unblock other people.
They make management look smarter.
They make average teams feel more coordinated than they really are.

And most importantly:

They create leverage that compounds.

A strong AI user does not just produce more output. They create prompts, templates, automations, operating procedures, and internal playbooks that other people can reuse.

So their value is not just their own output.

It’s the new system of work they drag into the company. That is why I think these people are worth way more than most org charts suggest.

If you have an IC who is implementing AI into every part of their workflow, you should think about them differently.

Not as a normal employee. As infrastructure. As a force multiplier.

I think a lot of companies are massively underestimating this.

Not 10% better.

Not “a bit more productive.”

10x more valuable.

What I would do

1. Identify your top 5 AI-native operators immediately.

Not the people talking about AI the most.

The people quietly using it to get absurd amounts done.

You probably already know who they are.

They are the ones sending better memos faster.
The ones with cleaner thinking.
The ones who come to meetings with the first draft already done.
The ones who somehow keep eliminating steps.

2. Turn them into internal teachers.

Ask them to document exactly how they work.

What tools they use.
What prompts they rely on.
What gets automated.
What still requires human judgment.
What a “good” AI workflow looks like inside your company.

If you do this well, one great AI operator can upgrade ten others.

3. Give them harder problems, not more busywork.

Do not turn your best AI users into cleanup crews.

Use them on strategy, customer insight, GTM experiments, internal systems, and cross-functional bottlenecks.

Point them at the highest-friction parts of the business.

Let them redesign the work.

4. Promote for leverage creation.

In a normal org, we reward ownership.

In an AI-native org, we should reward scalable ownership.

Who is building repeatable systems?
Who is making other people faster?
Who is changing the slope of execution?

That is the game now.

5. Weed out the Non AI People

Can you imagine if your entire organization was made up of these people? I’m seeing it first hand. I’m truly living it. I’m involved in companies with a few AI native IC’s (across every type of role BTW) and companies where Claude is the legit operating system of EVERY employee.

There is a serious difference between the two. And I don’t want to be in organizations that are the former. So If your a little hardcore like me, you may want to start thinking about weeding out the people that are still criticizing the use of AI. If they haven’t got it now, will they ever? Do you want to pay for them to catch up? If they ever do?

In an AI-native org, we should reward scalable ownership.

That is the game now.

The takeaway

The companies that win over the next few years will not just “adopt AI.”

They will identify the people who are best at using it, then reorganize around them.

That’s the opportunity.

And that’s also the warning.

If you have a few of these people in your company, treat them like gold.

If you don’t, go find them.

Fast.

If you made it thus far! You’ve found the easter egg.
Snag a free ticket to the Vertical Software Summit. Just use code FREE4FOUNDERS!

Grab Your Free Ticket

Vertical Software/AI Founders, Operators, Investors only please.
If you’re not your submission won’t be accepted.

Do me a solid and forward to a friend :-)

Nic and I sat down with Kyle Liu, General Partner at Bling Capital — built around helping founders find, define and scale PMF — on the two shifts every operator is feeling but few are naming: enterprise buyers are behaving like consumers (they’ll try anything, they’ll churn on a dime, they want one bill not ten), and the old PMF signals — $1M ARR, logo lists, a clean Series A deck — quietly stopped meaning what they used to mean.

The line between consumer and
enterprise is collapsing.

For twenty years there was a clean story. Consumer products were tried in seconds, loved or churned. Enterprise products were sold over quarters, integrated for months, and stuck around for years. That story is dead. In AI, your enterprise buyer behaves like a consumer — they’ll spin you up on a Tuesday and rip you out the following Tuesday — and the old PMF signals never accounted for that.

Kyle’s seen this from the inside. Bling’s product council — 100+ heads of product and growth at top tech and AI companies — used to be a hiring and advisory bench. Now they’re acting like consumer users for portfolio companies: trying everything, switching constantly, telling you within a week if your product earns a permanent seat in their stack. Buying committees flattened. Procurement got bypassed. The “champion + economic buyer + IT” map you learned at the last enterprise SaaS company doesn’t necessarily describe the buyer room anymore.

That collapse cuts both ways. The good news for operators: a single product-led demo can land a six-figure pilot in days. The bad news: a Series A AI company can go from a $30M customer to $0 the next year. Kyle’s seen it. The cost to build collapsed, the speed to market collapsed with it, and the feedback loops shortened — so 20 funded competitors now sit where three used to. $1M in ARR no longer proves anything. Defensibility, the thing every founder used to worry about later, is on the agenda at the seed pitch.

So what does PMF actually look like in the AI age — for founders shipping right now, not investors writing memos a year later?

That’s the lens Kyle brought in.

Earlier signal. Cohort discipline at week ten. A “press release at Series A” you can defend line by line. Gross profit, not ARR, as the honest scorecard.

And a gut-check on whether the enterprise customer who just signed will treat you like an indispensable system — or like a Claude tab they’ll close next quarter.

Watch on Youtube

Listen To Episode Here

This Weeks Vertical Titan: Kyle Lui (Partner @ Bling Capital)

Kyle was an operator first. He founded Choice Passenger, a sales performance management company that was acquired by Salesforce and grew inside it into the Salesforce HR Cloud — roughly $40M in ARR across 1,000+ enterprise, mid-market and SMB customers. From there, eight years investing at DCM, and in 2022 he joined his first-ever angel — Ben Ling — as GP at Bling Capital.

Bling is a product-led seed fund with a 100+ strong product council of heads of product and growth at top tech and AI companies — increasingly doubling as early design partners for portfolio startups. Kyle’s portfolio includes Spellbook (vertical AI for legal), Alma (AI-native immigration law firm), Miniva (vertical AI for high-volume manufacturing), and PartBay (used auto parts — the “is this market even big enough?” company).

The frame he keeps coming back to: cohort data over storytelling. Even ten weeks of data is signal worth pulling — not to declare PMF, but to find the bright spot inside a semi-horizontal product and aim the whole company at it.

And the line every founder should write on a wall: he’s looking for “learn-it-alls, not know-it-alls.” Meet a founder on a Monday, meet them again the next Monday — if they’ve already talked to half the people you introduced them to and re-shaped the plan around what they heard, you’re probably looking at a generational team in formation.

Watch on YouTube

Listen To Episode Here

What PMF actually looks like in the AI age.

Nine lessons every operator should re-internalize before they trust their next ARR chart — including the consumer-style behavior now showing up in every enterprise deal.

#01. Size the company, not the market — in gross profit, not TAM

Bling’s market-sizing exercise doesn’t ask for a TAM slide. It asks how the company gets to $100M and $500M in gross profit: how many customers, what percentage of the market, over what timeline. The numbers aren’t arbitrary — $100M GP is roughly what a seed fund needs from an 8–10% stake to return the fund, and $500M is the multi-stage equivalent. It’s an ICP and willingness-to-pay question dressed as market sizing.

