Oct 28, 2025 · 39m · y-combinator
From Idea to $650M Exit: Lessons in Building AI Startups · Y Combinator
0:00 / 0:00
▶ Watch on YouTube →
gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions
In this Y Combinator presentation, Casetext co-founder and CEO Jake Heller shares key frameworks for building high-impact AI startups, drawing lessons from Casetext's $650 million acquisition. He guides founders on targeting massive paid labor markets, building enterprise-grade products through rigorous evaluation frameworks, and pricing AI as labor replacement.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. How this is scored →
The partners as informed peer 0.0 Guest teaching 0.0 Guest disagreement 1.8 The partners pushing back 0.0
speaking balance: gold is the partners, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 24:38 Rejecting VC Consensus on Sales vs Product
Jake forcefully dismisses standard venture capital orthodoxy, calling the claim that marketing outweighs product quality 'fucking bullshit' based on his own decade of startup experience.
Hardest push from the partners ▶ 30:50 Reframing the Question on CompetitorsWhen asked how to navigate existing competition when choosing an industry, Jake rejects the premise immediately, stating that founders should not care about competitors at all.
Biggest teaching moment ▶ 20:10 Demystifying AI Reliability and Prompt EvalsJake educates the audience on why most AI startups fail after raising seed capital on demos, laying out the unglamorous two-week prompt evaluation grind needed for enterprise-grade accuracy.
The partners hold their own ▶ 38:58 Rebutting the 'Thin Wrapper' AnxietyWhen challenged about becoming an indefensible GPT wrapper, Jake counters immediately by explaining that the sheer complexity of fine-tuned prompts, checks, and data pipelines makes copying genuine products nearly impossible.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | The partners as informed peer | Guest teaching | Guest disagreement | The partners pushing back | Why |
|---|---|---|---|---|---|---|
| How to Pick an Idea: Target Paid Labor and Expand TAM | 0 | 0 | 1 | 0 | Jake delivers a solo lecture explaining that founders should pick startup ideas by looking at what tasks companies currently pay human labor to perform, expanding the traditional software TAM by orders of magnitude. As a monologue without a host present, host-side scores are zero. | |
| Building AI Products: Domain Expertise and Workflow Engineering | 0 | 0 | 2 | 0 | Jake breaks down workflow engineering and the necessity of domain expertise, criticizing developers who rely on complex frameworks when straightforward deterministic Python pipelines suffice. The segment is an instructional monologue. | |
| The Critical Need for AI Evaluations | 0 | 0 | 2 | 0 | Jake outlines the rigorous evaluation process required to move AI products from 60% demo reliability to 97%+ production reliability. He emphasizes that sleepless iterative prompt tuning on objective benchmark datasets is where most competitors fail to follow through. | |
| Marketing, Selling, and Customer Success for AI Startups | 0 | 0 | 3 | 0 | Jake forcefully rejects standard VC advice that sales and marketing matter more than product quality, recounting how product excellence drove organic inbound demand at Casetext. He also warns founders about relying on pilot revenue that fails to convert. | |
| Q&A: Navigating Competition and Market Selection | 0 | 0 | 2 | 0 | An audience member asks about selecting markets with existing competitors. Jake dismisses worrying about competition entirely, explaining that target markets are multi-trillion dollar opportunities and competitors are often surprisingly weak. | |
| Q&A: Founder Focus Across Startup Stages | 0 | 0 | 1 | 0 | Michael asks how founder focus should shift across funding stages from seed to exit. Jake answers with self-deprecating clarity that founders should obsess solely over product-market fit at every single stage rather than getting distracted by secondary business functions. | |
| Q&A: Targeting Massive Problems for Maximum Impact | 0 | 0 | 1 | 0 | A young founder who exited at 14 asks what problem to tackle next. Jake reflects on his early mistake of targeting narrow legal software before LLMs and encourages working on the largest, most ubiquitous problems imaginable. | |
| Q&A: Pricing AI Services Beyond Human Capabilities | 0 | 0 | 2 | 0 | Sabod and another audience member ask about pricing novel AI capabilities and avoiding being classified as a generic wrapper. Jake briskly explains that intense technical and domain workflow integration creates natural defensibility. |
Statements from this episode (16)
Heller: Casetext received early access to GPT-4 in summer 2022
“Because we were so focused on large language models and were researching deeply in this space, we got really early access to GPT-IV. Like summer, 20, 22.”
