Half Baked Newsletter August 13th.jpg

The Idea

In June 2023, Anton Osika was a 32-year-old CTO at Depict, a Y Combinator startup in Stockholm selling AI product recommendations to e-commerce sites. GPT-4 had shipped three months earlier and Osika had spotted something nobody had properly tested. The model could write a decent function, but nobody had asked it to write an entire codebase from one sentence. He spent a weekend building a command-line tool that did exactly that, then pushed it to GitHub on a whim. It trended worldwide within days. Tens of thousands of stars followed, along with a waitlist of 27,000 people asking for a version that did not require a terminal. That waitlist was the real discovery. Copilot only helped people who could already code. No-code builders like Bubble collapsed the moment an idea needed a database. The gap sat between them: production-grade code, ordered by conversation. Days after posting, Osika rang Fabian Hedin early on a Saturday and asked him to go for a walk. This was the beginning of Lovable.

The Execution

The lesson?

Osika did not build a model. He built the layer on top of everyone else's models, four months after GPT-4 made whole-codebase generation possible, and aimed it at the 99% of people who cannot code. That is a market nobody could serve in 2022 and everybody wanted in 2024. Lovable went from $0 to $500m ARR with fewer people than most Series B companies have in sales. Being early to a capability beats being first to a category.