In the AI era, Kyle now adds a third number: $1B. Multiples have come down, growth rates have gone up, competition has gone up, and the power law got even more extreme. A billion in gross profit is the bar a multi-stage fund needs to see to be excited. If your model can’t credibly find that ceiling, your cap table tops out before you’d like it to.

Action item: Run the $100M, $500M and $1B GP exercise for your business. If you can’t credibly hit the second number, you’re not seed-VC-fundable. If you can’t see the third, you’re not multi-stage-fundable.

#02. PMF as the Amazon press release you can’t write yet

At seed, true PMF is unknowable. What’s knowable is the early signal of it. Bling’s exercise: imagine you just raised your Series A 12–18 months from today and write the Amazon press release that announces it. What claims are in there? Customer love, engagement acceleration, early distribution proof points. Then break each claim into OKRs and work backwards.

Every generational company starts as a wedge — Uber was ‘five times the cost of a taxi and people can’t get enough.’ Stripe was a clean API. The vision can be enormous, but the wedge has to be brutally specific, and the press release forces the conversation about which wedge actually proves itself first.

Action item: Draft the press release. Break each claim into measurable proof points. Treat the doc as your PMF roadmap, not a marketing exercise.

#03. Cohorts at week ten — find the bright spot, don’t prove PMF

Whenever there’s data — even ten weeks of it — Bling pulls cohorts. Not to certify PMF, which would be premature, but to find the one or two slices where engagement or retention is meaningfully ahead. That’s especially powerful for companies with semi-horizontal products debating between adjacent verticals.

The job is to cut through the noise of a market with 20 funded competitors and find the customer profile where urgency is real. Once you see it, the whole company should bend toward it: positioning, hiring plan, design partners, the next product surface. Sound discipline beats storytelling — and at seed, that discipline is most of the moat you have.

Action item: Pull cohort retention, expansion and engagement on every customer slice. Pick the bright spot. Re-orient the company around it.

#04. Your enterprise buyer is a consumer now — and ARR is hiding it

In vertical SaaS, hitting $1M ARR meant something — it usually came from roughly twenty discerning customers with 80–95% likelihood of retention. In vertical AI, Kyle has seen a Series A company go from $30M in one customer to $0 the next year. The reason: enterprise buyers now behave like consumers. They trial fast, they champion fast, and they churn fast when the next demo is shinier — exactly the way an individual flips between Claude, ChatGPT and whatever they read about on Twitter that morning.

That’s why ARR has stopped being the honest scorecard. For operators, the numbers that actually predict survival are gross profit, gross-margin trajectory, customer concentration, and second-cohort retention. If a single customer is more than 20% of revenue, that’s not a logo — it’s a runway risk.

Action item: Don’t celebrate ARR until you’ve stress-tested concentration and second-cohort retention. Track GP, not bookings.

#05. Moats fall first — then rebuild around workflow and trust

Kyle is blunt: moats are falling. In the early days founders shouldn’t worry about defensibility because they have nothing to defend — what matters is execution speed and getting people to pay attention. But over time, the durable moats start to look familiar: data, workflow lock-in, and trust earned inside traditional industries that don’t actually want to switch once they’re happy.

Harvey, Legora, Spellbook didn’t out-model Claude — they out-executed everyone else and embedded into workflows the foundational labs aren’t going to touch. Developers are actually the hardest customer to retain, because they care about marginal performance. Traditional industries, once they trust you, are the opposite — they don’t want to re-evaluate every six months.

Action item: Year one, optimize for speed of learning. Year two, consciously pay down workflow depth and switching cost — not just ship features.

#6. Be the router, not the reseller

Users will pay a ~20% tax for a product that picks the right model for each task. Cursor is the canonical horizontal example; every vertical AI company has the same opening. The router is a clean wedge — high frequency, immediately valuable, and a beachhead for the workflow you actually want to own.

But if your core value prop is the same thing the LLM does well — pure document review, for instance — the labs can eat you. Routing is a wedge; it can’t be the whole company. The constellation of features around it is what makes ‘just use Claude’ a non-option twelve months in.

Action item: Start as a great router. Use that to capture workflow, data, and trust. Then layer the features that make a single-LLM fallback genuinely worse.

A quick word from our incredible partner,
Parafin - the leading embedded capital partner.

• 4.8 star rating on Trustpilot, with over 500 reviews [source]

• 71% CSAT rating across all partners

• 84 NPS

• Serving top Fortune 500 companies, including Amazon, Walmart, and DoorDash

Check Out Parafin

#07. Sell against the labor budget — the biggest unlock of the cycle

Kyle has watched board conversations rewire in real time. R&D line items now combine token spend and headcount in the same row — and traditional industries are even further along. They feel labor replacement viscerally. Alma, an AI-native immigration law firm in Kyle’s portfolio, compressed 40 attorney-hours of visa work into roughly 4. Miniva, in high-volume manufacturing, is targeting roles factories literally cannot find people for.

That changes what counts as a venture-scale market. Vertical AI for private aviation brokers sounds laughably niche — until you realize the early contracts are seven figures because the budget is coming out of payroll, not the software stack. The categories that didn’t pencil under SaaS economics often pencil beautifully when the comparison is loaded labor cost.

Action item: Price against fully-loaded labor cost. If buyers are still benchmarking you against SaaS seats, you haven’t told the right story yet.

#08. Gross margin still matters — eventually

Negative gross margin from compute is forgivable; everyone assumes inference costs trend down. Negative gross margin from a giant services bench is not — that’s just BPO with a venture multiple on top. Compute costs come down. People costs don’t.

You can’t invest a dollar to make 80 cents forever. There has to be a clear inflection where the business model becomes viable on its own — and a credible path from today’s gross margin to a steady-state target you can actually defend in a board meeting.

Action item: Map the path from today’s GM to a steady-state target. If you can’t draw that line, you don’t have a business yet — you have a subsidy.

#09. Hunt narrative violations — the obvious bets are already crowded

Big funds coming downstream has created consensus vortices around a few hot categories. Meanwhile the biggest outcomes of this decade — SpaceX, Anthropic, OpenAI — were all unloved at the moments that mattered most. The ‘stupid’ vertical that 10 VCs pass on may actually buy you the longest stretch of low competition and a clean shot at owning the names that matter in your market.

If everyone you respect is already in your category, your moat is now execution speed and very little else — a brutal place to live. If they all think it’s stupid, do the work twice. That’s where the real outcomes hide.

Action item: Pressure-test your category against the consensus list. If you’re on it, ask what would have to be true for you to win on execution alone.

