Linear: A Vertical Software & Vertical AI Newsletter

It’s truly the best few days of the year in vertical software and vertical AI.

This Weeks Vertical Titan:
Mike Duboe (GP @ Greylock)

Mike Duboe is the kind of investor you want when the model itself is in flux, because he’s not a thesis guy, he’s a mechanics guy. He doesn’t defend marketplaces on faith; he can take one apart and tell you which gears still turn.

That operator pedigree matters. Mike ran growth at Stitch Fix from pre-scale all the way through its public debut, meaning he lived the two hardest problems in network businesses at ground level: manufacturing liquidity when no one knows you exist, and turning that liquidity into a compounding machine rather than a series of one-off campaigns. He left about a year after the IPO and joined Greylock, where he focused on “network businesses — so marketplaces and consumer networks,” which he says “led me to accidentally becoming more of a vertical software and application investor.”

That accident is our gain. People who think seriously about market structure are rare. People who have actually operated one are rarer. Mike is both, and it shows every time the conversation drifts toward the parts of the model everyone else takes for granted.

His core view on what’s happening right now: AI isn’t creating “AI-native marketplaces.” It’s making more markets addressable to the marketplace model than ever before. Transactions that were previously human-to-human, high-complexity negotiations — the kind that lended themselves to brokerages rather than streamlined marketplaces — are now things agents can handle. And in verticals where no online catalog has ever existed, AI creates the ability to build one that didn’t exist before. That reframe — addressability expansion, not marketplace replacement — is the most useful lens I’ve heard on this topic in a while.

He’s also refreshingly honest about where the model is exposed. His take: “Nowadays, it’s going to be insufficient to just charge for the match.” That line drives the whole playbook section below.

1. The match fee is exposed — move downstream of the transaction

For decades, the core value of a marketplace was the match, and the match is what you got paid for. Mike’s argument is that agents quietly threaten that: if agents can match supply and demand on both sides without a marketplace interface, then charging for the match itself gets compressed. Expect take-rate compression in some categories.

But this is an opening, not a death spiral. The monetization moves downstream of the transaction — credit guarantees, exception handling, verification, traceability. Those are areas where agents are actually smart, and where the marketplace can own something agents can’t easily replicate.

The takeaway: If you’re building a marketplace or a vertical platform that touches commerce, stop building your P&L around match fees. Build it around what happens after two parties find each other — because that’s where durable value (and defensible revenue) is heading.

2. The proprietary catalog is the new system of record

Mike’s most specific investment insight came from High Stock, a liquidation marketplace for excess inventory. The unlock wasn’t the technology — it was the discovery that a whole “dark market” of cross-border distributors quietly buy near-expiration US beauty products and move them overseas. These buyers have no websites, no online catalogs — they exist entirely offline, found through conferences and phone calls. By onboarding them, High Stock ended up with proprietary SKUs in a market that had no legible catalog.

That’s the pattern Mike looks for in B2B: “How are you actually going and creating the catalog versus just piggybacking one that exists?” If the catalog is already legible to agents, agents can bypass you. If you built the catalog nobody had, you’re the system of record. As I put it in the conversation, that’s a proprietary data moat in the agentic era.

The takeaway: Pick verticals where the supply side has never been digitized. The messy, offline, relationship-based parts of a market aren’t a problem to route around — they’re the wedge. If agents can already see the catalog, you don’t have a business; you have a UI.

3. Differentiated logistics beats liquidity when you’re the challenger

Palm Street built a consumer marketplace in categories Whatnot overlooked — plants first, then reptiles. On paper, Whatnot had the liquidity lead. What pulled sellers over? Fulfillment. Shipping a live plant — let alone a reptile — is genuinely non-trivial, and Palm Street’s specialized logistics expertise was enough incentive for sellers to move despite the liquidity gap.

Mike’s framework: what comes downstream of the transaction ends up being a reason to use a marketplace instead of going direct through agents. Liquidity gets you in the door; operational specialization keeps you in the building.

The takeaway: If you’re going second into a marketplace where the winner has liquidity, don’t fight the match — fight the fulfillment, the exception handling, the compliance, the physical weirdness the winner doesn’t want to deal with. Those are moats agents can’t talk their way around.

4. Start with vertical SaaS on one side, add the marketplace later

The classic marketplace-to-vertical-SaaS pivot is well-trodden. Mike lived the data point: Faire, Pepper, Bori, Resnovie — most of the B2B marketplace theses from his early investing days ended up eschewing the marketplace model entirely, building vertical software first, and monetizing on SaaS and payments rather than the match. Pepper eventually layered on an ad network at real scale. But the core model is SaaS, not match fees.

The interesting question the three of us landed on: does AI change the odds? My answer in the episode was that the two models start to blend into one. You get one side of the ecosystem with classic vertical AI tooling, and then over time you spin up the other side — the way a law-firm tool could eventually push clients to lawyers and take a commission on top of SaaS fees. The line between “vertical AI product” and “vertical marketplace product” is going to dissolve, because at some point you’re both the workflow and the transaction.

The takeaway: Don’t feel obligated to solve two-sided liquidity on day one. Enter with the easier side and a valuable tool, build trust and data, and treat the marketplace as an expansion product — not the founding premise. The cold start problem isn’t gone; AI just made it optional to confront immediately.

5. Consumers won’t hand over the transactions that feel good

Here’s the counterweight to all the agentic-commerce hype. Mike thought about this deeply during his Stitch Fix years, and his conclusion: a lot of consumers don’t actually want frictionless shopping. The instinct to automate everything misses that shopping itself is the product for a meaningful slice of purchases.

His split: where the objective is clear, where search and matching is easy, and where shopping feels like work — agents should and will take it. Think replenishment, commodities, the stuff you automate on Amazon anyway. But where the process of shopping is expressive and fun — fashion, discovery, surprise — AI’s role is to make the experience better, not to remove it. That’s why Whatnot became the outlier marketplace of the past decade: it was fun.

The takeaway: When you design for agentic buying, sort your catalog into “chore” and “joy.” Automate the first. For the second, build AI that delights rather than transacts — because the moment you turn a joy purchase into a utility purchase, you’ve destroyed the value you were supposed to capture.

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.

6. Trust systems were built for humans. Rebuild them for agents.

One of the sharpest one-liners in the conversation: “Most marketplace trust systems — I think about eBay stars — were built for humans. What’s the equivalent built for agents?” The answer is probably verified outcome data — not star ratings, but auditable performance: did the thing arrive, did it match spec, did the counterparty perform? That’s the currency agents will trust.

This has a practical implication for the “dark marketplace” dream — agents transacting with agents while you sleep. The real constraint isn’t technology; it’s that the agent needs to think like you’d think. Abstraction of buyer judgment is the hardest and most valuable asset in the whole system, and you can only build it with deep engagement over time.

The takeaway: If your marketplace or vertical platform has a reputation layer, start asking what the agent-native version of it looks like. Verified outcome data, performance guarantees, and exception-handling records will matter more than human star ratings in an agent-mediated world.

7. Embeds ads are the vertical AI monetization story

This was the section of the conversation where everyone leaned in. Mike’s example: Open Evidence, the clinical AI company, which converted point-of-time physician usage into massive ad revenue essentially overnight — because pharma has huge ad budgets and there has never been a better moment to reach a doctor than while they’re actively reasoning through a case at the keyboard.

I’ve been sold on this since Uber turned on embedded ads and added roughly a billion to the bottom line. And generative AI is the perfect vehicle for it. You’re serving one industry. There are a finite number of companies desperate to reach those users. The user is mid-workflow, talking to an expert (the AI), with intent visible in every token. An ad served there isn’t a banner — it’s contextually relevant information, delivered at exactly the right moment, tied to what the user is already asking about. It’s a no-brainer.

8. Clear ad labeling is a trust feature, not a betrayal

There’s a reflexive worry that ads inside AI products will break user trust. Mike takes the opposite view, and I think he’s right: when it’s clear what is and isn’t an ad, it builds confidence that the default responses are unbiased. Transparency becomes a feature.

The open question, as he framed it, is whether anyone has built the right native ad unit — the one that enhances engagement rather than trading it away. He points out that Instagram ads arguably make Instagram more engaging, which is an absurdly high bar. Right now, most AI ads are effectively search ads (which is why Open Evidence works). The unlock is a native unit nobody has invented yet — creative, bespoke, per-publisher. My favorite idea from this stretch was Luke floating an API key anyone could drop into a vertical AI app to turn on embedded ads in ten seconds flat. The SSP layer for consumer AI apps is probably early. The B2B version is lower volume and higher value — and, for now, inherently consultative.

The takeaway: Vertical AI founders should treat embedded ads as a first-class revenue line, not a sellout. Ship contextually relevant, clearly labeled placements; watch the engagement data; and expect that the first five ad formats you try aren’t the right ones.

9. Growth talent is the scarcest resource in the AI era

Mike grew up in the early-2010s school of systematic growth thinkers — people who engineered compounding loops instead of running tactics. His observation: in the AI era, basic products are trivial to ship, so what differentiates winners is structural distribution, network effects, and durable brand. That makes systematic growth talent more valuable than ever — and there’s a shortage, because technical founders historically under-valued distribution, so the incentive to train the next generation of growth thinkers collapsed.

The current wave of AI-native marketers are scrappy and fast, but “lean more tacticians versus systematic growth thinkers.” And since the tactics are changing faster than ever, being a tactician without principles atrophies quickly.

The takeaway: If you're a vertical AI founder, the best headcount decision you'll make this year is hiring someone who thinks in systems about distribution — not someone who can spin up channels fast. And if you're a growth person: the window for tacticians is closing. The people who can abstract the principles will own this era.

10. The marketplace underwriting checklist, rewritten for agents

Mike ended with the most useful segment of the whole episode: the Bill Gurley / Jeff Jordan evergreen marketplace attributes — and what needs rewriting for the agent era.

Still evergreen:

High fragmentation on both sides
High-frequency transactions (or a credible path to frequency)
Getting into the payment flow
A founder with an edge on aggregating demand — typically spiking on industry know-how or growth
A point of view on disintermediation — which in this era means “going to agents”

Getting rewritten:

Take rate assumptions — the match fee is exposed (playbook #1)
Reputation systems as lock-in — less confident; agent-native trust isn’t proven yet
The 10x interface experience — matters less; he’d rather underwrite a founder’s understanding of market structure and their ability to spin up liquidity fast

That last shift is worth sitting with. For a decade, the marketplace bar was product magic. Mike is saying the new bar is market-structure literacy: do you understand the shape of the market, where the supply hides, who holds the inventory, why they’ll move? Can you generate liquidity faster than anyone else? Because interfaces are replaceable now. Market understanding is not.

That’s the episode. As always, the whole conversation is worth your time — Mike learns in public, which is the best kind of guest. If you’re building in vertical AI, vertical SaaS, or anything touching B2B commerce, this one has quite a few frameworks to steal!

See you next Wednesday.
— Luke Sophinos

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

Linear #193: The $100M Race -- How Vertical AI Is Eating a Decade in Three Years

Mon, 07 Sep 2026 01:02:07 GMT

The Vertical Software Summit is ~60 days out...

If you’ve been waiting to grab a ticket, now is the time
(and if you need a free one reply to this and we will get you one).

Grab a ticket here

It took Procore 14 years to reach $100M in ARR. It took Harvey three.

Legora went from zero to $100M in ARR in eighteen months. OpenEvidence hit $150M in five. MagicSchool put up $10M in its first year of monetization. GC AI went from $1M to $10M in under twelve months.

Meanwhile, the last generation of vertical winners, the ones we all study, needed a decade on average to do what this generation is doing in thirty-six months…

This is not a rounding error. It is a structural change in how fast vertical software businesses can compound, and it changes how you should think about every vertical market on earth…

The Two Clocks: Legacy Vertical SaaS vs Vertical AI

Twenty years of vertical software history produced a clear tempo. Build the system of record, grind through multi-year implementations, expand seat by seat, attach payments, cross $100M sometime around year ten, go public a few years after that.

Here is what that clock actually looked like:

Veeva was the sprinter of its generation: seven years to $100M with fewer than 200 customers and only $12M raised. That was considered one of the greatest executions in software history.

Procore was the grinder: a decade to reach $5M, fourteen years to $100M, twenty-two to a billion. Clio needed 14 years to hit $100M and crossed $500M in May of this year, eighteen years after founding. Doximity, a genuinely elite business, took ten years to cross $100M.

Median time to $100M ARR for the vertical software greats: roughly a decade.
Median time to $500M for the ones that got there: well into the teens.

Now look at the clock the vertical AI generation is running…

Harvey crossed $100M in ARR in August 2025, three years after founding. By this summer it was at $350M. If the line holds, Harvey gets to $500M in under five years. Clio just did it in eighteen.

And then Legora broke the clock entirely: $100M of ARR in eighteen months from founding, in the same vertical as Harvey, with half the ramp. Eighteen months is faster than Wiz, faster than Deel, faster than nearly any software company ever measured — and it happened in law, a category the last generation needed a decade and a half to crack.

OpenEvidence is the outer edge: $50M to $150M in ARR in five months, faster than three quarters of every SaaS company ever measured. It reached 300,000 doctors in about a year. Doximity needed a decade to assemble that same audience.

That is the part worth internalizing. The audience was compounding for years while the revenue line sat flat at zero, and monetization was not a build — it was a switch. When a vertical AI product owns the daily habit of a licensed, high-value professional, the revenue ramp is not a function of go-to-market. It is a function of the day someone decides to charge.

Distribution is pre-built.
The wallet got bigger.
Adoption friction collapsed.

The three mechanisms

Distribution is pre-built. The vertical SaaS generation spent twenty years dragging laggard industries onto cloud and mobile. The AI generation inherits buyers who already purchase software, already hold budget, and already know their workflows are broken. Harvey never had to convince a law firm that technology matters. A decade of legal tech did that for them.

The wallet got bigger. Vertical SaaS sold seats, and seat budgets are small. Vertical AI sells completed work: a drafted motion, a resolved claim, a documented encounter. That spend comes out of labor and outsourced services budgets, which dwarf software budgets in every vertical. The AI company is not competing for the incumbent’s line item. It is eating the line item above it.

Adoption friction collapsed. The legacy winners fought implementations measured in years. The AI winners ride an interface that feels like a conversation and returns value in the first session. MagicSchool signed up six million educators without a sales force worth mentioning. That motion did not exist in 2011.

Two honest caveats, because they matter.

Speed is not margin. Harvey’s revenue carries real model costs underneath it in a way Veeva’s never did. A $100M AI ARR and a $100M SaaS ARR are not the same business, and the market will eventually price the difference.

And the vertical still sets the floor. GC AI compounds 23% a month because in-house legal budgets are centralized and decisions are fast. The model does not override the market’s metabolism. It removes the excuses.

That is not a faster horse. It is a different race.

The best vertical software businesses of the last cycle were decade-long builds, and the very best of them, Clio included, needed the better part of two decades to reach $500M. The best vertical AI businesses of this cycle are compressing that into three to five years.

Let’s look at the chart in aggregate…

This may be my favorite chart in newsletter history.
What a time to be alive…

The Map Was Already Drawn

In 2007, Peter Gassner left Salesforce to build software for one industry: pharmaceuticals. Everyone told him the market was too small. Veeva now does $3.4B in revenue and trades around $33B.

In 2011, Aman Narang and his co-founders started building point of sale for restaurants, a category investors had written off as a services-heavy graveyard. Toast now runs at roughly $5B in annual revenue.

In 2012, Ara Mahdessian and Vahe Kuzoyan started building software for plumbers and HVAC shops out of a garage, because their parents ran trades businesses. ServiceTitan crossed $1B in revenue this year.

Each of these stories has the same shape. A founder picked a market that looked boring or broken from the outside, spent a decade proving the wallet existed, educating the buyer, and wiring up the workflow. Then the business compounded for another decade.

Here is the part nobody is pricing in: every one of those stories is also a prospectus for the next one.

Every public vertical software company above a billion in revenue is a proof statement. It proves the demand exists. It proves the buyer can be trained. It proves the workflow is budgeted. Those are the three hardest things to establish in any vertical, and the incumbent already paid for all of them.

