Founders Should Chase Secrets

What's next for Vertical AI

Public and private market commentators spilled plenty of ink last week about the SaaS market correction. With the software indices down ~22%, the punditry alarm was warranted for a change.¹ The sell off has wiped out hundreds of billions of value. Founders, take note.

I agree with the teams at Meritech and Altimeter: investors have changed their long term expectations about SaaS companies’ terminal values.² The rationale of the “SaaSpocalypse” was the threat of competition: vibe coders (maybe in SMB), AI native start ups (yes), and AI model providers (definitely!) will displace many legacy software companies. Public markets gave a coherent, albeit painful, message to management teams. “Just do better on AI, or else.”³

But what of Vertical SaaS?

A simple basket of vertical SaaS stocks — Veeva, Procore, Toast, AppFolio, and Guidewire — outperformed horizontal SaaS names like Salesforce, ServiceNow, CrowdStrike, Datadog, and Snowflake over the last couple weeks.⁴ But the results are relative. Vertical SaaS multiples have cratered, too.

So is Vertical SaaS also “over”? My unsatisfactory answer: it depends. That was my view before the sell off, so dear readers, you’ll have to trust me.⁵

Vertical AI is fine, right? Right?

The prevailing wisdom has been that Vertical AI is next gen Vertical SaaS. But we’re at a turning point in these AI-native, industry-first companies. We’re in a new era of Vertical AI. Call it “Vertical AI 2.0” in the local parlance.

The core trait of Vertical AI 1.0 was not a lack of ambition. It was assuming the model layer with some unique workflows would remain the scarce asset.

Vertical AI 1.0 was model-first, office focused, and constrained to a few industries. Vertical AI 2.0 is data-first, extends beyond the desk, and applies to nearly every industry.

Vertical AI 1.0 began in late ‘23: novel approaches to LLM-friendly industries like legal, healthcare, and finance. Incredible companies and exceptional founders the likes of Ambience/Abridge, Legora/Harvey, and so forth. Many of these companies are and will continue to dominate, despite frenemy competition with model providers. It will be exciting to watch.

But for many of the other companies, their Vertical AI 1.0 strategy appears to have been: piggy back on model updates with maybe some fine tuning as the sole means of product improvements. Free riding is not a dominant (product) strategy. In recent weeks, Anthropic launched a suite of new finance tooling, OpenAI released Healthcare and then Frontier, and so forth.

Vertical AI 2.0

Vertical AI 2.0 is broader than before. Founders are exploring new markets and approaches previously thought too small or impossible in Vertical AI 1.0. It’s time to “boil the ocean,” as Garry Tan has urged.⁶

Vertical AI 2.0 is about finding where defensibility lives. And defensibility is a moving target across office and non office (i.e., field) workflows.

For inside the office, more defensibility will accrue to markets with an arcane mix of accounting, legal, & regulatory rules and unique access to data. Founders need to have a different product strategy than Vertical AI 1.0. Eli Dukes recently coined the term “Systems of Training” for these kinds of companies:

“I have a hunch that the companies who model themselves as an applied system for producing the data recipes and decision logic for training models will ultimately accrue far more value than those who model themselves more as product-centric companies focused on the UX around AI for a Vertical. This is not because products don’t matter, but because the last non-commodified layer left are these data recipes.”

Differentiated Data Recipes, Undifferentiated Compute

Eli is on to something, particularly for office based workflows.⁷ Product workflows alone will not create deep moats.

But for those deskless users, Vertical AI 2.0 products will have a different look and feel. Many jobs in the field require days working in hard to reach places, “offline mode,” and physically nimble problem solving that seems out of reach for even the next generation of humanoid robots & models.

This is Moravec’s Paradox in practice.⁸ Put simply, robots have a harder time with mundane physical tasks than intellectually challenging activities like chess etc. In the intervening 5 (or 10? 15?) years until the robots arrive, we will see new and different application form factors in the field: wearables that finally work, and passive voice / video to capture data on mobile devices to run computer vision and other models. Mobile apps won’t be deprecated, but they will feel fresh. Data capture GUIs will shift from the pick lists everyone hates to just, well, pixels. Time to fire up those Swift projects on Claude Code.

The winners of Vertical AI 2.0

The answer is simple, not simplistic. The winners will be the most exceptional founders who can find secrets. Before exploring the winners, who won’t win in this new phase: the prototypical Vertical SaaS-cum-AI investor.

Around 2016 – well after the nonobvious Vertical SaaS investments in Veeva etc. – a new type of Vertical SaaS investor emerged from the primordial soup of South Park. Decidedly non technical, this investor sought out Vertical SaaS because it was comprehensible. Size a market, find a founder, define a ‘control point,’ eventually find a way to ‘embed fintech,’ and voilà a unicorn. This approach succeeded when building software was hard.