Where vertical AI is already a consensus trade — and where the real outcomes still hide.

Big funds coming downstream has created consensus vortices around a handful of categories. Kyle’s read on where the auction is hot — and where the next category-defining companies are most likely to be quietly built.

Kyle’s most provocative line:

The pattern is uncomfortable but consistent. The hype-cycle category of any given year almost never produces the biggest outcome of that year. The biggest outcome comes from the category everyone agreed was dead, niche, or already won. If your whole thesis is in the consensus column, you’re not actually playing for power-law returns — you’re playing for second place.

Watch on YouTube

Listen To Episode Here

Thanks for reading LINEAR. I reply to every email…
Have any questions, feedback, comments? Let us know, we work for you!

Shaped by lessons from decades of building and scaling 20+ SaaS platforms, it turns payments into a true growth lever. Read the guide to see how top platforms get it right. Read Now.

An Exciting Announcement…

Today I’m excited to announce the second annual Vertical Software Summit.

We spent the last six months pulling together the BEST in vertical software and vertical AI — founders and executives who have actually built, scaled, and in some cases monopolized their corners of software…

• Fred Voccola: The only guy (that I know of) who runs TWO billion dollar vertical software businesses, Kaseya (IT services companies) and Simpro (Trades)

• Founder/CEO of OpenGov: A 15 year journey building the OS for local government and he just sold for $1.8B to Cox

• The Founder/CEO of GCAI: One of the fastest growing Vertical AI companies out there. I’ve never seen growth quite like them…

• The Founder/CEO of GlossGenius - probably worth $1B+. Massive business in the beauty software space.

• The Founder/CEO of Valsoft - He’s bought 100+ vertical SaaS co’s and is approaching $1B of ARR. Monster business.

• The Chief Business Officer of Service Titan. $1B+ of ARR. Self explanatory :-)

• The Founder of a $5B+ vertical marketplace.

• 30+ other Founder/CEO speakers from $1M-$1B in ARR

• Leading CxO’s on AI, FinTech, Customer Success, GTM, Product, and more

The founders who have actually dominated categories don’t speak in abstractions. They talk about the ugly parts:

• The pricing decisions that looked wrong until they worked.

• The distribution channels everyone ignored.

• The product lines they killed.

• The implementation bottlenecks they underestimated.

• The one integration that mattered more than ten features.

• The part of the workflow they thought was peripheral that ended up becoming the real wedge.

That’s where the value is.

Not in hearing polished startup advice you could get anywhere else. In hearing how category leaders actually built businesses that became hard to dislodge…

The conference is built entirely around the folks actually shaping their categories: the founders building vertical businesses and the operators scaling them.

And when you get enough of those people in one place, the conversation is something special…

If you’re a founder, you’ll leave with a better sense for what actually separates a good vertical business from a category-defining one.

If you’re an operator, you’ll hear how leading CxOs think about expansion, implementation, org design, AI adoption, and maintaining advantage as a company scales.

And then there’s the other part that makes this room unusually valuable: it’s not just founders and operators showing up…

The entire vertical investment and M&A community shows up wanting to meet YOU.

We will have ~100 ultra high quality investors and acquirers in the room that know and believe in Vertical Software and Vertical AI.

That’s why this is more than a conference.

It’s a place where the people building the category, scaling the category, and funding the category all collide in the same place.

And that’s when patterns emerge.
That’s when the quality of conversation changes.
That’s when you stop getting generic content and start getting real context.

If you’re serious about vertical software or vertical AI, that’s the kind of room that can change how you build.

So, today’s newsletter is short and sweet, but I encourage you to spend your usual time reading it to check out the Vertical AI event of the year!!

Hope to see in Miami November 4-5th, 2026…

PS - it will sell out (last year we had to turn away ~100 folks who waited until last minute to grab a ticket. Don’t do that!)

Check out the conference

Mike Droesch of Bessemer Venture Partners sits down with Nic to discuss where Voice AI is working, where it’s overcrowded, where the blue oceans still are, and why voice may become the next great application layer in vertical software, not just "AI that answers the phone."

Voice is an interface shift.

Voice isn’t a slicker chatbot with audio. It’s a much higher-bandwidth interface than typing, clicking, or filling out forms — and when you increase the bandwidth of data flowing in, the quality of the output changes dramatically too.

Mike’s argument: voice becomes interesting when it stops being a feature and starts becoming a serious way to gather context and take action. We’re not talking about AI that answers the phone — we’re talking about a new layer in software that captures more nuance, makes better decisions, and eventually owns entire workflows that used to require humans.

There is a lot changing in AI right now. New models. New wrappers. New infrastructure. New noise. But if you forced Mike to pick one category with the potential to reshape software more than people realize, it would be this one. Not because it’s trendy. Not because the demos are cool. Because voice changes the amount and quality of information that can flow into a machine — and that quietly changes everything downstream of it.

That’s the lens Mike brought into this conversation. Not “AI that answers the phone.” Not a better IVR. A new context layer for software, and eventually a new action layer on top of it. If he’s right, voice ends up being one of the most important new wedges into vertical software we’ve seen in a long time.

Watch Episode Here

Listen To Episode Here

This Weeks Vertical Titan: Mike Droesch (Partner @ Bessemer)

Mike came up as an engineer before venture, joined Bessemer close to nine years ago, and got pulled into vertical software at exactly the moment the category started proving itself. Since then he’s built investment roadmaps across marketplaces, supply chain, vertical SaaS, and now vertical AI.

He’s had a front-row seat to the rise of companies like Abridge, Rilla, and Vapi — meaning he’s thought about vertical voice AI as much as almost anyone in the market. The difference in his framing is that he doesn’t treat voice as a narrow subcategory. He treats it as an interface shift.

One of his biggest surprises of the AI era: the fastest-adopting sectors have been regulated ones — healthcare, insurance, financial services, legal. They care deeply about workflow structure and control, and when voice gives them tight gates, identity validation, and rigorous agent monitoring, they actually move faster than the markets you’d expect to.

And the line every founder in this category should write on a wall: “The infrastructure stuff is table stakes. Distribution matters more than it ever has because it’s so easy to get to feature parity with other products.” Latency and reliability are the price of admission. Distribution, workflow ownership, trust and evals are the moat.

Watch Episode Here

Listen To Episode Here

Voice AI changes the economics of software when it actually works. Traditional SaaS made labor more productive. The best voice companies start to replace it — and that moves the budget conversation from the IT line item to payroll and broader opex.

When voice actually works, the economics of software change.

A few observations from Mike on where Voice AI stops looking like a demo and starts looking like a category:

From IT line item to payroll line item.

In the traditional SaaS world, a lot of companies sold tools that made labor more productive. In the Voice AI world, the best companies don’t just support labor — they start replacing parts of it. That changes where the budget comes from. Instead of selling into an IT line item, you’re selling against payroll and broader operating expense.