Which means the correct way to map vertical AI opportunity is not to chase what is trending. It is to look at where vertical software already produced massive winners and ask: who is the AI-native successor here, and does one exist yet?

Run the scoreboard:

Now notice where the vertical AI winners showed up first: medicine and law.

Doctors got OpenEvidence. Lawyers got Harvey, Legora, and GC AI. Why those two? Because Doximity spent a decade assembling the physician graph and proving doctors monetize, and Clio spent eighteen years proving legal budgets are real, real enough to build a $500M ARR business on top of them. When the AI products arrived, those markets were pre-sold.

The software winner is not the competition.
It is the proof of wallet.

And the successor’s ceiling is higher than the incumbent’s. Veeva sold software, a tool budget. Whoever builds the Veeva of AI sells labor and outcomes, an operating expense budget. The second wallet is five to ten times the size of the first.

Gassner proved pharma pays. Narang proved restaurants pay. Mahdessian proved the trades pay. Twenty years of vertical software history drew the map, marked the wallets, and trained the buyers.

The vertical AI generation does not need a better map. It needs to move faster on the one it was handed. And as this week’s numbers show, it already is…

The Best Value Creation Levers Don’t Stand Alone

The best Vertical SaaS companies I’ve worked with all have one thing in common: they don’t try to do everything.

Instead, they’re remarkably disciplined about where they invest. They evaluate every product decision, strategic initiative, and investment not only by its immediate return, but by what it makes possible next.

That’s a different way of thinking about strategy. If you spend enough time talking with software founders and private equity investors, you’ll notice a common pattern. Nearly every strategic conversation eventually turns into the same question:

Where should we invest next?

The answers are familiar. AI. Embedded payments. Pricing optimization. International expansion. New financial products. Services. Each has a compelling business case, and each has the potential to create meaningful enterprise value.

But I think we’re starting to ask the wrong question. Instead of asking which initiative will deliver the biggest return, leaders should ask:

Which investment will make every future investment more valuable?

Every Vertical SaaS company has more opportunities than it has engineering resources, investment capital, or organizational capacity. The challenge isn’t identifying good ideas. It’s deciding which investments create the greatest strategic optionality over the next five years.

One thing I’ve noticed is that the strongest SaaS companies rarely chase trends. They make a handful of strategic investments that continue paying dividends as the business evolves. Looking back, those decisions often seem obvious. At the time, they usually required saying no to a dozen other attractive opportunities.

Some initiatives create value. Others strengthen the platform in ways that make future initiatives more valuable. The best ones do both. Those are the investments that compound over time.

Take AI as an example. AI can automate workflows, improve customer support, and increase productivity. But its impact changes dramatically when it’s built on top of rich operational data, well-defined workflows, and years of customer context.

The technology matters. The foundation matters even more.

The same principle applies across many of the strategic decisions software companies are making today. Better operational data improves AI outcomes. Deeper workflow ownership strengthens customer retention. Financial infrastructure creates opportunities for embedded finance. Transaction data unlocks new insights.

Viewed independently, these look like separate initiatives competing for budget. Viewed strategically, they’re connected capabilities that reinforce one another.

Embedded payments is one example. For years, we’ve talked about payments as a value creation lever, and I think that’s still true. It can generate new revenue, strengthen customer relationships, and contribute to enterprise value.

But perhaps the bigger opportunity isn’t payments alone.

It’s what becomes possible once payments become part of the platform. Transaction data becomes more accessible. Financial workflows become more connected. Future financial products become easier to introduce. Each new capability increases the value of the next.

That’s why I believe the most successful vertical SaaS companies will increasingly think less about individual initiatives and more about building platforms that create future opportunities.

These are exactly the kinds of conversations I’ll be having with Mark Passifione and Daniel Burton of Xplor Pay during our executive discussion, Building More Valuable Vertical SaaS Companies, on September 10. If these questions are on your mind too, I hope you’ll join us.

→ Reserve Your Spot

The best value creation levers don’t stand alone. Neither do the conversations that shape them.

The next generation of Vertical SaaS leaders won’t win because they identified more value creation levers than everyone else.

They’ll win because they invested in capabilities that made every future lever more powerful.

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

Linear #192.5: Forward Deployed Is a Chainsaw with Quinn Litherland (Founder/CEO @ Revin)

Wed, 02 Sep 2026 14:53:00 GMT

Great in the right hands, a disaster in the wrong ones. Quinn Litherland on Forward Deployed — who to hire, how to pay them, and how to templatize it.

Every vertical AI company with a seed round is suddenly “forward deployed.” Palantir’s own executives made fun of the copycats on an earnings call: sales engineers in costume. And there is a reason everyone claims it — the model works. But it works like a chainsaw. Great in the right hands, a disaster in the wrong ones.

This week Nick and I sat down with Quinn Litherland, founder and CEO of Revin. They build verticalized automation for home services: HVAC, plumbing, roofing, remodeling, pool services. Think AI Native Services for Mid-Market/Enterprise Field Services. Three years old, just over $10M raised, out of New York. Their agents answer calls at 9pm, qualify homeowners, run a $30,000 roof sale across a four-week omnichannel campaign, and book the job back into ServiceTitan. Their FDEs sit in a random state in the U.S. every week running $250M remodeling businesses. And Quinn has thought harder about how to make forward-deployed actually scale — and how to avoid the two traps that kill companies that try it — than almost anyone we have had on the show.

This episode is a masterclass in the FDE motion: who you hire, how you pay them, where they sit, how you templatize the role, and how you keep from becoming a services shop with venture funding and a prayer.

Vertical Software Summit

Last chance before ticket prices go up September 1st!

We are already 80% sold out so if you’re considering coming to the Vertical
Software Summit this November, now is the time to grab your ticket.

All information at the link below.

It’s truly the best few days of the year in vertical software and vertical AI.

This Weeks Vertical Titan:
Quinn Litherland
(Founder/CEO @ Revin)

Quinn walked into one of the most fragmented markets in software: home services, a $650B-plus industry with 2.5 million providers running systems older than the startups selling to them.

Revin targets mid-market to enterprise: $25–30M family businesses that scaled regionally, plus PE platforms like Vertex doing $600M+ across 25–26 brands. ACVs run $70K to $250K. A roofing company pays that because a single new job is worth $50,000, and Revin can close five new customers in month one that would have leaked through the funnel.

The strategic insight worth studying is the reframe. Revin is not selling software. They are selling labor. Their workflows and agents are employees to these businesses. And that changes everything downstream — because a remote employee who is onboarded badly and doesn’t understand the business will never run operations effectively. You cannot just ship a dashboard and wait.

Revin is not selling software. We are selling labor.
Quinn Litherland, Revin

So Revin embeds. Deep, fast, from day one of a customer going live. At some point they are essentially running your business, and ripping them out becomes nearly impossible. The product’s atomic unit isn’t the UI. It is the agents, learning directly from every conversation, improving constantly. Quinn calls the endgame “headless companies”: a $500M roofing business run by five people and a stack of agents.

The core product insight is more interesting than the product itself. A roofing job isn’t a transaction; it’s a $30,000 sale that happens over a four-week omnichannel campaign. Someone has to answer the phone at 9pm, qualify the homeowner, follow up while they’re deciding, maintain context across SMS, email, and voice, and then hand off a warm, fully-qualified lead. Historically that “action layer” was people — call center reps doing the same motions for decades, leaking lead after lead. Revvin replaced that with agents that own the workflow end to end: inbound speed to lead, qualification, job-type matching, booking back into ServiceTitan or whatever the system of record is.

That is the vision. The engine that makes it real — and the reason this episode matters — is the forward-deployed motion. That is what this newsletter is about.

01. The Forward-Deployed Motion
Deploy, maintain, expand — in that order.
Renewals are the scoreboard.

That is the Revin mantra. And I love it.

Deploy: go live as fast as possible with the highest quality outcomes.

Maintain: don’t vanish at month three, which is where most FDE-heavy deals die.

Expand: if a customer is buying more workflows, they are happy, and “you crush these three workflows, now do these six” becomes the natural motion. When LTV goes from $70K to $250K to $1M, the renewal stops being a negotiation and becomes a planning session.

Hire founder-shaped operators, not sales engineers.

The profile is the story. Paul, one of Revin’s best FDEs, ran a bootstrapped car detailing business in Austin doing about $500K a month, built his own tools, exited, and joined. Today he manages a couple billion dollars of enterprise value in roofing and works with the C-suites of these businesses. Quinn’s line: Paul could leave tomorrow and run a roofing roll-up better than 95% of the private equity folks he meets. Implementation skills, expansion instincts, industry fluency, and deep empathy for operators because he was one. This is not a sales engineer. This is a founder who chose your company instead of starting his own.

Pay that person like an owner — because they are one.

Quinn points to Decagon as the model. An FDE who can make a million dollars this year if every account renews, plus a cut of expansion, will act like the CEO of their book of business. Renewal-based commission turns your most expensive cost center into a revenue engine. And the economics work: hire one of these people instead of the old pod of PM, engineer, CSM, and sales rep. Pay $250–300K plus commission and you replaced six heads with one. But you have to get the hire right, because a miss is more expensive and more painful than anything we have seen in software.

Put forward-deployed under product, not engineering.

The field is where the signal lives. Your FDE should be able to say, “I have seen this five times today, we need to solve it.” Under engineering, you build what you think is right instead of what the trenches are telling you. Revin’s FDEs are the product org’s nervous system. They are in middle-of-nowhere Ohio every week, understanding these operators in a way a PM staring at a dashboard never will. Handled well, the FDE motion forces you to actually understand the businesses you serve — stickier product, more durable value. Handled wrong, you are screwed.

Watch for the two traps, because both will get you.

This is the most important part of the conversation. Trap one: you become a services factory — running a services playbook on VC money, praying you will eventually automate the humans. Trap two: everyone is “forward deployed” and nobody is building repeatable systems — chickens with their heads cut off. Quinn’s hard truth: forward-deployed is often an excuse not to build great product, and a lot of teams will be screwed because no one is coming to save them.

The escape hatch is obsessive templatization. A year ago, onboarding meant flying to the customer and building from scratch. Today it is: “You are a $50M HVAC business, here are the semantics you will run into. 99% of the time we already know your qualification process, your job-type matching, your system of record.” The FDE handles the residual 1%. Every new customer must take measurably less human effort than the last. Otherwise you are not building a company. You are building a nonprofit consulting firm with venture backing.

The FDE role is getting less technical — and that is fine.

Teams are already using the term “AI deployment strategist,” and Revin’s newest FDEs are less and less technical. The core requirements: really smart product thinkers who can talk hand in hand with a customer, because they are embedded in the trenches. Quinn expects a wave of startups building internal tooling for this motion, which eventually lets the role migrate toward a CSM shell. The jobs-to-be-done don’t change — deployment, trust-building, feeding signal back to product — but the profile is shifting from engineer to operator.

Sell the person, not just the product.

Buyers of vertical AI are starting to demand it: “I don’t want to be your guinea pig. Who am I working with day to day?” Operators want to meet their deployment counterpart before signing, because they will change and improve these workflows together. Your FDE is now a sales asset. Show the operator the exact person who will sit in their trenches and perfect their workflows. If the motion is good, it sells itself. If it isn’t, your sales cycle will find out.

Stake out the action layer, and don’t stab the system of record bear.

For 20 years in home services, humans were the action layer. They interacted with a system of record, performed a task, and either automation pumped the data back or they copy-pasted it themselves. In the agentic era, the agent is the system of action. Quinn’s read: own the workflow and you win. It is the lifeblood of the business, and it matters relatively less where the data lives. And critically: do not announce you are now a field service management tool. The Podium/ServiceTitan blowup is the cautionary tale — Podium spent 18 months building a competitive FSM product, refused marketplace terms, and got shut off. Revin’s stance: be a great integration partner, pipe data back into the system of record, stay clear of the bear. If you own the action, you own the monetizable moment.

Purpose-built embedded capital for vertical platforms.

In low-ACV markets, the product has to be the answer — not more FDEs.

The uncomfortable question for anyone with a $5K ACV product: why does this need humans to embed it into companies that won’t pay enough to make the math work? If the buyer can’t pay for the deployment, the product has to get to a place where it doesn’t need one. Quinn’s first question to low-ACV teams is always, “Why is it so low, and what else could you provide that increases their outcomes?” And Nick’s corollary: the gap between “handles 90% of the workflow but you still log in” and “handles it soup to nuts” is enormous — $10–20K versus $100K, because offloading an entire process carries a major premium. If software ACV can’t grow, count the payments and transactional revenue attached to actions. The revenue is there; it’s just not in the SaaS line item.

The throughline of this conversation is that the hard part of vertical AI was never the model. It’s the trust, the buy-in, and the dirty work of embedding your people into businesses that have run the same way for 30 years — and then converting all of that into repeatable product before the services trap swallows you.

Quinn’s bet is that the teams who do that painful work properly are the ones who win the next decade of vertical software. I think he is right for AI Native Services Companies. Handled well, the FDE motion isn’t a cost of doing business. It’s the moat.

If you are building anywhere the action layer is still run by humans, this episode is required listening. Watch it on The Verticals YouTube channel, and subscribe to the newsletter right here.

See you next Wednesday.
— Luke Sophinos

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

Linear #192: Surrendering the UI - When You Should Go Headless

Mon, 31 Aug 2026 00:02:54 GMT

Today’s newsletter is sponsored by Xplor Pay.
Artificial intelligence is changing how software is built, customer expectations are evolving, and investors are rethinking what creates long-term enterprise value.

Join Xplor Pay and Luke Sophinos, Founder of Vertical SaaS Group, for an executive discussion exploring how AI, workflow ownership, platform strategy, and financial capabilities are reshaping the future of vertical SaaS. Register today and reserve your spot.

A Guide For When To Go Headless & When To Own The UI

The most dangerous assumption in vertical software right now is that owning the interface is always the prize. It was the correct assumption for twenty years. The company with the login page won.

The AI era inverts that logic, and most founders have not caught up.

A small number of horizontal AI platforms are going to own the general interfaces of the world. Everything else gets pushed into one of two camps: vertical AI operating systems that own deep industry workflows, or headless software that runs underneath everyone else’s screen. Both can be enormous. The difference is whether you choose your camp before the kings arrive, or get assigned one after.

This week: the framework for surrendering the UI layer, when to do it, when to fight like hell to keep it, and three vantage points on the same decision.

The Seven Kingdoms: where the interface is actually going

I wrote about this in Linear #174: The Seven AI Kingdoms. The short version: the AI software stack is reorganizing into seven distinct layers, not the flat SaaS landscape we came from.

Here is the mechanism worth sitting with. The horizontal kings are not going to go all the way down into every market. They cannot. Their economics forbid it. A platform worth a trillion dollars does not deploy its best model, its brand, and its distribution into a workflow that produces a few hundred million in software spend. The marginal cost of entering a big new market is basically a compliance wrapper, an industry landing page, and a sales team. For a small vertical, the same effort does not move their needle at all.

So the world filters cleanly. Healthcare, education, legal, financial services: the giants are already building the general UI there. Elevator repair, fire departments, art galleries, cemeteries: they will never come.

That means the interface layer of large white-collar verticals is being claimed by the kings. The interface layer of everything else belongs to the vertical AI operating systems: the industry-specific “Claude Cowork” that looks conversational on the surface but runs on real workflow depth, regulatory knowledge, and trust. And underneath both sits Kingdom VI, headless vertical software, quietly becoming a compounding infrastructure business.

Headless is not a consolation prize. It is a position.

Surrendering the UI Layer: When to Do It and Why

First, the distinction. The UI layer is the product your end user logs into. The interface owns the relationship, the daily habit, the data, the trust. Headless means you never see the end user. Your customer is a developer. They embed your engine inside their product, and their users never know you exist.

Examples everyone has seen from underneath: Plaid is headless despite every fintech app showing its logo on a bank selection screen. Twilio is headless. Stripe started as an API and only later earned the right to touch merchants directly. Shippo is headless. Pretty much everything in the e-commerce checkout flow you touch is someone else’s engine wearing the storefront’s clothes.

This is not new. Point Nine’s essay The Rise of Vertical APIs catalogued this a decade ago: Onfleet for delivery operations, Eligible for healthcare payments, Infermedica for diagnostic triage, Retsly for MLS data, Aftership for tracking, Commerce.js for storefront infrastructure. It split vertical APIs into two buckets: data APIs and infrastructure APIs. That essay aged perfectly. It just called the category too early.