Heller: Casetext pivoted with $20M revenue and 100 employees for GPT-4
“We were like twenty million dollars in revenue. We were doing great. I had like a hundred people and we stopped everything that we were doing and said, we're going to build something totally new based on this new technology.”
Heller: AI application TAM is 1,000x larger when capturing labor budgets
“Today, the actual amount of money that we already know people and companies are willing to spend is the combined salaries of all the people they're currently paying to do the job. And that number is like a thousand X bigger. You pay 20 dollars a month to solve…”
Heller: 30% to 40% of Casetext Employees, Including Coders, Were Lawyers
“It was helpful for us that, that I was a lawyer, my co-founders were lawyers, 30 to 40% of my company, even the coders were lawyers, because we actually lived it.”
Heller: Casetext built CoCounsel with plain Python, dismissing frameworks like LangChain
“And to be honest, a lot of the stuff that we built while building co-council was exactly like this. Every time you do this task, you're basically gonna take the same six or seven steps. And you don't need to have, frankly, like fucking Langchain or whatever. J…”
Heller: Systematic prompt engineering can replace fine-tuning in AI apps
“What you'll find without any, these are for fine tuning, without any like technical fine tuning, you can go so far with just prompting. If you're being really careful about this, you will find that the AI gets things wrong predictably.”
Heller: Casetext derived most evals from customer failures, not lab tests
“We've added much more evals at this point from real things that happened to real customers than the ones we came up with in the lab.”
Heller: Most AI app developers fail to run evaluations or study workflows
“If you just do these two things, you'll be like, 90% of your way there to building a better AI app than what most of the crap that's out there, right? Because most people never eval, and they never take the time to figure out how professionals really do the jo…”
Heller: Enterprise customers preferred predictable $6,000 annual seats over usage pricing
“When we asked our customers, they said, listen, I'd rather pay more, but make it like consistent throughout the year, then potentially pay less and pay per use. So our customers wanted to pay a 6000 dollars per seat. They want to proceed and they want to pay 6…”
Heller: AI startups face mass extinction as enterprise pilots fail to convert
“A lot of those pilots are not converting to real revenue, and there's going to be a mass extinction event as a lot of pilot revenue, it's like, instead of ARR, it's like PRR, like Pilot Recurring Revenue, or something like that, or not even recurring, just Pil…”
Heller: AI startups with superior UI will lose without strong customer enablement
“Your product isn't just the pixels on the screen. It's not just what happens when you click this button. It's the human interactions with your support, customer success, with the founder it's training, it's, you know, everything around it. If you don't get tha…”
Heller: No single AI company will win entire knowledge work markets
“First of all, for some of these spaces, the market is so big because we're talking about like how much, how many trillions of dollars are being currently spent on like marketing professionals or support professionals or whatever. There's not going to be a sing…”
Heller: Roles companies already outsource abroad are ideal targets for AI automation
“What are the kinds of roles that people are currently outsourcing, say, to another country? Right? If it's something that they're willing to do that for, then that's probably a pretty good target for what AI could take over.”
Heller: Targeting legal tech pre-LLMs was a mistake due to low software spend
“I actually feel like in some ways for us in the early days focusing on legal, Made sense for us because I knew legal, but also it's kind of a mistake because at the time the legal software industry Pre-LLMs is actually pretty small because it was like a fracti…”
Heller: AI competition will drive legal service costs to pennies on the dollar
“I think at first you can start charging when the human's charging and then you'll have competitors and they'll come in and they'll charge a little bit less. And then other competitors will come and they'll charge a little bit less. And it's kind of beautiful, …”
Heller: AI defensibility comes from the brutal complexity of building it
“Just build it, and as soon as you build it, you'll see how fucking hard it was to build it, how many little pieces you have to build, how many data integrations, how many checks, how fine-tuned the prompts need to be, how you have to pick your models super wel…”