Mike gave the example of a workflow that might support a roughly $30K ACV in its SaaS form but something closer to $150K when an AI agent actually performs the job end to end. That’s a massive unlock if it proves durable across categories — and it’s why the comp set for voice companies starts to look more like BPO and labor spend than next-seat-license SaaS.

Don’t build voice unless you can close the loop.

Mike said it directly: he would not want to spend his time building a voice agent company if he didn’t think it could do an entire job end to end. There are going to be plenty of products that sound great in a demo, have pleasant conversations, and then hand off to a human the moment things get complicated. Those might become features.

The monster companies are far more likely to be built by owning a full workflow, not a moment inside it. The frame for founders is simple: if the realistic ceiling of your product is “assistive in a slice of the job,” you’re building a feature for someone else’s platform.

The limit isn’t capability anymore. It’s permission.

Mike made a subtle but important distinction: the limiting factor is increasingly not whether the model can handle the interaction. It’s whether buyers trust the model enough to let it actually act.

We’re approaching a world where agents understand more than people expect. But connecting them to core systems, letting them move money, resolve a claim, complete a booking, or change a mission-critical record requires a different level of confidence. That confidence is not built by vibes. It’s built by governance, observability, and repetition. The buyer’s CISO is on your roadmap whether you invited them or not.

Every call is product data.

Mike’s view is that the compounding advantage comes from seeing real-time evals and tests run in production. Every interaction becomes more than a service event — it becomes product data. Every conversation generates signals. Every signal can be measured against expected performance.

Over time, the companies best at instrumenting, interpreting, and improving from those signals build a very real flywheel around quality. A lot of founders still talk as if the moat is the model. Increasingly, it’s the eval system wrapped around the model.

Escalation to a human isn’t failure.

Some people still talk about handoff like it represents failure. Mike’s framing is more practical. In real-world workflows, people usually call when something has already gone wrong, or when they’re in an edge case. The job isn’t to fantasize about 100% automation on day one.

The job is to know exactly where the boundaries are, solve what the system can solve confidently, and fail over elegantly when it can’t. In many verticals, that handoff is the difference between trust being built and trust being destroyed.

Platform + usage today. Outcomes when they’re measurable.

Outcome-based pricing gets attention, but the market today still mostly looks like some version of platform fee plus usage bundle. It’s legible to buyers, maps well to current products, and doesn’t require perfect agreement on what a “successful outcome” actually means. Outcomes will arrive first in use cases where they’re standardized and easy to verify.

One other line that stuck with us: what “fast” looks like today is crazy relative to five or ten years ago. Categories are forming in real time. Feature parity emerges in weeks. Speed itself becomes a meaningful — if temporary — moat. Teams that learn fastest, instrument fastest, and improve fastest buy themselves the time needed to build something deeper.

Nine lessons for building in voice AI.

Frameworks for founders and investors deciding what to build, what to back, and what to walk away from.

#1. Don’t build voice unless you can own the whole job

Mike said it directly: he would not want to spend his time building a voice agent company if he didn’t think it could do an entire job end to end. Pleasant-sounding demos that hand off to a human the second things get hard become features. Companies are built by owning a full workflow, not a moment inside it.

Action item: Map the job from first ring to final action. If your product can’t credibly close the loop, you’re building a feature for someone else’s platform.

#2. Sell against payroll, not the IT line item

Traditional SaaS made labor more productive. Voice AI, when it works, replaces parts of it. That changes where the budget comes from. A workflow that supported a $30K SaaS ACV can support something closer to $150K when an AI agent actually performs the job end to end.

Action item: Price against the loaded cost of the human doing the work today — not against the seat-license comp you’d benchmark to in 2021.

#3. Infrastructure is table stakes. Distribution is the moat.

Low latency, reliability, model quality — all required, none differentiating. Feature parity is reached in weeks now, not years. The deeper moat comes from distribution, workflow ownership, trust, evals, and the feedback loops generated by real calls happening in production.

Action item: Audit how much of your roadmap is infra plumbing vs. distribution and workflow depth. Rebalance accordingly.

A quick word from our incredible partner,
Parafin - the leading embedded capital partner.

• 4.8 star rating on Trustpilot, with over 500 reviews [source]

• 71% CSAT rating across all partners

• 84 NPS

• Serving top Fortune 500 companies, including Amazon, Walmart, and DoorDash

Check Out Parafin

#4. The limiting factor is trust to act, not model capability

The models can already understand more than buyers expect. The real question is whether buyers trust the agent enough to let it move money, resolve a claim, complete a booking, or change a mission-critical record. That confidence isn’t built by vibes — it’s built by governance, observability, and repetition.

Action item: Treat permissions, audit trails, and rollback as product surface area.

#5. Evals are the compounding moat

Mike’s view: the compounding advantage comes from real-time evals and tests running in production. Every interaction becomes product data. Every signal can be measured against expected performance. The companies best at instrumenting, interpreting, and improving from those signals build a real flywheel around quality.

Action item: Ship eval infrastructure on day one. The team that learns fastest from production calls compounds faster than the team with the shinier model.

#6. Graceful handoff is part of the product

Escalation to a human isn’t failure. People usually call when something has already gone wrong or they’re in an edge case. The job isn’t 100% automation on day one. It’s to know where the boundaries are, solve what the system can solve confidently, and fail over elegantly when it can’t.

Action item: Design the handoff like a feature, not an apology. The smoothness of the human transfer is where trust is built — or destroyed.

#7. Pricing is platform + usage — for now

Outcome-based pricing gets the attention, but today’s market mostly looks like a platform fee plus usage bundle. It’s legible to buyers, maps to current products, and doesn’t require perfect agreement on what a ‘successful outcome’ is. Outcome pricing arrives first in use cases where the outcome is standardized and easy to verify.

Action item: Don’t force outcome pricing before the outcome is measurable. Use platform + usage as the bridge from the SaaS era into the agent era.

#8. Regulated industries adopt faster than you’d guess

Counter-intuitive but true: healthcare, insurance, and financial services have been among the fastest-adopting sectors. They care deeply about workflow structure, control, and compliance. If voice lets them gate conversations, validate identity before releasing information, and monitor agent behavior rigorously, they move faster than less-regulated markets.

Action item: Lead with controls, not capability. In regulated verticals, the compliance story is the sales story.

#9. Hunt the messy categories, not the obvious ones

Scheduling, debt collection, recruiting — already crowded. The interesting opportunities are where voice enables something previously too nuanced, messy, or expensive to scale: legal intake, expert-network conversations, mixed-modality workflows combining voice with screen or imagery, field inspections in industrial, roofing, construction.