One thing up front so nobody over-rotates: these are not prerequisites. Not every point below has to be true for headless to be the right call. But at least some of them should be, and the more that hold, the stronger your position. If you have one of these in full force, that can be enough. If you have three, you are building infrastructure on purpose.

Here is when you should go headless.

1. You serve a massive, broad vertical, especially white-collar with heavy screen time.

That is where the giants go. And we can learn exactly who by looking at who they have already targeted.

If your vertical is on that map, the giants are building the general UI for your customer as you read this. Do not fight that at the surface. You do not out-brand OpenAI inside a hospital. You win underneath them.

But there is a caveat worth stating plainly, because this is where the real openings hide. This point only applies if you are serving the broad vertical at the same altitude they are. The moment you go deeper, the game changes. Be the AI for chiropractors, not the AI for healthcare. Be the AI for behavioral health practices, not the AI for hospitals. Be the AI for a specific specialty, a specific workflow, a specific back office, and suddenly you are not competing with the horizontal kings at all. You are operating in a category they will never care about.

The rule for that move is simple and unforgiving: your slice has to be different enough, and ideally much more complex, that the broad tool is 10x worse than your solution. If a general model with a landing page gets you 80% of the way, the giant wins, because they have distribution and you do not. If your depth makes their general tool look like a toy, you win, because no amount of model quality replaces workflow specificity. Depth is the exception that beats scale.

2. The math does not work for the small spaces.

Here is a number worth sitting with. A $100B+ company needs multi-billion dollar outcomes to move its own needle. Not a billion. Multiple billions. A billion-dollar outcome is the same to them as your smallest ACV customer churning is to you. Which is quite wild to think about.

That one sentence explains more about the AI landscape than any model benchmark. It is why the kings go after healthcare and education and legal and not elevator repair. It is why their roadmap is finite and yours is not. And it is the quiet promise underneath the entire “own the UI” argument. They cannot chase you if you are too small to matter to them.

3. Your primary end customer is a developer.

If the buyer is an engineer who wants capability embedded, you are headless whether you admit it or not. The Point Nine essay is the playbook: data APIs, infrastructure APIs, compliance as a service. When your users are other products, you are not in a UI business, so stop spending like you are. No marketing site can fix a product that was built to be embedded.

4. You can solve one problem incredibly well, and it is big enough to still be a massive business.

Shippo. One multi-carrier shipping API with 40+ carriers underneath: rates, labels, tracking, address validation, all over a single REST interface. It never tried to be the storefront. It made shipping disappear as a problem, and every e-commerce platform became more valuable by embedding it. One problem, infrastructure depth, enormous TAM. That is the template. The mistake founders make is assuming that going headless means going narrow in ambition. It does not. It just means the surface is someone else’s and the engine is yours.

5. The trojan horse rule: a dominant industry ERP already owns the workflow.

If the entire industry runs on one system and the operators open it every morning, you do not win the login. You go headless, integrate into their screen, and become the layer they cannot live without. You win by being impossible to remove from inside the incumbent, not by being the second tab nobody opens.

And the trojan horse is just the starting point. There is a longer game underneath it. You integrate to get as much data as possible. You learn the workflows from the inside, the ones that never made it into any requirements doc. You map every inefficiency and every user complaint. And then, when the knowledge is complete and the moment is right, you build those workflows 10x better and launch the competing UI layer, this time from a position of total information. Every roll-up in software history is this play. The headless phase is how you earn the right to eventually own the screen.

Now the other side…

When to not go headless.
When to push hard and own the interface.

1. Your market is too small to move the needle for a $100B+ company.

If the kings will never come, the interface is the biggest prize on the board and it is yours for the taking. Niche is not a limitation here. It is a moat. The very math that pushes the giants toward healthcare is the math that guarantees they leave your vertical alone forever.

2. There is no dominant industry-specific vSaaS yet.

Open field. The interface is the prize. Take it before someone else does.

And let me be direct here, because I know a lot of you do not believe this anymore: this is the classic vertical software play, and it still works. There are still a ton of markets with four or five players that each own maybe ten percent. The other half of the industry is still running on paper, spreadsheets, or frankensteining together horizontal tools that were never built for them. That is not a mature market. That is an open market wearing a mature market’s clothes. The AI layer makes this easier, not harder, because you can reach the paper-based half of the market without building a twenty-person onboarding team.

3. The dominant vSaaS is showing wobbliness.

Private equity bought it and squeezed it. Customers hate it. Data can be pried out of it. R&D has stalled. That is an incumbent that stopped compounding, and the UI is winnable. Entrenched is not the same as strong. Some of the best businesses in this category were built by watching the category leader get hollowed out by financial engineering and simply walking in the front door.

4. You are already the market-leading end-to-end system.

Do not surrender what you already own. Put the AI layer into your own interface and widen the gap. The headless argument is for people who do not have the login page. If you already have it, you are not a utility. You are the throne.

That is the full framework. Now three vantage points on the same decision.

The Vertical Software Summit 2026

It is not a payments conference masquerading as a vertical software conference.
It is not an investor conference not allowing other investors in.
It is the only conference by vertical software and ai founders for vertical software and ai founders.

The Summit is founders, operators, investors, and all the companies that support vertical businesses talking to each other, in one room, without the noise.

Don’t wait until last minute! This will sell out soon!

Real Examples In The Wild

Metriport: the headless checkmark

Metriport is an open-source, universal API for healthcare data. One API for complete patient data, built on FHIR, providing real-time access to medical records for more than 300 million Americans. Your customer is a developer building a health product, and what they buy is access to every patient record in the country without building 40 EHR integrations themselves.

Now check it against the when-to-go-headless list.

Healthcare is the purest example of a massive, white-collar, high-screen-time vertical. Clinicians live in EHRs all day. And the giants are already here: OpenAI, Google, and Microsoft are all shipping healthcare products, while the clinicians’ actual screen is owned by Epic and the entrenched EHR incumbents. You are not winning that UI, and it would be incredibly difficult to try. And that is assuming you are serving healthcare broadly. Go a slice deeper, behavioral health, therapy practices, a specific clinical specialty, and the calculus shifts again, because now depth can beat scale. Metriport is not trying to be that slice. It is making itself useful to every slice at once.

The primary customer is a developer. The one problem solved incredibly well: access longitudinal medical records from every provider, instantly. The trojan horse: integrate underneath whatever the clinician already uses, learn the workflows from the inside, and become the data layer every health product is built on. And the dominant industry layer, the EHRs, is so entrenched that the UI war is not just hard, it is pointless.

The evidence it works: Sollis Health uses Metriport to pull prior EKGs, operative reports, imaging, and lab results in seconds, which keeps patients out of the emergency room. The infrastructure runs at 99.999% uptime and processes over 100 million requests a day. SOC 2, HITRUST, HIPAA. A YC company building the data layer for healthcare. Open source as a trust wedge with exactly the developer buyer it needs.

Metriport is building the engine of healthcare, not the screen. In a market the giants are entering, that is the right side of the table.

First Due: the UI checkmark

First Due is the fastest-growing end-to-end platform for Fire and EMS. It consolidates NERIS incident reporting, ePCR, fire prevention, pre-incident planning, scheduling, assets and inventory, hydrants, training, incident command, and mobile response into a single application with one login. Their own words: the last piece of software your agency needs to buy. More than 2,800 agencies use it, including Charlotte, Mesa, Seattle, and FDNY.

Now check it against the when-to-own-the-interface list.

The market is too small for a $100B+ company to ever care. Fire departments are thousands of fragmented agencies with mission-critical needs and small budgets. OpenAI is never building a fire records management system. The math is definitive. Even a great outcome in this market would be a rounding error on their P&L, which means the interface stays available for whoever earns it.

There was no dominant end-to-end vSaaS in the category. The legacy tools are siloed, disconnected, and universally complained about. That wobbliness is the opening, and First Due is consolidating the field with one modern, cloud-native platform that replaces the patchwork module by module. The fragmented incumbents each hold their little slice, and the other half of the market is still doing this on paper. Classic vertical software play, still working, exactly as described above.

And it is becoming, or already is, the market-leading end-to-end system. Once the fire chief logs into First Due for incident reporting, scheduling, training, and hydrant testing, that interface is the defensibility. The data lives there. The daily habit lives there. The AI layer gets added there, on top of a workflow the company already owns end to end.

First Due is building the screen on purpose. In a market the kings will never attack, the interface is not a battleground. It is a fortress.

Field Services: the interface is a red ocean, the engine is a blue ocean

Field services is the most visible vertical AI battlefield on earth right now. And that is exactly why you should think carefully before joining the fight for the login.

The market is enormous and proven. ServiceTitan went public with roughly $772 million in ARR and a market cap north of $6 billion. Beneath it: Housecall Pro for the small shops, Jobber for the middle, and a long tail of Workiz, Kickserv, FieldEdge, Service Fusion, mHelpDesk, FieldPulse, ServiceTrade, Fieldy. Above it, Salesforce Field Service and ServiceNow FSM push down from the enterprise. The software market alone is $6B+ and growing double digits.

And now the AI layer is piling on. Hundreds, probably thousands of AI upstarts, all with different logins, all pitching themselves as the AI-native operating system for the trades. Every single one trying to own the interface.

To be clear, I am not arguing the incumbents are unbeatable. I think they get beaten. ServiceTitan is a public company now, and public market pressure has a way of pulling category leaders toward financial engineering over innovation. The trades are exactly the kind of market where an AI-native upstart eventually takes the crown.

But here is the question nobody wants to answer: what are your odds of being that upstart?

If there are a thousand startups aiming at this interface, one or two win. Maybe. The other 998 get decimated. The category is enormous, the prize is enormous, and the competition is a massacre. That is not a thesis. That is a lottery ticket with extra steps.

Now flip it. Instead of fighting for the login, go headless. Hook into the model layer: build on Claude or GPT and become the vertical intelligence nobody else bothers to build. Or hook into the workflow layer: integrate into ServiceTitan, Housecall Pro, or Jobber and become the AI capability they cannot live without.

That is a genuinely blue ocean. Why? Because the people fighting the interface war have no time to build the engine. Because the incumbents, even the ones who will eventually lose, need AI depth right now and cannot build it all themselves. Because the horizontal platforms want vertical reach and cannot do it alone. Because the one or two winners who eventually take the UI will need the best engine available, and it does not have to be theirs.

And here is the beauty of the trojan horse. You integrate into whoever owns the screen today. You get the data. You learn the workflow. You become structurally important. And when the UI war finally shakes out, whether the winner is ServiceTitan, an upstart, or a horizontal king, you are already inside. You do not care who wins the surface, because you own the layer every winner has to build on.

Everyone is fighting to be login number one thousand and one. The smarter play is to be the engine inside whichever login wins.

Metriport surrendered the screen and became the engine of healthcare.

First Due is building the screen on purpose in a market the kings will never enter.

And in field services, thousands of startups are fighting to be the one login that survives, when the smarter play is to be the engine that every login, winner or loser, has to build on.

The interface is not always the throne. Sometimes the throne is underneath.

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

Linear #191.5: The 5 New Models That Replace SaaS with Yoni Rechtman (Partner @ Slow)

Wed, 26 Aug 2026 14:06:36 GMT

Sell the the finished project, not the hammer
(where the customer has to do it themselves).

SaaS was stupidly easy to model. Recurring revenue, 80% gross margins, land-and-expand, negative churn. Investors could underwrite it in their sleep. And that, says Yoni Rechtman, is exactly why it’s over.

Here’s the uncomfortable math underneath the whole conversation: a normal business spends 1–3% of its revenue on software. It spends the other 97% on work; labor, contractors, BPOs, services. For fifty years that imbalance held because software was small but hard to make, and being small was survivable.

Two things just changed at once. Software got easy to build. And software got able to do things, to execute the work instead of merely representing it. You can now sell the hammered nail instead of the hammer, and the market for hammered nails is 30 to 100 times bigger than the market for hammers.

In the episode, Yoni walks through the five models he thinks actually work going forward:

1. Service to differentiate software (FDE) You take responsibility for the outcome, not just the tool. “I’m going to change my software around your business process, and I’m going to guarantee that it’s going to do that — and I’ll create that guarantee through a level of service.” This decommodifies the value proposition and lets you sell expertise and “AI transformation” through a different line item, not just a web app. It’s the forward-deployed engineer (FDE) model.

2. Software to differentiate a service (neo-firms) The inverse: you’re running a service business, but software is what makes you faster, cheaper, and higher quality than every other provider of that service. You own the delivery and the liability — which is why Yoni says he’d rather underwrite one of these than a “post-trained model in a nice UI.”

3. AI rollups: Buying existing services businesses and using software + AI to compress their cost structure and standardize delivery — a “growth buyout” wrapper. This is the model Slow has written decks about, where the software isn’t the product; the service company is the product and AI is the margin expansion tool.

4. Agent networks Start in single-player mode — do a unit of work for money, get paid to acquire customers — then flip into multiplayer mode once you own a population. Phoebe (home care) is the example: it starts as labor orchestration filling shifts, and the network of caregivers becomes the asset it can resell to new demand. The end state Yoni calls a “for-profit union.” This one, he says, simply “wasn’t possible before agentic AI” — the 2019 “come for the tool, stay for the network” attempts failed because there was no platform shift underneath.

5. Hardware to differentiate software Software moated by a physical device — the hardware is the wedge that keeps competitors (who only have API keys) out. This is the Metropolis-style play he references: physical infrastructure that software then operates and monetizes.

The throughline on all five: each one escapes the old SaaS trap of selling a tool into the 1–3% of revenue companies spend on software, and instead sells or owns an outcome in the other 97% — the labor and services line items. As he puts it, you can sell the hammered nail instead of the hammer, and the market for nails is 30–100x bigger.

Vertical Software Summit

All information at the link below.

It’s truly the best few days of the year in vertical software and vertical AI.

This Weeks Vertical Titan:
Yoni Rechtman (Partner @ Slow)

Yoni Rechtman is a partner at Slow Ventures, a seed and pre-seed fund. Ask him to describe the firm and he’ll say “generalist,” but that undersells it. Slow’s actual filter is narrower and more interesting: they’re basically hunting for new business models and capital structure wrappers for software, which, as Yoni puts it, “often winds up leading us towards vertical software or vertical AI, in the broadest sense.”

The portfolio proves the thesis. Slow was a seed investor in Metropolis, which used acquisitions to build its way into parking and is now a $5 billion AI company. They backed Recurrence, maker of Phoebe, the home-care platform positioning itself as a three-sided agent network. They led Ando’s seed round in restaurant staffing. The throughline isn’t an industry — it’s structure. Every one of these companies is trying to escape the software business model and own an outcome instead.

What led him here is one observation with violent implications: no one has ever woken up and wanted to buy workflow software. People want invoices sent, claims filed, shifts filled. The only software businesses that ever got to sell outcomes — Visa selling transactions, Uber selling rides, Google selling clicks, Facebook selling customers; were network-effects businesses. As he puts it, “I am starting with a recognition of what most people and businesses in the world want to spend money on, which is not software.” Everything else in this conversation follows from that.

01. Sell the hammered nail, not the hammer.

The core mechanic behind the entire episode: software is priced against the 1–3% of revenue companies spend on software, but the value of an outcome lives in the labor line item. The nth AI receptionist for dental clinics isn’t ripping because it’s better software — it’s ripping because the buyer compares it to a human they’d otherwise pay, not to their practice management system. And here’s the underrated reason that’s an easy sale: businesses love firing contractors. “You pay 7% of your revenue to a medical billing firm — I’ll do the same work for 5%, better and faster.” That drops straight to the bottom line.

Action item: Recalculate your TAM (and your pitch) against the labor or services line item you replace, not the software line item you improve. Position yourself as the contractor replacement, not the software upgrade.

#02. The income statement gets worse. Take the trade anyway.

Yoni is brutally honest about the new model’s economics. Inference cost hits gross margin. Switching costs collapse when the user is an agent swapping API calls instead of a human learning a UI. You can’t amortize R&D as long because competitive pressure is relentless. “Along basically every line of the income statement kind of gets a little bit worse — but the markets get a lot better.” His prediction: software companies get much bigger, run lower margins, and produce far more absolute dollars of revenue and cash flow.