What this investing approach always missed: the search for secrets and the founders who can find them.

The Secret Seekers

Secrets are the market insights that, when paired with novel technical solutions, can create generational companies. As a founder, finding secrets does not require having worked in said industry a priori but does require having observed, listened, and learned. A lot. This work is painstaking, uncomfortable, and limitless–what Paul Graham famously called a “schlep” in 2012.⁹ Not much had changed in the intervening 15 years of schleping: one must take red eye flights, stay in seedy motels, and grapple with contradictory product feedback from (sometimes) ornery users.¹⁰

However, what has changed in Vertical AI 2.0: building with agents and models for specific industries. Non deterministic systems are particularly challenging for these types of customers who expect things to “just work, damn it.” These are not fellow startups or Big Tech. And compared to copilots, building agents is even harder, because founders have to attempt to displace said ornery users.

The schlep is worse than ever before, but the size of the prize is greater. The Vertical SaaS 2.0 market opportunity is orders of magnitude greater than Vertical SaaS and Vertical AI 1.0. The schlep will be worth it.

The willpower to find the secret alone is not the path to success for founders. Succeeding requires a new technical solution to solve problems that have not been addressed before, or done well by incumbents. The founders need to push the tech – in this case agents, computer vision models, and even some classic UI/UX – to its limits.

In Vertical AI 1.0, there was an overcorrection toward teams who were at the bleeding edge of the models (good on them!) but little affinity for, or unique insight into, the actual market.¹¹ These half-formed secret seekers created an explosion of funding in specific corners of the economy, creating products with high gross dollar churn and little in the way of differentiation. For instance, the 11th attempt of some purportedly ‘cracked’ team to build a healthcare scribe, despite having spent little time around provider offices. Depending on the market, the model providers pose a threat to these companies, too. These founders did not schlep, so they never found secrets.

So, the best secret seekers for Vertical AI 2.0 have a few traits. They have the determination, technical ability, and intellect to locate the secret(s) and then build magical product experiences, for agents or humans or both. What drives them: an near unsettling desire to find those exact secrets, at this moment in time.

Investors will no doubt chase Vertical AI 2.0. The best founders will chase secrets.

---

Endnotes

1.The BVP Nasdaq Emerging Cloud Index & iShares Expanded Tech-Software Sector ETF | IGV

2. Clouded Judgement 2.6.26 - Software Is Dead...Again...For Real this Time...Maybe? & Time’s Up for SaaS (Grow Faster or Vanish)

3. I’m reminded of what Andrew Lo at MIT calls the “wisdom of the crowds” vs. the “madness of mobs” The wisdom of crowds versus the madness of mobs: An evolutionary model of bias, polarization, and other challenges to collective intelligence

4. I prompted GPT 5.2 to build me a horizontal vs. vertical comparison chart using the BVP EMCLOUD index. The model did a pretty good job selecting the baskets and showed a 500 bps outperformance by the Vertical SaaS companies. So l/s hedge fund readers, please don’t roast me on the narrow comp set; it’s an example.

5. While not investment advice (actually), here’s my simple decision tree to think through public and private market Vertical SaaS at the moment. Is the company performing well in recent quarters against plan? If not, you’re toast. Then, is the company founder led? It better be, because the moral authority required to drive change here is immense. We’re still not done…Is the founder locked in and ready for the fight? If she’s not, a slow march toward irrelevance will ensue: years of elevated churn, reduced NDR, and pricing pressures. Not dead but comatose: the once mighty “system of record” becomes a SQL database. Lastly, is the team in place today – at the mid level, in the trenches – able to plan, design, and ship using the models? And can they do it quickly enough to develop durable and reportable AI revenue that can blunt AI-native competition and even the model providers in some Verticals? If all those hold true, the odds are fine for some Vertical SaaS providers, including the private ones. Buckle up.

6 https://garryslist.org/posts/boil-the-ocean

7. I’m still formulating my views here. I’m not sure Eli is entirely right on the product piece in the short term. Presentation and engagement layers still matter; product elegance, regardless of who uses it (agent or human or both) still matters. “You know it when you see it.”

8. The Physical Intelligence team has some interesting observations here: https://www.pi.website/blog/olympics

9. https://paulgraham.com/schlep.html. My approach to “secrets” is bottoms up (Paul Graham) vs. more tops down (Peter Thiel).

10. Peter Fenton recently defined great entrepreneurs similarly on Jack Altman’s podcast, Uncapped (#41)

11. H/t to Ramy Adeeb for this general observation