Action item: If your category has a clean ROI deck and ten well-funded competitors, you picked the easy wedge. Go find the operationally dense one.

Where voice is already a knife fight — and where the real opportunity still hides.

One of our favorite parts of the conversation. Mike’s read on where founders are piling in, and where the next category-defining companies are most likely to be quietly built.

Mike’s most provocative line:

That sounds extreme until you think about how bad most current service experiences are. If the agent is faster, more accurate, always available, and actually resolves the issue, the old assumption that humans are always preferred quietly breaks down.

Voice AI isn’t a feature added to software.
It’s a new chapter in software itself.

If you’re building — or backing — voice AI today, Mike’s worldview comes down to four commitments:

1. Pick an economically meaningful job. Not a demo, not a feature. A workflow with real labor cost attached, where ownership of the job rewrites the budget.

2. Build for trust to act. Governance, observability, permissions, audit trails. The buyer’s confidence to hand over the keys is the actual product.

3. Instrument great evals. Treat every call as product data. The compounding advantage is the eval system, not the underlying model.

4. Earn the right to own the workflow. Start with a wedge you can credibly close, expand into adjacent steps, and let graceful handoff carry you until the system can carry itself.

Not every voice company will matter. Not every demo will become a business. But the founders who pick economically meaningful jobs, build real trust, instrument great evals, and earn the right to own the workflow end to end — those are the ones with a shot at building very big companies.

Watch Episode Here

Listen To Episode Here

Thanks for reading LINEAR. I reply to every email…
Have any questions, feedback, comments? Let us know, we work for you!

Shaped by lessons from growing and scaling 20+ vertical SaaS platforms, this guide explores what AI readiness actually means for software providers. Read the guide.

Most software is going headless

For the last 20 years, most software value was packaged inside the UI.

That was the bundle.

You sold the workflow.
You sold the dashboard.
You sold the reporting layer.
You sold the place where the user logged in and did the work.

The product was the screen.

And that worked because humans were the operators.

Humans had to know where things lived. They had to click through tabs, navigate record pages, update fields, run reports, approve tasks, move work from one system to another, and string together a process using the interface as the coordination layer. That made the UI incredibly important, because it wasn’t just presentation — it was behavior-shaping infrastructure.

But agentic software changes that.

If an agent can understand intent, retrieve context, call the right tools, reason across multiple systems, and execute a workflow on the user’s behalf, then the software no longer needs to be primarily optimized for a human operator sitting inside the product. It increasingly needs to be optimized for machine access, orchestration, permissions, and reliability. OpenAI’s ChatGPT agent is explicitly built to move across connectors, websites, browsers, APIs, and terminals to complete tasks. Anthropic’s computer use capability similarly shows models operating software by looking at the screen, clicking, typing, and navigating interfaces designed for people.

That’s the core headless thesis.

A lot of software won’t necessarily die.
It will just lose its head.

Meaning: the software still exists, but the human-facing interface is no longer the primary source of leverage.

Instead, the software gets reduced — or elevated, depending on how you look at it — into its underlying components:

• structured data

• workflow logic

• permissions

• audit trails

• compliance controls

• APIs, MCP tools, and commands agents can call

That’s not a small shift. That is a fundamental reordering of where value sits in the software stack.

The other problem is if the AI UI layer wants to do what your headless software does, they will do it, and you will die….

That’s why becoming the AI UI layer in verticals is where i’m spending my time.

The cleanest example of this showing up in the market is Salesforce’s Headless 360 announcement.

Salesforce said the quiet part out loud: instead of burying capabilities behind a UI, expose them so the platform is programmable and accessible from anywhere. They describe Headless 360 as the capabilities agents need most, exposed as an API, MCP tool, or CLI command, across the breadth of Salesforce workflows. They also say plainly that if your platform requires humans to click through UIs to make progress, it is not ready for the agentic enterprise.

That is a massive statement.

Because Salesforce is one of the most important systems-of-record companies in enterprise software. For decades, “using Salesforce” meant working inside Salesforce. Open the app. Navigate the object. Update the record. Run the workflow. Now they are telling the market that the future is not just humans working inside the application — it is agents acting on the system directly. Trailhead makes the same point even more clearly: in the agentic enterprise, agents don’t click through pages, buttons, and tabs; they call APIs, invoke MCP tools, and run CLI commands directly.

This is why I think so many point solutions and horizontal SaaS tools are about to get pushed down the stack.

If your product’s main value proposition is that a human logs in and manually performs a repeatable workflow, that position is getting weaker. If the steps can be abstracted, coordinated, and executed by an agent, then the UI becomes less of a moat. The value migrates below the interface into the quality of the data model, the richness of the workflow graph, the trust and permissioning layers, and the ability to safely support machine execution. a16z’s framing is useful here: in the agentic era, defensibility shifts downward into data models, permissions, workflow logic, and compliance, and upward into networks, proprietary data generation, and real-world execution.

That doesn’t mean every software category gets commoditized overnight.

Some products will remain highly human-centric for a long time. Some workflows will stay collaborative, exception-heavy, or emotionally nuanced enough that the interface still matters a lot. And many companies will live in a hybrid world where humans and agents work side by side for years. Showpad’s response to Salesforce is instructive here: they explicitly say the future is both humans and agents, and they’re building dual-track — strong UX for people while enabling agent access in parallel.

That’s probably the pragmatic transition path for most of the market.

But even in that hybrid state, the power structure changes.

Because once users stop starting in the native app, the real battle is no longer just over the system of record.

It’s over the AI UI layer.

Who owns the place where the user expresses intent?

Who owns the conversational surface?
Who owns the orchestration layer?
Who decides which tools get called, in what sequence, with what context, and with what permissions?

That is the new choke point.

And everyone serious in software can see it.

OpenAI is pushing toward a world where ChatGPT becomes a layer that can act across apps and workflows. Microsoft is explicitly combining agents and workflows so AI can reason where processes are ambiguous and workflows can execute where consistency matters. Salesforce is trying to make its system programmable for agents while also shipping an Experience Layer that lets rich interactive components render across Slack, mobile, ChatGPT, Claude, Gemini, Teams, and other clients.

That Experience Layer point is especially important.

Salesforce describes it as a UI service that separates what an agent does from how it appears, so interactive components can render natively across multiple surfaces. In plain English: build the logic once, but let the interface show up wherever the user already works. That matters because it means the fight is no longer “which app do users log into?” It becomes “which layer owns the interaction, even if execution is happening elsewhere?”

That’s why “headless software” is not just an architecture idea.

It’s a market power idea.

In the old model, the best product often won by becoming the place where work happened.

In the new model, the winners may be the ones that:

• own the deepest context

• control the critical workflow

• enforce the trust layer

• and sit closest to the point where intent becomes action

That is a very different game.