Action item: Stop underwriting your business — and stop letting investors underwrite it — on SaaS-style margin profiles. Model it against a labor-sized market and be upfront that the margin structure is different. It’s the on-prem-to-cloud trade, on a bigger magnitude.

#03. Service to differentiate software — FDE as a pricing unlock.

The first model in the framework. The reason forward-deployed engineers have come into vogue isn’t Palantir’s cultural shadow — it’s that FDE “is allowing people to make bigger promises and therefore charge more money.” Most industries are reticent buyers because they don’t trust the promises: sales says nice things, post-sales says “we’ll do a training.” That’s why ACVs cap out at $50K instead of $500K. When you say “I’ll change my software around your business process and guarantee it works,” you decommodify the value proposition — and you sell into a different line item entirely, because right now enterprises are buying “AI transformation,” expertise, and future-proofing, not web apps.

Action item: If your ACV is stuck because buyers don’t trust the software alone to deliver the outcome, add a service layer that guarantees it — and charge for the guarantee, not the features.

#04. Ownership is the dividing line between software and a service.

Yoni’s test for any AI company: are you responsible for doing the work, or not? There’s no blurry middle. “Harvey could not simply decide to be a law firm — because then it would be a law firm.” He’d rather underwrite a company that owns delivery, quality, and liability — even at a lower likelihood of success — than “a post-trained model presented to you in a nice UI.” His comparison: Crosby definitionally adds value above its inference relative to Heartbeat, because Crosby owns the outputs and the quality of its work.

Action item: Ask which side of the ownership line you’re on. If you’re on the UI side, remember the customer is buying the model’s capability — and the margin accrues to whoever owns the outcome.

Purpose-built embedded capital for vertical platforms.

#05. A BPO with tokens is a wedge, not a terminus — and the burden of proof is on you.

Yoni’s honest read: most vertical AI right now is “building BPOs with tokens instead of people.” That’s a great wedge and a murky end state, because someone can always come along and do the work cheaper, “either because their cost structure is better, or they’re just willing to tolerate lower revenues than you are.” His framing for founders and investors alike: “the burden of proof is on you for why there’s moat and terminal margin and terminal value, not on me to demonstrate that there isn’t.”

Action item: If your only moat is price, write the plan for what happens when the undercutting starts — self-improving systems that hit escape velocity, trust and brand, network effects, or distribution. “We do it cheaper” is not a plan.

#06. Everyone has the same API keys — a better prompt is a bet against model progress.

Yoni’s one piece of religion: “We all have access to the same API keys, so no matter what you do, you have to substantially differentiate tokens and code.” Better prompts are not a long-term differentiator. His exhibit A is Jasper, which was “selling a series of skills and prompts for copywriting” and got eaten the moment models absorbed the skill. Even “proprietary data” is fragile — the model labs are actively buying that data to fold into the weights. The PowerPoint corpora people called proprietary a few years ago? “The labs just bought it.”

Action item: Assume any differentiation expressed as a prompt, a thin fine-tune, or a dataset a lab could purchase has a short shelf life. Durable value has to live in workflow ownership, process, distribution, or network position.

#07. Get paid to acquire customers — the single-player to multiplayer motion.

The most novel model in the framework: agent networks. Start in single-player mode with a clear, legible value prop — doing a unit of work for money — and use that revenue and customer access to acquire a large population, then flip into multiplayer mode. The last generation of network businesses spent their way through the cold start problem (free pizza, subsidized rides, signup cash). The new generation gets paid to acquire customers. Yoni’s answer for why “come for the tool, stay for the network” failed in 2019: there was no platform shift underneath — “no profound emerging capability, just two separate kinds of capabilities put together.” Agents that can actually do the work are the shift.

Action item: Before you start, be specific about the second thing you’ll be able to do once you own a population of customers. If you don’t have a conviction about it, you’re just running a services business with extra steps.

#08. The for-profit union.

Slow’s portfolio company Recurrence (the Phoebe product) is the model in action. It starts as labor orchestration for home care — filling shifts for agencies. In doing so, it gets in front of all the caregivers, the entire supply side. The obvious next step is a talent agent: resell that supply to new sources of demand. The agency pays them to build the network; then the network becomes the business. Yoni calls the end state a for-profit union — “if you can serve a valuable universe of supply and control the attention of that valuable universe of supply, there’s a huge amount to do with it.” And one thing agents are genuinely good at: managing attention and negotiating.

Action item: Look for businesses where doing paid work plants you in front of a concentrated supply pool. The network asset you unlock is worth more than the work you’re billing for.

#09. What the customer thinks they’re buying is what you’re selling.

When Nic argued the service/software line is blurry and useless, Yoni pushed back hard: “It’s not unclear for the customers. Whatever they think they’re buying is what you’re selling.” If they think they bought software, you get priced and sold like software. If they think they bought a service, they expect ownership, liability, and someone to call. The customer’s belief dictates your promises, your channel, and how you charge — “all of those things are really dictated by what the customer thinks they’re buying.”

Action item: Audit every touchpoint — sales collateral, pricing page, onboarding — for what the customer would conclude they bought. Fix the mismatch between what you think you’re selling and what they’d say they bought.

#10. The compounding asset is the entire game.

Yoni’s test for whether any of this holds up: from customer 1 to customer 20, something has to get better. “You want to say every customer we get, the product is better and more valuable.” If every engagement is de novo — “every customer is its own snowflake and butterfly” — there’s no compounding asset and you don’t have a venture business; you have a high-end consulting shop. His most contrarian suggestion for where that asset can live: productize the process, not what flows through it. The first scaled consulting businesses — BCG, Bain, McKinsey — were selling a new technology of scientific management, not outputs. “Process and product are distinct until they’re not.”

Action item: For every new customer, ask what the engagement teaches you that makes the next one cheaper, better, or faster. If the honest answer is "nothing," you're not building an asset — and the process you use to deliver may be more productizable than the thing you hand over.

Sell the hammered nail, not the hammer.
Own the work, not the workflow.
Get paid to acquire customers.

Four things worth taping to the wall — whether you’re building the next vertical AI company or underwriting one:

The money is in the labor line item, not the software line item. Businesses spend 1–3% of revenue on software and the other 97% on work. Software can finally do that work, which means the addressable market just got 30 to 100 times bigger. Position against the contractor you replace, not the incumbent software you improve — businesses love firing contractors.
Ownership is the dividing line. Either you’re responsible for doing the work or you’re not, and the customer always knows which one they bought. Harvey can’t simply decide to be a law firm, because then it would be a law firm. A company that owns delivery, quality, and liability is worth more than a post-trained model in a nice UI — because the margin accrues to whoever owns the outcome.
The burden of proof is on you. Most vertical AI right now is a BPO with tokens instead of people. That’s a wedge, not a terminus — someone can always do the work cheaper. Assume there’s no moat, terminal margin, or terminal value until you prove otherwise. Everyone has the same API keys; a better prompt is a bet against model progress. Take the uglier income statement if it buys you a labor-sized market.
Networked verticals get built by people who get paid to build them. The last generation burned capital through the cold-start problem. This one inverts it: do a unit of work, earn the customers, then flip single-player into multiplayer. Recurrence’s home-care story is the template — get in front of all the caregivers, and the end state is a for-profit union. But it only compounds if every new customer makes the product better. If every engagement is de novo, you’re a consulting shop with a data-lake accessory.

If there’s a single through-line to steal from Yoni this week, it’s this: the best software companies of the next decade sell outcomes into the 97%, structure themselves as services where ownership is the moat, get paid to acquire the networks that make them defensible, and never mistake access to API keys for a business. That’s the compounding formula.

Big thanks to Yoni for opening up the Slow Ventures thinking to the rest of us. If you’re building or backing the software layer, the services layer, or the agent-network layer of a vertical, this episode is required listening. Forward it to a founder or an investor in vertical AI. If you’re building something interesting in the space, my inbox is open.

See you next Wednesday.
— Luke Sophinos

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

Linear #191: The 1000 Question Rule & How To Build The ChatGPT of Your Industry

Mon, 24 Aug 2026 00:24:06 GMT

Vertical-Specific GPT’s

A lot of vertical AI companies are not losing to incumbents.
They are losing to ChatGPT.

The founders who understand this are quietly building something different. They are not shipping features. They are shipping answers. Better, deeper, more specific answers than any horizontal model can produce, for the exact questions their customer asks every single day.

If your vertical AI product cannot out-answer GPT on the top 1000 questions your customer actually asks, you do not have a moat. You have a wrapper with a login screen.

This week is about how to actually build that moat, why the Vertical Software Summit is the only room where operators are talking about it honestly, and which vertical GPTs are already proving the model works at scale.

Prompting Your Way to an Epic Vertical AI Use Case

Most vertical AI products die at the prompt layer. The team ships a chat interface, wraps a horizontal model, adds a domain-specific system message, and calls it a product. Six months later the same customer asks the same question in ChatGPT and gets a comparable answer. The wrapper is exposed. Retention collapses. The team blames distribution.

Distribution is not the problem. The answer is the problem.

Here is what the operators actually winning in vertical AI are doing. This is the full playbook.

Step 1: Map the top 1000 questions in order of importance.

Not 10. Not 100. One thousand.

The number is not arbitrary. Below 100 you have a demo. Below 500 you have a feature. At 1000 you have a product. Above 1000 you start covering the long tail where retention actually lives.

How to build the list:

Pull two years of support tickets. Every ticket is a question in disguise.
Transcribe every sales call, onboarding call, and QBR from the last 12 months. Extract the questions the customer asked, not the ones your team answered.
Sit inside the customer for a full week. Not a demo day. A real operating week. Watch what they Google. Watch what they ask their coworker. Watch what they paste into ChatGPT when they think nobody is looking.
Scrape the industry subreddits, the private Slack groups, the closed Facebook groups, and the trade forums. That is where the real questions live.
Interview 25 customers with one prompt: what did you ask an expert this week that you wish software could answer.

Then rank them. Two axes. Frequency and economic weight. A question asked 40 times a week by a billing manager beats a question asked once a quarter by the CFO. Order the list. That order is your build sequence.

This list is not a document. It is your product roadmap. Everything downstream of this list is execution.

Step 2: Build a massively detailed prompt for each question.

Twenty pages is not a joke. It is the floor.

The best vertical prompts in production today are structured like operating manuals written by a senior domain expert who is also a systems thinker. Every one of them contains, at minimum:

Domain framework. The mental model an expert uses to approach this question. Not the answer. The way of thinking.
Vocabulary layer. The exact terms of art, the acronyms, the regional variations, and the terms that mean different things in different sub-verticals.
Regulatory context. What the customer is legally allowed to do with this answer. What disclaimers must appear. What jurisdictions apply.
Workflow context. Where in the customer’s day this question gets asked. What tool they are in when they ask it. What the answer needs to plug into next.
Output format. Text, table, filled form, diagram, deck, doc, cited citation list. Different questions want different artifacts.
Edge cases. The top 20 ways this question can be asked incorrectly, incompletely, or with hidden assumptions.
Graceful degradation. What the answer looks like when 30 percent of the required data is missing. What it looks like at 60 percent missing. What triggers a human handoff.
Source hierarchy. Which sources are authoritative, which are supporting, which are disqualified.
Anti-hallucination rules. The specific claim types this prompt is not allowed to make without a citation.
Tone and register. How a senior person in this vertical actually talks.

If your prompt fits on one screen, you have not done the work. You have written a system message.

Step 3: Test every model against every prompt.

GPT, Claude, Gemini, open source frontier models, image models, deck generators, doc generators, spreadsheet generators, form fillers. Do not have a favorite model. Have a favorite outcome.

For every question in the 1000, run the prompt through every viable model. Score the output on:

Factual accuracy against a domain-expert answer key.
Format fidelity to what the customer actually needs.
Latency at the point of use.
Cost per answer at scale.
Failure mode when the input is broken.

Some questions want Claude for reasoning. Some want GPT for structured output. Some want Gemini for long context. Some want an image model to produce an annotated diagram. Some want a deck generator to produce a client-ready output in 30 seconds. Match the model to the answer, not to the vendor relationship.

The customer does not care which model ran. The customer cares that the answer was right.

Step 4: Benchmark every answer against horizontal AI.

For each of the 1000 questions, paste the raw question into ChatGPT, Claude, and Gemini as a naive user would. Capture that answer. That is your baseline.

Your product must produce an answer that is 10x better than the baseline. Not 20 percent better. Not 2x better. 10x.

Ten times better means one or more of the following:

Cited where the baseline is not.
Structured for the workflow where the baseline is prose.
Compliant where the baseline is generic.
Current where the baseline is stale.
Personalized to the customer’s data where the baseline is universal.
Delivered inside the tool where the baseline requires a copy paste.

If you cannot articulate the 10x on a per-question basis, that question is not yet a moat. It is a coin flip. Coin flips lose to whoever has the better distribution, and horizontal AI has better distribution than you do.

Step 5: Build the master spreadsheet.

This is the artifact that separates operators who talk about vertical AI from operators who ship it.

Columns:

A: Question ID
B: The question in the customer’s actual words
C: Ranked importance (1 to 1000)
D: Frequency per customer per month
E: Economic weight (dollars touched by this question)
F: The full prompt (link to the 20 plus page doc)
G: All similar questions that route to this same prompt (typically 5 to 30 variants)
H: The winning model for this question
I: Output format
J: Latency target
K: Cost per answer
L: Horizontal AI baseline answer (link)
M: Your answer (link)
N: The specific 10x delta (one sentence)
O: Score against baseline (1 to 10)
P: Last tested date
Q: Owner
R: Retest cadence

That spreadsheet is your real product spec. The UI is downstream of it. The pricing is downstream of it. The GTM is downstream of it. The moat is inside it.

Step 6: Retest on a cadence.

The horizontal models get better every quarter. Your 10x lead compresses if you do not defend it. Set a mandatory retest cadence. The top 100 questions retest monthly. The next 400 retest quarterly. The bottom 500 retest twice a year. If any question drops below 10x against the current frontier, it moves back into the active build queue.

This is not a project. It is an operating discipline.

The uncomfortable truth: being in the workflow is not a moat anymore. Answering domain questions better than any horizontal model, on a per-question basis, benchmarked and defended, is the moat. If a horizontal model can match your answer, your customer will eventually notice, and your product will collapse into a login screen on top of someone else’s intelligence.

The work is boring. The 1000 questions are boring. The 20 page prompts are boring. The spreadsheet is boring. That is exactly why it compounds. Nobody else is willing to do it.

The Vertical Software Summit 2026

There is a reason we built this thing.

For a decade the vertical software conversation has been fragmented. Scattered X threads. A handful of newsletters. One or two tracks buried inside larger horizontal conferences. A lot of private group chats and off-the-record dinners. The category that quietly produces most of the durable software value in the world had no home.

The Vertical Software Summit is that home.

What makes the format work is what it refuses to be.

It is not a payments conference masquerading as a vertical software confernece.
It is not an investor conference not allowing other investors in.
It is the only conference by vertical software and ai founders for vertical software and ai founders.

The Summit is founders, operators, investors, and all the companies that support vertical businesses talking to each other, in one room, without the noise.

The lineup tells the story:

Founders who built systems of record in industries most tier one VCs would not touch. Pool cleaning software. Woodworking software. Pizza shop software. Laundromat software. The unglamorous categories that produce 40 percent EBITDA and 110 percent net revenue retention.
Acquirers & Investors who understand that a boring vertical with those numbers is worth more than a horizontal darling burning cash at scale.
Operators who have survived the SaaS Crash, the SVB bank crisis, the SaaSpocalaypse and are now walking through the software to AI transition in real time. This is the room where founders learn what the second transition actually costs, whats working, from the people currently paying for it.
AI-Native companies that are showing us how fast growth can look when you launch a net-new vertical AI product.

The programming is centered around:

War stories and hard lessons learned from the folks that have monopolized their verticals.
Playbooks we can all take into our own businesses that are already proving to work in adjacent industries.
CxO’s who can go deep in particular areas — AI-powered sales enablement, GTM, Customer Success, etc.

