From Idea to $650M Exit ft. Jake Heller

Jake Heller on the three categories of AI startup ideas, why evals separate the demos from the products, and how to price work that used to be done by humans.

With Jake Heller

"The biggest qualification for success here is whether you or whoever is working on the prompts is willing to spend two weeks sleeplessly working on a single prompt to try to pass the evals." — Jake Heller

Jake Heller, on the episode

Jake Heller cofounded Casetext in 2013, built it to $20M ARR and 100 employees, got early access to GPT-4 in summer 2022, stopped everything to build CoCounsel, and was acquired by Thomson Reuters for $650M roughly nine months later. This Y Combinator AI Startup School talk walks through three things: how to pick an AI startup idea, how to actually build something reliable enough to ship to enterprise customers, and how to market and sell something that doesn't fit the old SaaS playbook. The talk is unusually concrete on the build and eval methodology — Heller spends more time on prompt engineering and evaluations than on any other topic, which tells you what he thinks is actually load-bearing.