And it is bad news for thin point solutions that are mostly UI wrappers around relatively generic functionality.

If your wedge is “we have a slightly better dashboard,” you should be nervous.

If your moat is proprietary context, embedded workflow ownership, regulated execution, payments, domain-specific data, or a mission-critical system of record, you are in a much better position. Because in a headless world, the closer you are to the trusted execution layer, the more durable you become.

So my take is pretty simple:

Nearly all software is moving toward some degree of headlessness.

Not because interfaces vanish completely.
Not because apps stop mattering.
Not because every workflow becomes fully autonomous.

But because the center of gravity is shifting.

From screens to systems.
From navigation to orchestration.
From UI to execution.
From software you use… to software agents use for you.

And the fiercest battle in software over the next few years may be the battle to own that new AI UI layer.

The Story of AIDoc & Their Massive Series E Round

How a radiology wedge became a clinical AI operating system

Aidoc is one of the clearest examples of what a real vertical AI winner looks like.

Not a generic copilot. Not a horizontal wrapper. Not “AI for healthcare” in the abstract.

It started in one painful, high-stakes workflow, earned trust where mistakes actually matter, and then expanded into platform, workflow, and enterprise infrastructure. That’s the playbook.

1) Founding story: start where the pain is brutal

Aidoc was established in 2016. The company says its founding team came out of deep AI and defense-tech backgrounds, with Talpiot-trained founders bringing algorithmic, computational, medical, and research expertise into one company. CEO and co-founder Elad Walach says he began in the elite Israeli Defense Force technology program Talpiot and later led AI research in the Israeli Air Force; co-founder and CTO Michael Braginsky brought a medical-image-processing background and helped define a roadmap covering more than 75% of common acute CT pathologies. In other words: this was not “let’s bolt AI onto healthcare.” This was a technical founding team choosing a vertical where accuracy, speed, and workflow integration actually change outcomes.

The original wedge was smart too. Aidoc first focused on helping radiologists reduce turnaround time and improve quality by flagging acute anomalies in real time. That’s important. They did not start by promising to reinvent all of medicine. They started with one clear buyer pain: overloaded imaging workflows, rising scan volume, and costly delays in diagnosis. That’s classic vertical software strategy—begin with a sharp wedge inside a mission-critical workflow, then expand once trust is earned.

2) Product: from point solution to operating layer

Aidoc’s first chapter was essentially workflow-aware clinical AI for radiology. But the current product story is much bigger.

Today the company is built around two core assets:

1. CARE — its clinical foundation model

2. aiOS — its enterprise AI operating system for health systems

That product evolution is the whole story.

A lot of AI companies get stuck selling a single model for a single use case. Aidoc appears to have used the initial radiology wedge to climb into infrastructure. aiOS is described as the layer that runs, orchestrates, and governs clinical AI across the health system. It integrates with PACS, EHR, mobile, worklists, and care tools; supports urgency-based triage; provides validation, drift detection, override tracking, and analytics; and is positioned as the way hospitals move from pilot projects to enterprise-scale deployment without “sprawl.” That’s a huge jump up the value chain.

The most important recent product milestone is FDA clearance for what Aidoc calls healthcare’s first comprehensive foundation-model AI triage solution. The company says CARE now supports a workflow combining 11 newly cleared indications with 3 previously cleared ones, with mean sensitivity of 97% and mean specificity of 98% in the FDA-reviewed pivotal study. More strategically, Aidoc says this is the first FDA clearance of a comprehensive set of double-digit acute indications powered by a single foundation model. That matters because it suggests the company is no longer just shipping separate narrow models. It’s building a scalable clinical intelligence layer.

This is what I like most about the product.

They are not selling “AI magic.” They are selling:

• earlier detection

• faster triage

• cleaner workflow routing

• better care-team coordination

• governance

• enterprise deployment

That is how vertical AI becomes durable.

3) GTM: enterprise, multi-stakeholder, and trust-heavy

Aidoc’s go-to-market motion is exactly what you’d expect in a serious healthcare vertical AI company: slow at first, then powerful if you get it right.

This is not bottoms-up PLG. This is not swipe-a-card SaaS. This is enterprise clinical workflow software with regulatory weight.

The company’s public materials point to health-system sales centered on measurable operational outcomes: improved efficiency, shorter length of stay, faster diagnosis, reduced silos, and ROI. It also appears to sell into multiple service lines—radiology, neurology, vascular, cardiology, emergency care—which is important because it expands ACV after the initial wedge. Once you are trusted in one workflow, you can use the same platform and integration layer to expand across adjacent clinical domains.

The integration strategy is a big GTM advantage. Aidoc says aiOS is vendor-agnostic and connects with PACS, VNA, worklists, EHR, scheduling, and communication interfaces. It also says it is available within Epic’s App Orchard and can integrate with Radiant to provide acuity-based feedback in radiologists’ workflows. In healthcare, integration is distribution. If your product lives inside the clinical systems teams already use, adoption friction drops and your ROI story gets easier to prove.

I also like the partner strategy. Aidoc has both reselling partners and AI application partners. That means it is not just selling a closed suite; it is positioning aiOS as the orchestration and governance layer through which other clinical AI applications can be deployed. That is a much more ambitious GTM posture than “buy our algorithm.” It says: make us the control plane for AI inside the health system.

4) Customer success: this is where the company seems to separate itself

Most AI companies talk model performance.

Aidoc talks rollout, workflow, governance, and speed to value.

That’s a tell.

On the aiOS page, Aidoc highlights several scaled deployments:

• Advocate Health: 69 hospitals, 8M+ annual imaging studies, 22-site rollout, nearly 63K patients/year projected to benefit

• Hartford HealthCare: 500+ care locations, first sites live in 3 weeks, 17 FDA-cleared algorithms deployed plus care coordination and patient management tools

• Mercy: 1,000+ facilities, 12 use cases deployed system-wide in 3 months, urgent and incidental findings triaged in under 5 minutes, down from 20

Those are not vanity metrics. Those are customer-success metrics.

They show that Aidoc understands a truth most AI founders miss:

In the enterprise, the model is not the product. The implementation is the product. The workflow change is the product. The internal buy-in is the product.

The Hoag case study reinforces this. Aidoc highlights how radiology leaders built internal buy-in by starting with pulmonary embolism and vascular workflows, demonstrating shorter stays, improved outcomes, and cost savings, then expanding once the ROI case was clear. That is textbook vertical expansion: prove value in one high-pain workflow, create champions, then widen the footprint.

If I were underwriting this business, that’s what I’d focus on.

Not just “does the model work?” But:

• how fast can they go live?

• how many use cases can they layer in?

• how sticky is the workflow once deployed?