The bet underneath the event is simple. Vertical software is not a niche of the tech industry. It is the majority of durable software value creation. The market has priced this in at the incumbent level. It has not priced it in at the emerging company level. That gap is where the next decade of returns lives.

The people already in the room know this. The Summit is where they compare notes.

If you build, invest in, or acquire vertical software companies, this is the room you should be in. I hope to see you there! Grab your ticket, come enjoy Miami in November, and do it before tickets sell out!

The Vertical GPTs That Are Actually Winning

The fastest growing AI companies in the world right now are not horizontal chat products. They are vertical-specific GPTs that decided, early, to out-answer the horizontal models on a narrow, high-value domain. Every single one of them, whether they call it that or not, has been running some version of the 1000 question playbook.

Harvey did it for elite law firms first. The product is not a chat interface. The product is an internal framework that reflects how BigLaw associates actually draft, review, cite, and defend legal work. Horizontal models produce plausible legal text. Harvey produces text a senior partner will sign. That difference is worth every dollar the top firms pay for it, and it is why Harvey now sits inside a large share of the AmLaw 100. The moat is not the model. The moat is the encoded understanding of how legal work actually gets produced inside a firm, and the citation and review infrastructure that makes the output defensible.

MagicSchool did it for K-12 teachers. The horizontal models can technically write a lesson plan. What they cannot do is write a lesson plan that maps to a specific state standard, differentiates for three reading levels in the same classroom, aligns with the district’s approved curriculum, and outputs in the exact format the teacher’s LMS expects. MagicSchool encoded that specificity across dozens of daily teacher workflows and became one of the fastest adopted education products of the AI era, now used by millions of educators across tens of thousands of schools. Teachers do not stay because the model is better. They stay because every answer lands inside their actual workflow.

OpenEvidence did it for practicing physicians. Founder Daniel Nadler had already sold Kensho to S&P for around 700M. He did not need to build this. He built it because he saw that the highest frequency, highest stakes questions in American medicine were being answered either by memory, by outdated UpToDate articles, or by physicians pasting patient scenarios into ChatGPT under the desk. OpenEvidence mapped those questions. Cited every answer to peer-reviewed medical literature. Made it free at the point of use. Monetized, quietly, through embedded pharma advertising to a captive audience of prescribers. The result is one of the fastest ARR ramps in software history, roughly 150M in run rate, reached faster than almost any software company on record. A meaningful percentage of practicing US physicians now use it daily. The horizontal models cannot compete on this domain because they were never given the citation infrastructure, the regulatory awareness, or the workflow context that OpenEvidence encodes by default.

ChipAgents did it for semiconductor design. This is the vertical where horizontal AI fails most visibly. The vocabulary alone requires years of encoded context. RTL, verification, timing closure, DFT, tapeout. A general purpose model produces confident nonsense the moment you push past surface level questions. ChipAgents encoded the domain, integrated with the EDA toolchain, and gave semiconductor engineers an answer layer that actually understands what they are asking. The company raised 21M to keep expanding that answer surface, and its wedge tells you exactly where vertical AI has structural advantage: the more specialized the vocabulary and the more expensive the mistake, the wider the moat.

TrunkTools did it for commercial construction. Construction is the perfect vertical AI target and the perfect vertical AI graveyard. Every project generates thousands of pages of specifications, drawings, RFIs, submittals, and change orders. Every question a superintendent asks in the field has a definitive answer buried somewhere in that document set, and no horizontal model can find it because horizontal models were not trained on a specific project’s document universe, do not understand the specification hierarchy, and cannot cite the exact page a general contractor needs to defend a decision. TrunkTools built the answer layer on top of that document set. A super in the field asks a question. TrunkTools returns the exact spec section, the exact drawing detail, the exact RFI response, cited to the page. The 1000 questions in construction are things like: what is the required fire rating for this partition, what submittal governs this fixture, what is the latest revision of this drawing, which RFI answered this clash. Horizontal AI cannot touch any of that. TrunkTools can, and the general contractors and subcontractors adopting it are compounding a data advantage horizontal models will never replicate.

The pattern under every one of them is identical. They did not win because they had access to a better base model. They won because they took the boring, unglamorous work of mapping their customer’s real questions, building the answer infrastructure around them, and defending that infrastructure with citations, workflow integration, and continuous benchmarking against the horizontal frontier.

The horizontal models are impressive. The vertical products are indispensable.

That is the difference between a demo and a business, and it is the difference between an AI company that will still exist in 2030 and one that already quietly does not.

See you next Sunday.

Luke

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

Linear #190.5: What the leading tech underwriting platform can teach vertical founders with Sahill Poddar (Founder/CEO @ Parafin)

Wed, 19 Aug 2026 14:03:35 GMT

You’re running out of time!

Ticket prices go up September 1st, and may even run out by then. We are already 75% sold out so if you’re considering coming to the Vertical Software Summit this November, now is the time to grab your ticket!

All information at the link below.
It’s truly the best few days of the year in vertical software and vertical AI.

Underwrite the platform.
Own what nobody wrote down.

Most VCs in the country have a thesis on vertical software and vertical AI. Which translates to them essentially making educated bets. Sahil Poddar isn’t. As the co-founder and CEO of Parafin, he powers embedded capital inside some of the largest platforms on earth — DoorDash, Worldpay, Mindbody, Amazon, Gusto, and dozens more. That gives him one of the rarest seats in tech: he’s underwriting vertical software companies from the inside. He sees which platforms compound, which stall, which have real merchant lock-in, and which are one AI cycle away from being replaced.

This week’s conversation is the closest thing you’ll get to a live scan of the vertical SaaS ecosystem — what separates the great platforms from the median ones, why sequencing into financial services matters more than ambition, how AI is quietly rewriting underwriting and moats, and why “blue-collar” vertical software might be the most defensible category left standing. If you build, invest in, or sell into vertical software, this one’s essential reading.

Three things in this conversation should change how founders and investors work:

First, distribution — the thing that decides whether a vertical fintech line ever pencils: direct-to-SMB is a graveyard and embedded is the unlock. Marketing a loan selects for the borrowers you least want. Pushing a pre-approved offer through the platform a merchant already opens every morning selects for the ones you do — and hands you granular transaction data to underwrite with instead of a personal credit score.

Second, the separator between best-in-class vertical SaaS and the median: the great platforms act as collective bargaining agents for their merchants. They negotiate payments and capital on behalf of thousands of small businesses and repackage the result as a core service. The merchant data shows the payoff — a hard split between the one-to-two feature cohort and the six-plus feature cohort, with financing very often the stepping stone across.

Third, the moat question every vertical software board is asking wrong. If everything worth knowing about your industry is written in a document an AI can read, AI is coming for it. Legal, finance, accounting — exposed. Pizza shops, home services, anything running on knowledge that only exists in practice — defensible, and even more so once you move past pure software into capital, supply and physical goods.

This Weeks Vertical Titan:
Sahil Poddar (Co-Founder & CEO @ Parafin)

Sahil’s path is not the standard fintech founder arc. He grew up in a small-business family in India, then went deep on the sciences — bachelor’s, master’s, and eventually a PhD in particle physics. That work pulled him into data science, which pulled him into Meta, and then — somewhat serendipitously — into Robinhood, where he built and led major pieces of the growth, revenue, payments, and risk infrastructure. It’s also where he met the co-founders he’d eventually start Parafin with.

What Sahil saw at Robinhood was a renaissance in consumer finance. You could download five or six apps from the App Store, move money instantly, and get a best-in-class experience. But he noticed the sector driving ~40% of the U.S. economy — small and medium businesses — was quietly being left behind. Payments had been digitized by Square and Stripe. Credit had not. SMBs had exactly two ways to fund growth: retained earnings or the debt markets. Equity wasn’t a realistic option. And the debt markets, in Sahil’s words, were unfair.

That’s the wedge Parafin was built on. But Sahil’s insight wasn’t just “lend to SMBs.” It was the opposite of the conventional playbook. He looked at the graveyard of direct-to-SMB lenders — Kabbage, BlueVine, OnDeck — and concluded the whole model was structurally broken: marketing-led acquisition attracts adverse selection, you end up competing on ad creative, you have zero proprietary data to underwrite on, and you lead with none of the trust that lending requires.

So Parafin flipped it. Instead of going direct, they went through the platforms — embedding capital inside the software that SMBs already trust, already live inside, and that already has asymmetric transaction data on them. DoorDash knows exactly how much Nic’s restaurant sold last quarter. Mindbody knows exactly how a yoga studio is trending. That data becomes the underwriting engine, and the platform becomes the distribution engine. The result: zero CAC, higher conversion, lower loss rates, and enterprise churn that has stayed at effectively zero since day one.

Today Parafin runs three products — working capital, BNPL, and platform-embedded financing — has built its own transformer model (Paraformer) for time-series credit analysis, and just shipped a bank-transaction AI system that can produce a credit limit from a single Plaid token. Their BNPL line — starting with Gusto for payroll deferral — is growing 3x faster than anything they’ve ever launched. He’s building the underwriting layer for the entire vertical software economy. And because of the seat he sits in, he can see the pattern behind every great vertical SaaS company operating today.

Live Workshop Tomorrow @ 4pm EST.
How to Build An Epic Customer Conference

If you aren’t yet doing a customer conference you should. No matter your size.
I would add seven figures of ARR every time we did it and I’m bringing on the exact guy who organized mine to teach us all how to knock it out of the park. Hope you’ll join us!

01. Act as the collective bargaining agent for your customers.

Across Parafin’s entire portfolio of platform partners, the single strongest predictor of a great vertical software company is this: it aggressively negotiates with vendors on behalf of its customer base — payments, lending, insurance, supplies, marketing. The harder the platform bargains, the more valuable it becomes to the SMB.

This is what makes a vertical platform look and feel like a franchise. The median platform negotiates to maximize its own take rate. The best ones negotiate to maximize what their customers get, and take the compounding retention that follows. It is the Costco instinct applied to vertical software.

Action item: Founders: in your next vendor negotiation, write down what your customer gets before what you get. If that list is shorter, you’re monetizing your base instead of representing it. Investors: this instinct is the cleanest qualitative separator between best-in-class vertical software and the median.

#02. The wedge gets you in. The feature stack keeps you there.

Every great vertical software company gets its foot in the door with one aha moment — one workflow, one wedge, one feature. But durability comes from stacking features over time until you are the operating system for that business.

Sahil’s data shows two clusters of customers forming on every platform: those using one or two features, and those using six or more. The 6+ cluster is where nearly all the LTV lives, and churn there is a fraction of the rest of the base. Growth is not a funnel problem — it’s a feature-depth problem.

Action item: pull your own feature-count histogram this week and manage to the 6+ cluster, not to logo count. Investors: ask any vertical platform for that distribution. The shape of it is the growth model.

#03. Sequence financial services with discipline, not ambition.

Sahil is blunt: if a platform has less than roughly $1B in payments volume, layering in lending usually doesn’t make sense yet. Parafin routinely turns platforms down and tells them to go make the core product better first.

Payments, then lending, then deeper embedded services is a ladder, not a menu. Your platform has to be sticky and scaled before it can be a distribution channel for capital. Otherwise a financing line is a press release attached to a thin base — and one that consumes the team’s attention at exactly the wrong moment.

Action item: don’t bolt on financial products until your core has the volume and retention to make them a channel rather than a distraction. Investors: ask for the sequencing ladder, not the ambition slide.

#04. Financial products are the on-ramp to your highest-LTV cohort.

Once a customer takes capital through the platform, they adopt more products across the ecosystem — fast. Financing is one of the strongest gateway features in vertical software, not because of the spread, but because of what it does to the rest of the account.

The mental model shift is to stop treating embedded finance as a revenue line and start treating it as the fastest known path into the 6+ feature cluster. Priced that way, the decision changes: you optimize for adoption and trust, not take rate.

Action item: measure your financial products by downstream feature adoption and retention, not by their standalone margin. Investors: ask what happens to net revenue retention after a customer’s first capital event.

Purpose-built embedded capital for vertical platforms.

#05. Sell the MVP. Then delight into the expansion.

One of the sharpest lessons from Parafin’s early enterprise wins: always sell the smallest version of the product that still delivers real value. If the buyer will pay for the MVP, everything you build on top becomes delight instead of obligation.

Luke shared the inverse — an early $1M ACV deal where he oversold, promised the world, and spent years rebuilding trust. Overselling a whale is a debt that compounds at a punishing rate, and you pay it back in the exact quarters you needed for building.

Action item: strike every unbuilt feature from the enterprise deck. If the deal dies without them, you have a hypothesis with a contract attached, not a wedge. Investors: ask what shipped on the day of signature, not what’s on the roadmap.

#06. Enterprise sales works best when it’s cerebral, not aggressive.

Parafin has won some of the biggest logos on earth with what Sahil calls a ‘very cerebral, non-salesy’ motion: lead with data, lead with analytics, and be willing to say ‘you might not be ready for this yet.’

That intellectual honesty compounds. In enterprise, being the party willing to walk away wins more deals than being the party willing to promise anything. In a market saturated with AI-generated outreach, the scarce asset is a human who will tell a buyer the truth about their own numbers.

Action item: put one honest disqualifying sentence in your next enterprise pitch — the deals you lose to it were never going to activate anyway. Investors: signed logos are not the milestone in vertical software. Activation is.

#07. When your first whale asks you to jump, jump — but never go single-threaded.

Parafin’s first partner was a public DoorDash. On the deciding call, DoorDash asked how big the team was. Sahil answered, ‘Everyone’s on the call.’ Five people. They won by moving faster than any incumbent could.

In parallel they landed partners across other verticals so they were never dependent on one relationship. Balance founder-mode intensity with the discipline of a second and third thread — big partners go quiet for reasons that have nothing to do with you: a reorg, a roadmap freeze, a new CFO.

Action item: pull out every stop for the whale, and keep two other flagship accounts in flight so no single partner’s pause becomes your pause. Investors: concentration isn’t automatically a red flag — an idle roadmap caused by concentration is.

#08. Increase your surface area for luck.

Every early-stage vertical business has a cold-start problem. Sahil’s solution was volume of exposure: talk to as many prospective partners as humanly possible, show up with the most compelling product flow in the room, and give them zero reasons to say no.

Not a deck — a working flow. Absorb the feedback, tighten it, and run it again. Luck in a vertical market is mostly a function of pipeline density and the quality of what you put in front of people.

Action item: count your live prospect conversations this quarter. Under twenty and your problem isn’t conversion, it’s exposure. Investors: at inception stage, surface area and follow-through are the two signals you can actually underwrite.

#09. Your moat is what AI can’t learn from a document.

Sahil’s test is the sharpest moat question in vertical software right now: is your industry’s knowledge written down? If the expertise lives in manuals, statutes, and documentation, a model can absorb it and the software layer gets commoditized. If it lives in practice, relationships, and undocumented judgment, you have time.

Screen time is the second axis. Verticals where the operator lives in software all day are the most exposed to a generalized interface; verticals where the work happens off-screen are harder to reach. Score yourself honestly on both.

Action item: ask whether your value is institutionalized in documents or embedded in practice — if it’s written down, you need a layer beyond software. Investors: this is the best one-question screen for a vertical portfolio today.

#10. Insights are cheap. Actions are the product.

Parafin explicitly declined to build an ‘AI financial insights’ product. Their view: an insight without an action isn’t enough. What changes with modern AI is that the model can now take the action — tool calling, agentic payments, computer use — and that’s the moment there’s a real product instead of a dashboard.

Their filter for every AI idea is two gates: does it save the customer time or money by completing a task, and does the company have a genuine right to win in that category? Most ideas fail one or both. Meanwhile the highest-return AI work has been pointed at the core — underwriting, fraud, reconciliation, projections.

Action item: run every AI roadmap item through both gates and kill anything that clears only one; build the action, not the dashboard. Investors: ‘right to win’ is the question most vertical AI decks never answer.

Underwrite the platform, not just the merchant.
Do the task, not the dashboard. Own what nobody wrote down.

Four things worth taping to the wall — whether you’re operating a vertical platform or underwriting one:

1. Distribution through the workflow owner changes the business. Founders get a better borrower pool, better conversion, better loss rates and data no direct competitor can see. Investors should treat that as a different business model, not a cheaper CAC line.

