This post summarizes the Y Combinator Lightcone episode Vertical AI Agents Could Be 10X Bigger Than SaaS (November 2024). The numbers and predictions are the hosts’ own. They are an investor argument for a category, not a forecast we have checked.
Key Takeaways #
- One host argues there could be $300 billion-plus companies in vertical AI. It is a prediction.
- The hosts say competition among AI model providers is good for founders and consumers.
- The case is made by analogy with SaaS: the argument that vertical AI could eventually be 10 times larger rests on companies spending more on employees than on software.
- The advice for founders: look for boring, repetitive work you have seen up close.
The Rise of Vertical AI Agents #
The hosts talk about vertical AI agents: AI built to do one industry’s work, such as a whole team’s job in an enterprise. They say Y Combinator has funded a bunch of them and that the world has not caught on to how big the category could get. They also say that AI model competition, with several strong providers rather than one dominant player, is good for founders and consumers.
SaaS vs. Vertical AI #
The case runs by analogy with software as a service. A host says that over the last 20 years more than 40% of venture capital dollars went to SaaS companies, and that Silicon Valley produced over 300 SaaS unicorns in that period, “way more than every other category”. These are the host’s figures; the episode does not cite a source for them, and we have not checked them.
He credits the browser’s XMLHttpRequest, added in 2004 according to the host, as the missing piece that made rich web apps, and so SaaS, practical. His view is that large language models are a similar enabling step.
The Three Buckets of Startup Ideas #
Another host sorts the billion-dollar startups into three groups:
- Obviously good ideas: mass-consumer products such as docs, photos, email, calendar and chat. Much of the value in these went to incumbents.
- Ideas nobody predicted: the episode names Uber, Instacart, DoorDash, Coinbase and Airbnb. The hosts say incumbents did not try to compete until it was too late.
- B2B SaaS: where, in their account, most billion-dollar companies came from, because no single player dominated.
Why Incumbents Missed B2B SaaS #
A host’s explanation: each B2B product needs people who go very deep in one domain. His example is payroll. Why did Google not build a Gusto competitor? Nobody at Google, he says, understands payroll or has the patience for its regulations, and it is easier to focus on a few huge categories.
The Future of Vertical AI #
The hosts argue that vertical agents can do more than replace software. Companies spend far more on employees than on software, and agents could take over the data entry, approvals and clicking through software that people do now. From this, a host says it is “very possible” that vertical AI companies will be 10 times as large as their SaaS counterparts. That is a possibility the hosts raise, not a number anyone has measured.
Real-World Examples of Vertical AI #
The episode gives these examples of Y Combinator companies:
- Outset: uses large language models for surveys and qualitative research, a space where the host expects an incumbent like Qualtrics will not build the best AI product.
- Momentic: AI QA testing. The host says its pitch is not “makes your QA people faster” but that you may not need a QA team. See Momentic on Y Combinator.
- Salient: AI voice calling for auto-loan collections, replacing work done in call centers that the host describes as high-churn jobs.
Taken together, they show where a startup can sell: to a buyer who does not have to be convinced to cut their own team.
Finding Your Vertical #
If you are a founder, the episode’s advice is to look for boring, repetitive tasks. Start where you have personal experience or connections, since that is where you will see the real friction.
Conclusion #
The argument: vertical AI agents could repeat what SaaS did, by automating repetitive work in specific industries, and could produce large companies. Treat the sizes as an investor’s thesis. The part that holds up either way is the method: pick one industry, learn its dull work, and automate it.