• how much internal political capital does the champion need?

Aidoc’s public stories suggest they’ve gotten very good at that layer.

5) Revenue: likely large ACVs, land-and-expand economics

Aidoc does not publicly disclose revenue in its own materials, so any revenue discussion has to be intellectually honest.

What we do know is this:

• the company says it is deployed across nearly 2,000 hospitals

• supports decision-making for roughly 70 million patients each year

• analyzes more than 110 million patient cases

• and is clearly selling enterprise software into major health systems

A third-party database, Prospeo, estimates annual revenue at roughly $95M, but I would treat that as directional rather than verified fact. Still, even if you haircut that estimate, the public customer scale suggests this is no tiny experiment. This is a real enterprise vertical AI business with meaningful commercial traction.

My read on the revenue architecture is that Aidoc likely benefits from a classic vertical AI compounding motion:

• initial wedge into one service line

• expansion into adjacent use cases

• platform value from aiOS

• additional monetization through broader orchestration/governance footprint across the health system

That tends to create larger ACVs, better net revenue retention, and more strategic importance over time. That last part matters most. Once the product becomes the layer managing multiple clinical AI tools and workflows, ripping it out gets much harder. That’s when revenue quality improves.

6) Fundraising: massive round, but the more important signal is why it got done

Aidoc announced a $150M Series E in April 2026 led by Growth Equity at Goldman Sachs Alternatives, with participation from General Catalyst, SoftBank Vision Fund 2, and NVentures. The round brought total funding to over $500M.

That’s obviously a big round.

But the deeper signal is the narrative behind it.

Aidoc is not pitching “we have one impressive model.” It is pitching:

• a clinical foundation model

• an enterprise operating layer

• regulatory rigor

• scaled deployment

• measurable customer ROI

• and a path toward end-to-end workflow automation, including draft reporting

Elad Walach framed the round as “the end of the beginning” for clinical AI and said the capital will go heavily into R&D: expanding CARE to cover hundreds of diseases, upgrading aiOS to support scale, and investing in AI-generated draft reporting. That is exactly the kind of post-wedge capital allocation you want to see. Use the first phase to prove trust and workflow fit. Use the next phase to widen disease coverage, deepen infrastructure, and automate more of the value chain.

Also worth noting: Goldman’s quote emphasizes efficiency, shorter length of stay, and measurable financial returns. That’s investor-speak meeting buyer-speak. The company isn’t being funded on vibes. It’s being funded on the idea that clinical AI is moving from pilot project to operating necessity.

7) Why Aidoc matters

Aidoc matters because it shows what the best vertical AI companies are becoming.

They don’t stay point solutions.

They start as a wedge. Then they become workflow. Then they become system layer. Then they become control plane.

That’s the climb.

Aidoc started with radiology anomaly detection. Now it is trying to become the enterprise operating system for clinical AI across the health system. That is a much bigger business than “AI for scan reading.”

And there’s another reason this is important.

In vertical AI, the real moat is not just the model. It’s the combination of:

• workflow ownership

• trust

• deployment muscle

• integrations

• regulatory credibility

• and customer success at scale

Aidoc seems to understand that better than most.

Be intellectually honest about job security in an agentic world

Now let’s talk about an uncomfortable truth.

Too many companies are still speaking about AI as if it’s only a productivity enhancement.

That framing is becoming less credible by the month.

Agents are not just helping people write faster or summarize documents.

They are increasingly capable of doing real operational work:

• gathering information

• navigating systems

• coordinating tools

• executing repeatable workflows

• updating systems of record

• producing standard outputs

• handing work from one step to the next

And if that’s true, then companies need to be much more intellectually honest about labor.

Not every role goes away.
Not every function gets automated.
Not every org can move at the same speed.

But a lot of companies are keeping roles in place that they no longer really need in the same form.

They know it.

They just aren’t pulling the trigger.

I’ve seen so many companies with staff that they really don’t need anymore.

Why? Because saying “we can eliminate this role” is much harder than saying “we’re experimenting with AI productivity tools.” One sounds innovative. The other forces a real org redesign conversation. But sooner or later, that conversation is coming.

If a role is heavily concentrated in:

• repetitive information gathering

• system-to-system coordination

• standard process execution

• routine output generation

• structured follow-up and admin work

Then leadership should be asking whether that work should still live on the org chart, or whether it should now live inside an agent-enabled workflow.

That’s not fear-mongering.

That’s just operational honesty.

The companies that move first will redesign teams around agents and reallocate human talent toward judgment, escalation, relationship-building, and high-leverage decision-making. The companies that move slowly will keep carrying labor costs for work that the market increasingly expects software to handle.

So what should people do?

Don’t be the person defending the old workflow.
Be the person redesigning the new one.

Map where work is repetitive.
Identify where agents can already perform.
Figure out where humans still add real value.
Learn how to orchestrate tools, prompts, data, permissions, and workflows.
Show your org what AI-native execution actually looks like.

In short:

become the AI whisperer of your organization.

Because in an agentic world, job security won’t come from pretending disruption is exaggerated.

It will come from being the person most capable of helping the company turn that disruption into leverage.

Do me a solid and forward to a friend :-)

This week’s episode was a fun one. We flipped the format and put Nic Poulos in the hot seat, with Sierra Ventures stepping in to run the interview. And honestly, it made for one of the sharper conversations we’ve had yet — less founder storytelling, more raw investing frameworks, category formation, and what actually matters when you’re underwriting vertical AI at day zero.

Nic is a GP at Euclid, but the arc here matters: banking, startup/operator reps, AOL Ventures, co-founding Bowery Capital, then spending the last decade developing real pattern recognition around vertical software before leaning hard into what he sees as the next chapter: vertical AI. His core belief is simple: AI is not killing software — it’s creating a new generation of software winners.

And that’s what makes this episode so good. It’s not fluffy “AI is changing everything” content. It’s a practical conversation about where the wedges are, what actually creates defensibility, how the best founders validate demand before the product is done, and why fund size is quietly reshaping founder outcomes.

Watch Episode Here

Listen To Episode Here

This Weeks Vertical Titan: Nic Poulos (Co-Founder & GP @ Euclid Ventures)

Nic has spent 10+ years backing early-stage vertical software companies, with prior stops including Bowery Capital and now Euclid, which is explicitly built to invest pre-product, pre-traction, pre-everything with a focus on founder-market fit and inception-stage conviction. That background matters because his lens isn’t coming from theory — it’s coming from a decade of seeing how category leaders in vertical software actually get built.

What I liked most is that Nic is not chasing AI because it’s hot. He is chasing it because he thinks LLMs unlock new product wedges that vertical SaaS simply couldn’t access before: voice intake, document abstraction, automated drafting, workflow launch points, and products that can finally go after the much larger services budget sitting next to software spend.