2. Embedding means data flows in. If the user has to leave, apply cold and maybe get rejected, you built a worse version of direct inside someone else’s product. It’s the fastest way to tell a real integration from a partnership press release.

3. AI has to complete the task, and you need a right to win it. Insight is table stakes and worth nothing on its own. Most vertical AI roadmaps clear the first gate and quietly fail the second.

4. The moat test is whether the knowledge is written down. Documented industries are exposed. Practice-based, physical, off-screen industries are defensible — and get more defensible the further past pure software you go. Score every vertical in your portfolio or your roadmap on that single axis.

If there’s a single through-line to steal from Sahil this week, it’s this: the best vertical software companies act like franchises, negotiate like collectives, sequence financial services with discipline, and build moats in the parts of the business AI can’t read from a document. That’s the compounding formula.

Big thanks to Sahil for opening up the Parafin learnings to the rest of us. If you’re building or backing the workflow layer, the capital layer, or the AI layer of a vertical, this episode is required listening. Forward it to a founder or an investor in vertical software — and if you’re building something interesting in the space, my inbox is open.

See you next Wednesday.
— Luke Sophinos

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

Linear #190: The Branding Move Nobody Talks About in Vertical Software

Mon, 17 Aug 2026 00:01:30 GMT

Business-in-a-box / digital franchise

A fast primer on the model that makes the branding move possible.

Business-in-a-box vertical software is what happens when a founder stops asking “what workflow can I digitize” and starts asking “what does this operator actually need to run their entire business.” The answer is almost never just software. It is software plus services plus supplies plus payments plus marketing, delivered as one relationship.

Think of it as a digital franchise. All the operational infrastructure and scale advantages of a franchise, without the royalties, the brand handover, or the loss of independence. The operator keeps their name, their storefront, and their profits. The platform provides everything else.

Slice does this for independent pizzerias. Moxie does this for solo medspas. Odeko does this for coffee shops. Cents does this for laundromats. Each one earned the account with software, then followed the operator’s real spend into every non-software line they could serve.

That is the model.

Now here is the part almost nobody talks about…

You’re running out of time for the 2026 Vertical Software Summit. Tickets are close to 80% sold out! We’ll have 350+ vertical founders/operators/investors in Miami in November. If you need a free ticket let us know and we will get you one!

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

Once you own the whole stack, you get to change who you compete against.

Slice is the clearest example in vertical software history, and almost no one credits them for it. They do not position against Toast. They do not position against Square. They do not position against Silicon Valley AI. They position against Domino’s and Pizza Hut.

Read that again.

Their entire market story is “we give independent pizzerias everything the chains have, without making you sell your soul, your brand, your storefront, or half your profits to a franchisor.” That single framing does more work than any feature list ever could.

Look at what that positioning unlocks:

I. It reframes the buyer’s mental model
An independent pizza shop owner does not wake up shopping for POS software. They wake up watching a Domino’s open two blocks away and wondering how to survive. Slice walks in and says we are your answer to that, not another vendor.

II. It reframes the competitive set
When you compete against a franchise, features stop mattering. What matters is independence, brand ownership, margin retention, and dignity. Those are not things a POS competitor can counter-position against. They are structural.

III. It reframes pricing
If the alternative is handing over royalties plus marketing fees plus giving up your name, a monthly software fee plus supply attachment looks like a gift. You are no longer priced against SaaS. You are priced against franchise economics.

IV. It reframes retention
Customers do not churn from the thing keeping them independent. They churn from vendors. Slice is not a vendor. Slice is the reason they still own their business.

This is the branding move that turns a vertical SaaS company into a movement. And it only works because they own the full stack. If Slice were just software, the Domino’s comparison would sound absurd. Because they own software, supplies, ordering, marketing, and phone, the comparison is real.

Build the box, then use the box to move yourself into a category of one.

How to pull off the branding move in your own vertical

Most founders never attempt it because they were trained to position against other software. Wrong instinct.

Step one — find the villain your customers already fear

Not the software competitor. The real-world threat. For pizzerias it was Domino’s. For solo medspas it is corporate chains and franchise aesthetics groups. For independent coffee shops it is Starbucks. For laundromats it is national laundry brands and gig-delivery platforms. Ask ten customers what keeps them up at night. The villain is in that answer, not in your competitive matrix.

Step two — earn the right to make the comparison

You cannot say “we are your answer to Domino’s” if all you sell is a scheduling app. The claim has to be backed by the stack. Supplies, marketing, ordering, payments, phone. The more of the operating surface you own, the more credible the comparison becomes. This is why the branding move requires the business-in-a-box underneath it. One without the other collapses.

Step three — write the story in franchise terms, not software terms

Do not talk about features. Talk about what a franchise gives an operator and how you deliver the same thing without the handover. Scale. Brand support. Marketing power. Operational leverage. Then close with what you do not take. Their name. Their profits. Their independence.

Step four — price against the villain, not against SaaS

The moment you anchor pricing to franchise economics instead of software economics, your ACV ceiling changes. A pizza shop paying 8 percent of revenue to a franchisor will happily pay you a fraction of that for the same operational lift. You are not expensive software. You are cheap independence.

Step five — make the story public

The branding move is not an internal deck. It is the homepage. It is the sales pitch. It is the founder’s every podcast appearance. Slice does not hide the Domino’s framing. They lead with it. That repetition compounds into category ownership.

The shortlist, and how each one is approaching the move

The best way to see how this move works in practice is to look at operators who are already inside the businesses they serve. Not all of them have completed the branding move. Some are still building the box. Others have the box but have not yet named the villain with conviction. That gap is what creates the opportunity.

Each company below earned its first customer with software. Then it followed the money. Supplies, payments, marketing, ordering, logistics — whatever the operator was already spending on and doing badly. The question now is whether they complete the final step and tell the market, clearly, who they are really against.

The scorecard tracks two things. Stack depth: how much of the operating surface the company owns. Branding move: how explicitly they have named the villain and reframed the competitive set. The first gets you to scale. The second gets you to a category of one.

Slice

The reference case for the entire category. Software for independent pizzerias, plus boxes, cups, napkins, ordering, marketing, and phone answering. Positions explicitly against Domino’s and Pizza Hut. The message is not “better POS.” The message is “keep your shop, beat the chains.” No other vertical SaaS company has executed the branding move this cleanly.

We’re running a 60 minute webinar this Thursday at 4pm EST on…

How to Plan/Run an Epic Customer Conference

If you haven’t done one yet, they are a HUGE revenue unlock, no matter your stage…

Moxie

Medspa in a box. EHR, POS, scheduling, compliance tracking, discounted supplies, and business coaching. The branding move is emerging. They position against the loss of independence that comes with joining a corporate medspa group or franchise. The message is “own your practice, not someone else’s brand.”

Odeko

Coffee shop software plus supply delivery. The natural villain is Starbucks. The branding move is still underdeveloped here, and it is the single biggest storytelling opportunity left on the table for them.

Cents

Laundromat software running front-of-house, POS, pickup and delivery, and the operational layer for independent operators. The villain set is national laundry chains and gig-delivery platforms eating into local demand. The branding move is quietly forming.

The pattern across every one of these companies is the same.

They earned the account with software, then followed the operator’s real spend into every non-software line they could serve. The ones that will compound the most from here are the ones that also name the villain and take the branding move seriously.

Slice figured out that the software was never the product. The product was independence. Everything else — boxes, ordering, marketing, phone — was just the delivery mechanism for that promise.

That is the lesson worth stealing.

The takeaway

The wedge. In an AI age, the software is the wedge. It gets you in the door and earns the trust that everything else is built on.

The company. The business around the software is the company. Supplies, payments, marketing, ordering, and services are the real operating surface that creates the relationship.

The moat. The villain you name is the moat. When you change who you compete against, you move outside the feature-versus-feature death spiral and into a category of one.

Build the whole box. Then tell your customers exactly who it is really for, and exactly who it is really against.

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

Linear #189: The AI Roll Up Market Map, How It Pencils for VC's, The DNA That Wins

Mon, 10 Aug 2026 00:02:40 GMT

AI Roll Up Market Map: From Twitter to $4B & The Money Behind It

Twenty-four months ago the category did not have a name. Slow Ventures called them GBOs. A few operators were quietly rolling up service businesses and layering software on top. Matt Brown had a taxonomy. I was writing about M&A+SaaS and then Venture Buy Outs. Nobody agreed on anything. It was a Twitter thread and a Substack post.

A few of the operators worth knowing:

Want to see the full landscape? ai-rollup.fyi is tracking it and it is the best active directory going.

Then the platform funds moved.

General Catalyst built Creation, a $1.5B strategy dedicated to AI-enabled roll-ups. Ten publicly disclosed portfolio companies. Long Lake in HOA management. Eudia in legal. Titan in IT services. Crescendo in contact centers. Dwelly in UK real estate. The most systematic version of the play by a mile.

Thrive Holdings launched a $1B vehicle in April 2025 and by December had convinced OpenAI to take equity and embed engineers inside portfolio companies. Current (formerly Crete) in accounting. Savvy Wealth in advisory. Shield in IT MSP. Fewer bets, deeper integration.

8VC and Slow Ventures run leaner. 8VC backed Metropolis, seeded Sequence Holdings as a cross-industry holdco, and moved into stealth healthcare. Slow pioneered the low-upfront GBO model years before the category had a name.

a16z, Khosla, Bessemer, GV, Elad Gil, Felicis, Rockbridge, Bain all layered in over the last twelve months. Gil personally backed Long Lake and Special. Felicis led Adaptive Innovations in home health. Rockbridge co-backed AGI with Atomic. Bain is running a payer-focused merger in healthcare AI.

The pattern is simple. Twenty-four months ago this was six firms and a Substack. Today it is every top-tier venture firm plus most of the mid-market PE world plus a growing set of family offices and sovereigns.

The traditional PE-vs-VC boundary stopped mattering in this category. Both sides see the same thesis: labor-heavy services businesses, mispriced by traditional PE, unreachable by traditional SaaS, and transformable by AI-native operators willing to own the workflow instead of selling to it.

Two years ago this was a tweet.

Crazy how fast all of this moved…

If you’ve been reading the newsletter for a while and getting value out of it,
will you do me a really quick favor? A <5 minute favor to be exact…

I’m launching The State of Vertical AI: 2026 H2 Benchmark Survey.

It will be the definitive guide on how vertical software and vertical AI companies are actually building, monetizing, and scaling AI -- specifically the pre/post AI performance cuts nobody else is publishing.

Survey is 100% anonymous, takes <5 minutes, and every founder who fills it out gets the
full report back and a free ticket to the Vertical Software Summit in November!

Click the image below to access it :-)

A sincere thank you in advance!

How an AI Roll-Up Actually Pencils as a Venture Investment

The instinct most VCs have when they hear “buy a services business” is to reach for the PE calculator. Buy at 4x EBITDA. Improve operations. Sell at 8x. Fine returns. Not venture returns.

That is not the underwrite. Here is the underwrite.

Start with the acquisition math. A $10M revenue services business at 15% EBITDA margins throws off $1.5M in cash. In accounting, pest control, or HVAC, that business trades at 3-4x EBITDA in the middle market. Call it $5M enterprise value.

Now layer the AI operating platform. Real deployments in accounting and IT services are absorbing 30-70% of labor tasks. Labor is 55-65% of the cost structure. If you compress even half of that headcount cost into automation, EBITDA moves from 15% to 35-40%. That $10M in revenue is now throwing off $3.5M-$4M, not $1.5M.

You did not fire the humans. You let each of them carry 2-3x the client load, which solves the chronic labor shortage that was capping revenue growth in the first place. Revenue grows into the capacity. Now you have $15M in revenue at 40% margins. $6M in EBITDA on a business you bought for $5M.

Then repeat. Tuck in five more platforms of the same size at the same multiples. You now own a $60M revenue, $24M EBITDA operating business. Total equity deployed: roughly $30M plus some acquisition debt.

Now the part everyone misses.

Once you have run the AI platform across six acquired businesses, you no longer just have a services operator. You have a productized system. The workflows are standardized. The models are trained on real proprietary data. The compliance posture is defensible because you actually operate inside the regulated category.

That is when you launch the lightweight B2B product.

Take legal as the cleanest example. You roll up six mid-market law firms. You massively improve their throughput with an internal AI platform. Then you launch a B2B legal AI assistant to every business in America, priced at SaaS-like ACV, backed by the fact that your own licensed attorneys sit behind it for anything that requires actual practice of law.

Every horizontal legal AI startup gets stuck at the same wall: they cannot practice law, so they cannot close the loop. You can. Because you own the firms. The moat is not the model. The moat is the license, the data, and the operational spine underneath the software.

Now your $60M services business has a second revenue line: a $10M ARR B2B product growing at 200% a year, sold into a market ten times larger than the acquired firms could ever serve directly.

Do the same in accounting. Roll up CPAs, then launch an embedded AI accounting product for every SMB in America, backstopped by your own licensed accountants. Same play in insurance: roll up brokers, launch an AI insurance advisor with a real E&O policy and real licenses behind it. Same in home services: consolidate HVAC operators, then license the dispatch and diagnostics platform to independent shops nationwide.

The services business funds itself. The productized SaaS layer is the venture outcome.

The re-rate is where venture math actually shows up. Traditional services businesses trade at 6-8x EBITDA at exit. That gets you a $150M-$200M outcome. Fine. Not venture.

But AI-native operators that have crossed into productized software are being valued on a blended framework. The services revenue at services multiples. The B2B product at ARR multiples. Metropolis was valued at $5B+ before the SP+ deal closed. Long Lake reportedly cleared $100M in EBITDA and is being marked at software-like multiples. Crescendo is charging per resolved ticket, which is functionally software pricing dressed in a services wrapper.

Uber is a decent analogy. Uber was not valued like a taxi medallion company. It was valued as a technology platform that happened to own the workflow of moving humans through cities. The medallion comps said $2B. The tech comps said $60B.

Same trick. Different decade.

An AI roll-up hits venture math when three things converge: proprietary technology deep enough to defend the margin gains, a productized second act that reaches beyond the acquired footprint, and a narrative arc credible enough that the exit market prices the business as tech instead of services.

Miss any of the three and you have PE. Get all three and you have Metropolis.

How to Spot the AI Roll-Ups Destined for Greatness

Three filters. In this order.

One. The team is a triangle, not a solo act.

Every AI roll up worth backing has three roles staffed on day one. Miss any leg of the triangle and the company dies in a specific, predictable way.

The Forward Deployed Engineer builds the AI operating layer. Not fine tunes prompts. Not integrates SaaS. Actually builds proprietary systems that absorb workflow. This is the Palantir DNA the category runs on.

The PE/M&A operator underwrites deals, structures debt, and integrates acquisitions without blowing up culture. Nine out of ten founder led roll ups die here. They can build the tech and cannot close a deal, or they close deals and cannot integrate. Renovo Home Partners went bankrupt on exactly this failure mode.

The industry expert is the operator who ran a business in this vertical for 15 years and knows why the ugly workflow is actually ugly. Without this leg, the AI you build solves the wrong problem beautifully.

If the team is two engineers and a banker, pass. If the team is three operators without a builder, pass. The triangle is the moat before the moat.

Two. The margin move already happened. On real customers. Before the raise.

The signal that separates real from theater is whether the AI platform has actually moved EBITDA in a live business.

Not a pilot. Not a case study. Live P&L, live customers, and a margin curve you can point to on a chart.

Long Lake did this. Adaptive is doing it in home health. Metropolis proved it at SP+ within four quarters. The founders who cannot show the margin move on their own first acquisition are asking you to underwrite the thesis. The founders who can are asking you to underwrite the scale.

Those are radically different risks. Only fund the second one.

If you made it thus far, go check out the 2026 Vertical Software Summit. We’ll have 350+ vertical founders/operators/investors in Miami in November. Two days,

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

Three. The vision is big enough to flip the multiple.

This is the least obvious filter and the most important.

Every AI roll up eventually faces one question at exit: does this get priced as a services business or as a technology platform? A 6x EBITDA multiple or a 10x revenue multiple? The gap between those two outcomes is 5 to 10x on your return.

The vision has to be big enough to bend the multiple. That means a category large enough that the endgame is category ownership, not category consolidation. A productized second act that reaches beyond the acquired footprint. A founder who talks about the business the way Kalanick talked about Uber, not the way a PE operator talks about a portco.

Ask the founder what the business looks like in ten years. If the answer sounds like “we will be the biggest player in dental billing,” that is a nice PE outcome. If the answer sounds like “we will roll up the firms, productize the platform, and sell the AI layer to every business in America,” you might have a re rate on your hands.

Both can make money. Only one makes venture money.

Two years ago this was a tweet. Today it is a category. Two years from now half the entrants will be gone and the survivors will be very large businesses.

Pick carefully.

Until next Sunday.

Luke

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

Linear #188.5: Something is off in venture right now with Rick Zullo (GP @ Equal Ventures)

Wed, 05 Aug 2026 14:34:29 GMT

Before we dive in today, we have a quick ask!

Fill out our 100% anonymous annual benchmark survey and get a FREE ticket to the Vertical Software Summit this November AND access to the report when it drops.

It will help all of us understand what is working in verticals, specifically around AI.

Fill Out Survey Here

A quiet, structural rot….

Something is off in venture capital right now, and most people don’t want to say it out loud. Not in an obvious “bubble about to pop” way. More like a quiet structural rot; revenue multiples being applied to services businesses, GMV being called ARR, $100M valuations for companies that could be disrupted by two tokens from Anthropic, and a whole lot of capital looking for a home in anything that feels like a moat.

Meanwhile, at the early stage, according to Rick Zullo, there’s arguably more real opportunity than there’s ever been. Capital-efficient vertical companies are compounding quietly while the zeitgeist chases trillion-dollar outcomes.

This week, Nic sits down for a conversation with Rick Zullo, founder of Equal Ventures, that gets at this disconnect better than anything I’ve heard in a while. Rick runs a seed fund focused on four unsexy-but-enormous markets — insurance, supply chain, retail, and energy — and last year, somewhere between a third and half of his Fund 1 portfolio was free cash flow positive while still growing 100%+. Nobody’s putting that on a billboard. They should be.

This Weeks Vertical Titan:
Rick Zullo, Founder/GP of Equal Ventures

Rick’s path into venture is a classic operator-of-the-craft story. Consulting and private equity first, then an MBA at Columbia, where one of his early internships was at Bowery Capital — working for Nic Poulos, who interviewed him on this episode. Fourteen years later, he’s spent his entire career investing across vertical software, vertical marketplaces, and vertical services.

Eight years ago he started Equal Ventures. Today it’s a couple hundred million in committed capital, a few funds in, laser-focused on seed-stage investing in four markets: energy, insurance, retail, and supply chain.

What makes Rick interesting isn’t the fund. It’s more of his lens.

He describes himself as a value investor in a growth asset class — but he’s quick to clarify he’s not a Ben Graham “cigar butt” guy. He’s a Charlie Munger and Michael Mauboussin disciple. The translation: he’ll happily ride the J-curve and hemorrhage cash with a company for years, but only if the market structure and competitive dynamics suggest it can eventually build a durable moat and become a free cash flow monster. He spends more time thinking about moats before a company has shipped a line of code than most investors spend thinking about them at the Series B.

And that lens is exactly why he’s uncomfortable right now. Software used to have built-in defensibility — expensive to build, slow to ship, hard to distribute. Today, product has no moat, and a huge chunk of the market is investing in services businesses using software-era math. Rick thinks that ends badly for a lot of people. He’s not vague about it, and he’s usually armed with specifics.

He also has the receipts to back up the contrarian stance: a portfolio of “quiet compounders” throwing off real cash while growing triple digits, in a market that has decided those companies don’t matter because they can’t absorb a billion dollars.

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!

01. Underwrite to monopolization potential, not product pull

Revenue multiple investing in services has never worked. Services companies have cost structures, and even if you automate everything away, the long-term question isn’t gross margin — it’s net margin. And you can’t know net margin unless you understand the moat structure of the industry.

The Thiel test still applies: can you monopolize this sector? Rick’s blunt take — most service opportunities have zero chance of monopolizing their sector. Early product pull and temporary margins are not a business. If you’re at $10M of services revenue with great gross margins, that doesn’t matter candidly. What matters is whether the position becomes defensible at scale.

#02. Know whether you’re playing a greenfield or brownfield game

Software investing was largely greenfield — the 10x rule applied because you had to motivate a customer to adopt something de novo. Services investing is brownfield: you’re stealing share from incumbents, not creating a category. Brownfield means multi-homing customers, low switching costs, everyone driving down the same cost curve — and incumbents who are implementing AI like crazy too. Don’t underwrite a world where competitors don’t respond.

#03. Watch for “beta-driven product-market fit”

This was my favorite concept from the episode. When every company in a category starts working at the same time because of a technology or market catalyst — that’s not PMF, that’s beta. We saw it in supply chain in 2021. We’re seeing it in AI-native services now.

Corollary: if it feels too easy, it’s probably not valuable. And there’s a real case for being the fast follower, not the pioneer, in categories where everyone is subsidizing the customer at once.

#04. Treat early margin advantage as a wedge, not the business

Rick has a portfolio company that went 0 → $10M with 50% EBITDA margins in year one. His reaction? Great — now everything we talk about is how to convert that position into critical mass for the next stage of the business. The margin is the key to the door, not the house.

The Amazon lesson: Amazon wasn’t a bookseller. The wedge became a marketplace, then a platform, then a thousand other things. Right now, too many AI services and physical AI companies have the wedge but no second act — and that needs to be the plan from day one, not a bank shot later.

Purpose-built embedded capital for vertical platforms.

#05. Underwrite to the cost line that actually matters

In insurance, OpEx barely matters. The loss ratio is the business. Most AI-carrier startups have no clue what their loss ratio is until the book matures — and many are trying to outrun their book with growth. We watched this movie in 2021. Everyone got washed out.

Same pattern in mining: everyone wants to use AI to drill more surgically, but the core competencies that win are distribution and cost of capital — which the incumbents already own. And watch the accounting games: companies loading up on assets and putting the cost of capital below the line so gross margin looks pretty. That cost is real whether you report it or not.

#06. Reconcile every private company against public comps

WeWork had great revenue growth and positive margins — and horrific unit economics versus any legacy REIT. Rick applies the same test everywhere: profit per employee, revenue per employee, the specific ratios of the industry, against the incumbents. The 2021 EV and physical infrastructure cohort now trades at 2–5% of their IPO values. Public markets don’t care about your revenue multiple. They eventually reconcile everything.

#07. Physical AI moats are real — but only for cornered resources

The “data gusher” thesis works: if you own the cameras in the warehouse, the hardware on the ships, the sensors on the grid — those are scarce, cornered resources. How many ocean carriers are there? How many will work with multiple providers? Own the physical layer and competition gets structurally hard.

But recognize why the market loves these deals right now: they’re massively capital intensive, and the market is flooded with capital that needs a home. Capital intensity is a feature for mega-funds, not necessarily a moat for the company.

#08. Veeva couldn’t get funded today — and that’s the problem

Veeva raised $9M before going public and has been printing cash ever since. In today’s market, nobody wants that company because you can’t deploy billions into it.

The kingmaking era is real: run the math on 3-and-30 on a $10B fund and you understand why every good idea suddenly “requires” a billion dollars to pursue. The biggest funds look more like Blackstone than Benchmark — Rick’s comparison, and it’s accurate. Founders Fund has backed 600 companies where 6 really matter, with an average dollar-cost-averaged entry well above $1B. That’s a crossover/public-investor playbook (the Sands Capital model from the Amazon/Salesforce era), not seed venture. Neither is wrong. But they are not the same game.

#09. Founders: pick your investor by the outcome math, not the brand

One of Equal’s LPs uses a “dragon number” — take your average ownership and calculate the exit size needed to return the fund. Equal’s target: roughly a $570M outcome. A billion-dollar exit 2–3x’s their fund.

A mega-fund needs a $100B outcome. A billion-dollar exit doesn’t move their needle — which means your billion-dollar outcome doesn’t matter to them. If you want to build a capital-efficient company, investor alignment is existential. The wrong cap table closes doors.

#10. The quiet-compounder playbook still works

Equal is content to do a seed, maybe a small Series A, own 15–20%, and never have the company raise again. Invest at a $7M pre, wake up to $150M of top line and $20–30M of free cash flow. Their LPs — endowments, foundations, fund of funds who’ve seen the game — know compounding eventually wins.

In a market where hope is being passed off as strategy (”this could be a trillion-dollar outcome!”), picking your playbook and living and dying by it is the actual edge.

Hope is not a strategy.

The close

Here is what I keep coming back to. Everything working at once is not proof that the model is sound. It is the condition under which nobody has to check. Beta is generous, capital is cheap, gross margin is flattering, and the marks are all unrealized. In that environment discipline looks like underperformance, right up until the tide goes out and it looks like the only thing that mattered.

You do not need to call the top. You need to know which game you are playing and then actually play it. If you are underwriting to monopolization potential, say so, and eat the volatility. If you are building a quiet compounder, stop apologizing for the size of your market and go own it. The failure mode is not picking the wrong playbook — it is running one playbook and grading yourself on the other one’s scoreboard.

Beta lifts everyone. Structure decides who is still standing when it stops.

So pick your playbook, and live and die by it. Do the homework nobody wants to do. Be the nerd reading the book instead of sitting on the panel. In vertical software and vertical AI, the boring question is almost always the right one, and right now it is the only question with an edge left in it.

See you next Wednesday. If this one landed, forward it to a friend!
— Luke Sophinos

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

Linear #188: Open Source vs. Closed: Why A Bunch Of Us Are Renting a Ferrari For A Trip To The Grocery Store

Mon, 03 Aug 2026 00:03:09 GMT

Alright, let’s get to it…

I think most vertical founders are thinking about open source incorrectly.

They treat it as an identity, a movement, or a marketing narrative. It is none of those things. It is a build lever. And the founders I love are the ones who know when to lean on open weights, open frameworks, and open cores, and when to quietly route the hard parts through closed models they pay real money for.

The interesting question is not open versus closed.

It is how you compose them.

And if you’re using one or two closed source models for EVERY ITEM, you’re probably renting a ferrari for some trips to the grocery store…

How to think about deploying open source as a vSaaS founder without losing the plot.

Route cheap jobs to cheap models, hard jobs to frontier models

Start with a simple map. Draw every AI call in your product. For each one, write down three numbers. Cost per call. Volume per month. Consequence of a wrong answer.

That table tells you a lot about where open source might belong.

High volume, low consequence calls are usually open source candidates. These are the calls that quietly eat gross margin if you leave them on frontier APIs.

To simplify it, with these types of calls, you’re renting a Ferarri for a trip to the grocery store…

Moving them to open weights, hosted somewhere you control and possibly fine tuned on your vertical data, tends to pay for itself quickly at scale.

Low volume, high consequence calls are usually worth keeping closed. The final draft. The summary the user actually reads. The recommendation the operator acts on. The frontier tax feels worth paying here, because the quality delta is real and the volume is small.

Now the part I think most founders miss. Tie the two together with a routing layer you own.

Do not let the model choice leak into your product code. Every AI call in your app should ideally hit an internal endpoint, not OpenAI or Anthropic directly. That endpoint decides which model runs, based on the task, the customer tier, the latency budget, and the current cost profile. When a new open model ships that beats your current one on your evals, you swap it behind the endpoint and your product does not change.

This might be the highest leverage architectural decision in a vertical AI company right now. Founders who build it early tend to migrate work from closed to open as open catches up, and they compound margin every quarter. Founders who hardcode a single provider are usually one price hike or deprecation away from a bad quarter.

A few operator notes worth considering…

Build your own evals before you optimize your stack. You probably cannot decide what to route where if you do not know what good looks like on your workflow. Vertical evals may be more of a moat than the models themselves.

Consider open sourcing your commodity layer. The vertical data model, the workflow logic, and the customer specific fine tunes usually belong closed. Giving away the schema rarely pays off.

Do not fine tune too early. Prompting plus retrieval tends to beat a bad fine tune. Fine tuning makes more sense once you have volume, stable evals, and a clear cost or quality reason.

And in most verticals, your customers do not care which model you use. They care that it works, that it is fast, and that someone owns the outcome when it does not.

If you’ve been reading the newsletter for a while and getting value out of it,
will you do me a really quick favor? A <5 minute favor to be exact…

I’m launching The State of Vertical AI: 2026 H2 Benchmark Survey.

Click the image below to access it :-)

A sincere thank you in advance!

OpenSource: A Real Look At What It Can Do To Your Margin

Let me start with an observation.

Most of you are using 100% closed source models today. Gemini, OpenAI, Anthropic. You are wrapping their APIs, charging 10 to 20% on top of whatever they charge you, and calling it a vertical AI product. That is not a business. That is a passthrough with a login screen.

The uplift math is worth walking through, because I do not think most founders have actually done it.

Assume open source models can serve a meaningful chunk of your workload at roughly 10% of the cost of frontier closed models. That number moves around depending on the task, but for high volume interior work like extraction, classification, embeddings, and structured parsing, 10% is a reasonable directional number. Your customers are already paying you a price they are comfortable with. You do not have to lower it. You just have to change what sits underneath.

Now scale it across a real vSaaS.

Say you are a $20M ARR vertical AI company. You are running maybe $6M a year in COGS, and $4M of that is model inference on closed APIs. Gross margin sits around 70%. Not great for software, but typical for the current wave of AI native vSaaS companies.

Move 70% of your inference workload to open source at roughly 10% of the closed cost. That $2.8M of closed inference collapses to about $280K. You just added roughly $2.5M in gross profit without touching pricing, headcount, or product surface area. Gross margin moves from 70% to about 82%. On a software multiple, that swing alone is worth tens of millions of enterprise value.

Now run it forward. At $50M ARR with the same mix, you are looking at $6M plus in annual gross profit uplift. At $100M ARR, north of $12M. That is real money. That is the difference between a company that has to raise again and a company that compounds cash.

And the customer never notices.

The frame most founders miss is that AI COGS is not a cost of goods problem, it is a strategy problem. Every dollar of closed inference you leave on a workload that could run open is a dollar you are handing to OpenAI or Anthropic that should be sitting on your balance sheet. Multiply that across a portfolio of vSaaS companies and you are talking about hundreds of millions of misallocated margin across the category.

The founders who understand this are quietly rebuilding their stacks right now. The ones who do not are going to look up in 24 months and wonder why their gross margins never expanded and their competitors got acquired at 10x while they sat at 5x.

Real companies already doing this — and what they are saving

A caveat before I get into this. Most vertical software and vertical AI companies have not publicly disclosed their model stack. It is one of the more guarded operational details in the category right now, probably because it is a real source of margin advantage and nobody wants to hand competitors the map.

So I had to pull in a few horizontal companies alongside the vertical ones to get the pattern across. The principle travels. The vertical operators I have talked to privately are doing versions of the same thing. They just have not blogged about it.

Harvey (vertical AI, legal). Harvey confirmed they are training open source LLMs to encode law firm workflows, in partnership with the firms themselves. Co-founder Gabe Pereyra was explicit about the two goals: serve frontier intelligence across their product at an affordable price with a strong security posture, and give law firms the foundation to own their own intelligence. Then Fireworks and Harvey published the math. On the Harvey Legal Agent Benchmark, GLM 5.1 hit a mean score of 0.8921 for $121 across a 100 task run. GPT-5.5 hit a nearly identical 0.892 for $560. Claude Opus 4.7 hit 0.911 for $954. Roughly 8x cost delta for a rounding error in quality. When they wrapped an open source worker model around a closed frontier advisor, they beat the best closed model on the all-pass metric at 39% of the cost. This is the cleanest published example we have of a vertical AI company composing open and closed.

OpenEvidence (vertical AI, healthcare). The most used medical search engine in the US, trusted by 40%+ of American physicians, publicly stated their reasoning for owning the stack: “closed-source models offer strong general capabilities, but they impose constraints like non-secure data, high cost per token, reliability and latency issues, and the inability to optimize for domain-specific performance.” They now train custom, domain-specific open source models on peer-reviewed medical literature. Baseten disclosed they projected $1.9M in savings just on training infrastructure, before you even get to inference. This is the company that first scored 100% on the USMLE. It is not running on someone else’s API.