That’s also why the Sierra setup worked so well. Sierra Ventures has been around for four decades, manages more than $2B in AUM, and has a long history in early-stage enterprise investing. So instead of a typical founder interview, you get a real investor-to-investor conversation on how vertical AI is being underwritten right now.

The core thesis: the next big vertical AI companies won’t just be “AI wrappers.” They’ll sit at the point where work actually begins, own a critical workflow, expand from wedge to system, and capture dollars that historically went to labor, BPOs, or fragmented back-office services.

Watch Episode Here

Listen To Episode Here

Below — lessons from Nic:

1) Underwrite the team and market, not fake pre-seed precision

At inception, Nic’s point is brutal and correct: $100K ARR is not meaningfully different from zero if you’re trying to build a category-defining company. The real underwriting variables are the founder, the market, and whether the team has the right to win.

Action item: Stop optimizing your deck for revenue. Make the case for why this team, in this market, at this moment, is uniquely positioned to win.

2) “Earned insight” is still the cheat code

The best founders are not tourists. They’ve lived the pain, sold into the buyer, built adjacent tools, or grew up around the problem. That earned insight is what lets them spot a wedge that outsiders miss.

Action item: Audit your unfair advantage in plain language. If you can’t explain why you specifically should build this, neither can your customer.

3) Start at the authoring layer

This was probably the cleanest framework in the whole episode. The best vertical AI products often start where a record is first created: a phone call, an intake form, a contract hitting the inbox, a claim entering the system. Own that moment, and you can control everything downstream.

Action item: Map your industry’s workflow back to the very first keystroke. Build there, not where the dashboards live.

A quick word from our incredible partner,
Parafin - the leading embedded capital partner.

• 4.8 star rating on Trustpilot, with over 500 reviews [source]

• 71% CSAT rating across all partners

• 84 NPS

• Serving top Fortune 500 companies, including Amazon, Walmart, and DoorDash

Check Out Parafin

4) Don’t build “the agentic command center for X”

That’s the solution-first trap. If your product mostly vacuums data out of systems and summarizes it, you are building directly into the teeth of OpenAI, Anthropic, and every horizontal platform. That’s not great territory.

Action item: Pressure-test your wedge against ‘Claude 14.’ If a better base model flattens you, your moat is positioning, not product.

5) Sell before the product is ready

One of my favorite parts: Nic loves founders who can create commercial pull before the product is fully built. If customers will become design partners, sign LOIs, or start paying early, you’ve proven urgency — and urgency is gold.

Action item: Run a paid pilot or signed LOI before you write the second sprint of code. If the market won’t pull, the product won’t push.

6) Use cold discovery as a market stress test

Talking to 50–100 potential users is not a “nice to have.” It’s the work. And if a founder can convert cold outreach into active design partners quickly, that is often a stronger signal than polished demos or fancy decks.

Action item: Add up the labor and services dollars sitting next to the software line in your customer’s P&L. That’s your real TAM.

7) Go after services budgets, not just software budgets

This is one of the biggest insights in vertical AI. In many industries, services spend is multiples larger than software spend. If AI can replace headcount, manual coordination, outsourced processing, or consultants, the TAM expands dramatically.

Action item: Be ruthless about your wedge. Cut every adjacent feature until the one that earns the next conversation is undeniable.

8) Wedge first, then expand across workflow

The first product does not need to be the whole platform. It needs to be the right entry point. Once the wedge works, you expand into adjacent workflow, pull more actions into the system, and earn the right to become the operating layer.

Action item: Inventory the things that are obvious to you and unknowable to a horizontal team. Build the product around those.

9) Speed is a tool, not a moat

This is the nuance most people miss. Yes, speed matters. But no, speed alone is not defensibility. Speed is how you get to insight faster, ship the second product faster, and lock in workflow ownership before others catch up.

Action item: Before you sign, model the outcome your lead investor needs to be happy. If it isn’t the same as yours, walk.

10) Vertical nuance beats horizontal breadth

Nic’s implicit test is great: imagine a future where frontier models get much better. If your entire company gets flattened by “Claude 14,” you never had enough differentiation. Real moats live in vertical nuance, messy workflow logic, domain-specific edge cases, and system-level integration.

11) Become internal, not external

If the customer sees you as an outside service they can swap out, you’re vulnerable. If you become embedded in first-party workflow, connected to operational data, and tied to daily execution, you start to look like infrastructure.

12) Avoid the permanent services trap

Human-in-the-loop is fine early. Permanent dependency is not. Founders can absolutely use services to bridge the gap while models improve, but if the long-term model never escapes labor-heavy delivery, the ceiling gets lower fast.

13) The first technical hires matter disproportionately

The early engineering team doesn’t just write code — they set the culture, quality bar, and velocity norms for everything that follows. One great early hire compounds. One weak one lingers.

14) Don’t raise just because you can

This one should be printed and taped to every founder’s wall. More capital is not always better capital. Too much money too early can hide problems, delay hard decisions, and reduce the discipline that actually creates product-market fit.

15) Fund size is strategy

Nic’s critique here is sharp: if your investor only wins when you become a $5B+ company, that may not be aligned with what a great founder outcome actually looks like. Bigger funds don’t just change check size — they change incentives, exit math, and boardroom behavior.

The Anti-Playbook

A few traps from the episode that are worth calling out directly:

• Thinking “AI-native” means slapping a chatbot on old SaaS. That’s not reimagining workflow — that’s feature dressing.

• Believing speed alone will save you. It won’t. Speed only matters if it helps you discover and lock in something durable.

• Starting too far downstream in a workflow. If another system owns the first touch, you’re always negotiating from behind.

• Building for consensus instead of alpha. The biggest winners rarely look obvious when they start.

• Assuming a giant fund is automatically the best cap table addition. Sometimes it is. Sometimes it quietly destroys alignment.

We’re not watching the death of SaaS.
We’re watching category creation reopen.

If you’re underwriting — or building — vertical AI today, Nic’s playbook comes down to three commitments:

1. Understand the market deeply. Earned insight beats AI-native positioning every time. Tourists get filtered out by month three.

2. Wedge at the authoring layer. Own the moment the record is created. Everything downstream becomes negotiation from a position of strength.

3. Go after the services budget. The biggest outcomes will eat labor and BPO spend — not fight for another seat license next to the incumbent.

The founders who win here won’t be the ones with the fanciest AI positioning. They’ll be the ones who understand a market deeply, wedge into the workflow at the exact right point, create ROI fast, and then expand with ruthless focus.

Watch Episode Here

Listen To Episode Here

Thanks for reading LINEAR. I reply to every email…
Have any questions, feedback, comments? Let us know, we work for you!