Now the horizontal companies, included because they have published architecture details that most verticals have not, and the composition logic is the same.

Sierra (horizontal agent platform). Bret Taylor has been openly telling anyone who will listen that Sierra runs a “constellation of 15+ frontier and open source models.” Not one model. Fifteen. Open source models handle low latency tool calling and precision classification. Frontier models handle long context reasoning and empathy. He is on record saying there will not be one model to rule them all, and that any serious agent company will end up composing this way. Not a vertical company, but the pattern is exactly what a serious vertical AI stack should look like.

Pinterest (horizontal consumer). Their CTO went on record saying they are using open source AI at “less than 10% of the cost of leading, proprietary AI models,” with an order of magnitude reduction in inference costs on visual workloads. Their internal PinCLIP model outperforms state of the art open source multimodal embeddings by 30% on retrieval. Not a vertical company. But the shape of the argument travels directly. Fine tune open source on your proprietary data, beat the frontier on your specific task, save 90% on cost.

DoorDash and Instacart (horizontal marketplaces). Both use LLMs offline to enrich internal knowledge graphs and catalog data, while keeping runtime retrieval on classical, cheap infrastructure. The expensive intelligence runs once, batch, on their terms. The cheap runtime serves the customer. Same composition principle, just applied at horizontal scale.

The pattern is consistent across every company that has actually published its architecture, vertical or horizontal. The ones building enduring AI businesses are not picking sides. They are composing. Open source for volume, control, and margin. Closed models for frontier tasks where quality is non negotiable. A routing layer they own between the two.

I assume the vertical operators quietly doing this look exactly like Harvey and OpenEvidence.
They just have not written the blog post yet.

Always prioritize winning your vertical.
Optimizing the stack is not as important…

Model composition is a second-order problem.
Becoming the default choice in your industry is the first-order problem.

Everything above is real. The margin math is real. The routing layer is real. The Harvey and Sierra and OpenEvidence examples are real. And none of it will save you if you lose sight of the actual game.

The founders who lose in every scenario are the ones who spend more time composing models and less time building THE solution for their vertical.

Model strategy is a second order problem. Owning your industry is the first order problem. If you are not the default answer when someone in your vertical thinks about the workflow you serve, no amount of clever inference routing is going to fix that.

Open source is a lever.
Closed models are a lever.
Fine tuning is a lever.
Evals are a lever.
Levers are useful if you have already picked up something heavy.
They do nothing if you are still trying to figure out what to lift.

So use this piece the way it is meant to be used. When you have real volume, real workflows, and real customers who cannot imagine running their business without you, come back and optimize the stack.

That is when the 82% gross margin conversation actually pays off. Until then, every hour you spend on model composition is an hour you are not spending owning your vertical.

Own the industry. Own the workflow. Own the data. Own the customer. The model layer will still be there when you get to it, and by then it will matter more than it does today because you will actually have something worth optimizing.

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

Linear #187.5: Survive to Iterate with Alex Oppenheimer (GP @ Verissimo Ventures)

Wed, 29 Jul 2026 14:04:25 GMT

Before we dive in today, we have a quick ask!

Fill out our 100% anonymous annual benchmark survey and get a FREE ticket to the Vertical Software Summit this November AND access to the report when it drops.

It will help all of us understand what is working in verticals, specifically around AI.

Fill Out Survey Here

Survive to iterate.
Think in curves.

I’ve cancelled four software subscriptions this year and rebuilt them myself. My friend the barber is doing the same thing in ChatGPT between haircuts. And Alex Oppenheimer of Verissimo Ventures, NEA before that, Morgan Stanley tech banking before that, spent an hour with us walking through which VC priors actually survive when the median person walking into your shop is now, at least casually, a software developer. The answer is more nuanced than either the doom

take or the everything-is-fine take.

The macro update: the build-vs-buy line moves, but it never goes away. Small, tangible tools are on the cancellation list. Anything complex, cross-team, or regulated is more firmly in the buy column than it was a year ago, and the vendors on the right side of that line just got wider moats — not narrower ones. Specialization of labor is a hell of a thing. The HVAC owner should not spend half his week debugging his own dispatch tool. That’s still true. It might even be more true now.

The second shift is a founder one: the real debate is big vs. small, not vertical vs. horizontal. The vertical software cohort of the last decade quietly proved that the markets everyone thought were too small never were. AI is that same story again, at higher amplitude, because it’s now eating services budgets — 80% of GDP — instead of just software budgets. Retrospective pattern-matching is the tax you pay for not being early.

And the capital one: think in S-curves, not lines. The current fundraising market is bimodal — Notion memos priced at $80M, revenue rockets priced at $10B, everything in the middle is stuck in Death Valley. The pre-seed founder’s real job in four words is survive to iterate. The pre-seed investor’s job is to right-size the fund so a $500M outcome still works, because IPO nostalgia is a bad way to plan a decade.

This Weeks Vertical Titan:
Alex Oppenheimer and the case against the wrong argument

Alex ran early-stage at NEA for the better part of a decade before spinning up Verissimo. He is a mechanical engineer who wandered through Morgan Stanley, sat on eighty-plus cap tables, and now writes small checks with the specific contrarianism of someone who has watched three cycles in a row. His pattern recognition on cycle-vs-permanent is unusually calibrated, which is why the first thing he told me was that we’re all asking the wrong question. It is not vertical versus horizontal. It is not AI-native versus AI-later. The real axis is big versus small, and it is quietly doing all the sorting.

The build/buy line has moved before. It moved when Salesforce was born. It moved again when Airtable and Notion made every ops manager briefly believe they were a software company. It is moving now, further and faster, because you no longer need to know how to code to ship. But, and this is the part everyone selling the “software is dead” thesis quietly leaves out, the total surface area of things that could plausibly become software is expanding at the same time. The line drifts up. The chart underneath it gets taller.

Quick PSA -
90 Days Until The Vertical Software Summit

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!

Alex tells the story on himself. He churned his Salesforce subscription. It was $25K a year. He decided he could do better. He hired three developers to keep the replacement alive. The bill now runs closer to $80K a year, and he is on the hook for the roadmap. This is not an anti-vibe-code polemic. It is the oldest lesson in software procurement, dressed up in 2026 clothes.

Anything you build for yourself, you also own. The invoice moves from a vendor line to a headcount line, and the headcount line grows.

“I churned Salesforce. It now costs me eighty thousand a year to support what I replaced it with. That is the whole build-versus-buy debate in one anecdote.”

The frame Alex kept coming back to, and the one I have been thinking about since, is that vertical versus horizontal is not the interesting cut anymore.

Big versus small is. The prosumer app charging $12 a month for a lightly styled form builder is exposed. The complex, mission-critical, boring, un-sexy enterprise system that runs a manufacturer’s floor is not.

Both can be “vertical.” Only one survives the barber.

This maps neatly onto a shift Alex has been calling for years: the rise of the millennial decision-maker. Someone in their late thirties inherits, buys, or simply gets promoted into a traditional business.

They walk in, look at the tools, and ask why any of it looks like 1985. Ten years ago they had to wait for a vendor to build them something. Today they can vibe-code a passable v1 in a weekend. The incumbency you thought you had — because “these people don’t adopt technology” — is gone. And that is a demand-side story, not a supply-side one. It is not about the tools. It is about who is running the SMB.

Which brings us to the other Oppenheimer axiom, the one I’ve quoted in three conversations since we hung up: survive to iterate. Four words. He tells it to every pre-seed founder he backs.

The big funds, he argues — and I agree — give the opposite advice at that stage. They talk about scaling GTM, hiring aggressively, running the sales playbook. But at pre-seed the product isn’t baked, the market often isn’t ready, and the founder’s real job is to still be in the game when timing finally arrives. Timing is the single largest ingredient in startup outcome. You cannot control it. You can only outlast it.

“Your job at pre-seed is to survive to iterate. Not to scale. The big funds tell you the opposite, and at that stage they’re wrong.”

What to steal from Alex’s field notes:

#01. Bet on the sleepy strategic, not the loud one.

IPO windows will reopen — Alex is convinced the bar has been raised absurdly and will come back down — but the underrated wave of liquidity in vertical AI is the non-tech strategic acquirer. Devcon can’t build an AI construction platform. EvolutionIQ getting acquired in insurance is a preview, not an anomaly. The vertical incumbents without an AI strategy have three options: build (they can’t move fast enough), cannibalize (they won’t), or buy. Multiples will not stay low for long. If you are running an AI-native business inside a vertical whose largest player currently has no AI leader on their org chart, you are the acquisition they are quietly modeling.

#02. Survive to iterate — the pre-seed founder’s real job in four words

Alex has a line he gives every pre-seed founder he backs, and it is the exact opposite of what a $2B platform fund will tell you the day after they wire your seed check. Your job is not to grow. Your job is to survive to iterate. Because at genuinely early stages the product isn’t baked and the market probably isn’t ready either — and by the time it is, you need to still be alive. Big funds arrive with scaling advice, which means hiring advice, which means burn advice. That advice is a fine way to run out of runway before the market shows up.

This is the founder version of the S-curve. You’re investing pre-inflection. You don’t know exactly when the inflection lands. You just need to be there when it does, which means being there before it does, which means today it looks less interesting than it will in eighteen months. If your capital plan cannot tolerate that gap, you’re playing a different game than the one that actually pays out.

#03. Big vs. small is the real debate — vertical vs. horizontal is the distraction

One of the sharpest reframes in the conversation: for a decade every panel argued vertical vs. horizontal SaaS. Alex thinks the argument the AI era actually cares about is big vs. small. If your product is a small, tangible solution you could describe on a one-pager, you should assume a competent operator will vibe-code it for themselves this year. If it’s a complex, cross-team, non-core system that no single buyer wants to own in-house, you have a real business — vertical or horizontal makes very little difference.

This is why the cheap prosumer end of the market is bleeding first. Nobody is going to rebuild a full enterprise-grade platform in Claude Code on a weekend, and even if a few try, they will not scale it to a thousand users. But the $10–$20/month prosumer subscription with three features? That’s now on the cancellation list of every operator who owns a ChatGPT account.

#04. The line moves — but build vs. buy never goes away

Every company that employs software engineers has always been running the build-vs-buy calculation. AI has not killed the question. AI has moved the line — a barber can now ship his own booking system, a founder can cancel four subscriptions and rebuild them — but the question survives untouched. And the moment you actually try to run the DIY thing at any scale, the specialization-of-labor argument kicks in fast. Alex’s own example: churning a $25K/year Salesforce subscription and immediately paying $80K/year for three developers to hold up what he built himself.

The correct read isn’t that software is dead. It’s that the wrong software is dead. Anything a smart operator can hold in their head and reproduce in an afternoon of vibe coding is exposed. Anything that touches multiple teams, multiple regulators, real economies of scale, or real edge cases stays firmly in the buy column — and the vendor who owns it just got a wider moat, not a narrower one.

Purpose-built embedded capital for vertical platforms.

#05. Earned insight beats a Stanford resume — go do the job first

Alex tells college kids the same thing he was told about finance: you don’t want to study finance, you want to study the finance of something. Same rule for founders. If you want to build AI revenue cycle for hospitals, spend real time in a hospital before you spend anyone’s capital. The single quickest way to lose Alex on a first meeting is to pitch an idea you could have Google-searched into an existing public company.

This is where the from-industry founder still crushes the Bain-trained user-interviewer. Not because pedigree is bad — Alex ran mechanical engineering at Stanford, did Morgan Stanley, spent years at NEA — but because empathy for the problem is a load-bearing part of the product. Founders who wake up every morning going “why does this thing still not exist” build differently than founders who woke up going “what’s hot right now.”

#06. The barber is shipping software — respect it, but don’t confuse it with a business

The moment in the interview that stopped me: Alex pulls up a photo of his French barber, mid-cut, inside ChatGPT vibe-coding a full booking and scheduling system for the shop. That is the new normal. Every hands-on operator with a laptop and a curiosity can now materialize working software for their own workflow. This is unambiguously good — more people building, more real problems getting real solutions.

It is also not, by itself, a company. The barber loves cutting hair. The HVAC owner should not be spending half his week debugging his own dispatch tool. Specialization of labor exists for a reason: one team, focused on one problem for one hundred businesses, will always beat one hundred businesses each solving their own version of it at midnight. The question for founders is not whether people can DIY — it’s which DIY solutions collapse under the weight of the second, tenth, and thousandth user.

#06. Retrospective pattern matching — vertical markets are always bigger than they look

This is the sentence that reframes an entire decade of VC pushback: every generation of VCs is anchored about five years behind on which markets are big enough. The first cohort of vertical SaaS businesses were told their TAMs were too small. Software’s share of GDP kept marching up and the TAMs turned out fine. Supply chain? Too niche, too analog. Then a trillion dollars of freight went digital. AI is that story again, at higher amplitude — because it is now eating services budgets, not just software budgets, and services is 80% of GDP.

The operator implication is simple: if the loudest room-tone response to your market is “that seems small,” that is a leading indicator you’re early, not wrong. If the loudest response is “seven companies just raised in that space,” you are almost certainly late.

We’re doing a workshop tomorrow with a killer AI Product Expert who is going to talk about what it actually entails to build an AI Native Product.

The texture under the software nobody is going to vibe-code away.

From-industry founder + earned insight
The trucker who founded a freight platform. The millennial running a family business who demanded software that looked like Snapchat. The empathy shows up in the roadmap.
Complex, cross-team, workflow-deep
The kind of product you cannot describe on a one-pager. Multiple stakeholders, real edge cases, real economies of scale. The buy side of the frontier keeps expanding.
AI-native services in regulated niches
Fund admin, customs brokerage, licensed healthcare — the vendor gets paid for outcomes, owns the eval loop, and inherits a data moat the labs will never see.
Small-fund math that actually returns capital
$26M funds, 5% ownership, no billion-dollar fantasy required. Right-sized to the DPI the current exit environment can actually generate.

The common thread: earned insight into a job complex enough that a smart operator with a laptop cannot displace you in a weekend; priced and funded like a business that does not depend on an IPO that isn’t coming.

See you next Wednesday. If this one landed, forward it to a friend!
— Luke Sophinos

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

Linear #187: Pre-AI Software Incumbent? Here Are Your Five Paths Forward, The Story Of Cognite & Their Recent $3B+ Exit

Mon, 27 Jul 2026 00:02:07 GMT

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 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.

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 :-)

Linear #186.5: Sell the Fish NOT the Pole, with Jake Saper (Partner @ Emergence Capital)

Wed, 22 Jul 2026 14:03:09 GMT

Sell the fish, not the pole.

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.

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

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.

Purpose-built embedded capital for vertical platforms.

#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 :-)

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

Linear #186: AI Roll Ups: Two Years Later, Unveiling AGI

Mon, 20 Jul 2026 00:02:32 GMT

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.

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 :-)

Linear #185.5: Compliance is not a feature. It is a network.

Wed, 15 Jul 2026 13:59:04 GMT

The moat is not the rule.
It is how the rule gets enforced.

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…

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.

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.

Purpose-built embedded capital for vertical platforms.

#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.

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

Linear #185: Rule of 40 Deep Dive: Why It Still Matters, Public vSaaS R40 Analysis, How/When It WILL Impact You

Mon, 13 Jul 2026 00:03:33 GMT

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.

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

Linear #184.5: The data is the treasure. Everyone else is a pirate plundering for it.

Wed, 08 Jul 2026 14:03:09 GMT

Bury and protect your data treasure.
Then, go all-in on action.

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…

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.

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 “a big dumb database” 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 “is availability real or not” 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

#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…

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

Linear #184: Entrata's IPO = A New Public Vertical SaaS To Study

Mon, 06 Jul 2026 00:02:13 GMT

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.

ARR climbing to $574M with growth re-accelerating in Q1’26. Source: Entrata S-1.

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

Make payments mandatory. Optional products have 30% attach. Required infrastructure has 100%.
Price by transaction, not percentage. Fixed fees protect margin when interchange is low.
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.

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

Linear #183.5: What a founder, who sold his last company for $780M, is unlearning to build the next AI-native winner

Wed, 01 Jul 2026 13:59:19 GMT

Burn the instincts.
Own the distribution.
Build the hard thing.

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.

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.

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

#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.