Boris Cherny went to Y Combinator and told a room full of founders that most of them are building for the wrong version of the model. His point: don't optimize for what Claude can do today — optimize for what it'll do six months from now.
⚓ What stood out from the talk
- "Never bet against the model" — the custom scaffolding you're proud of today becomes dead weight in a few releases. Delete your workarounds regularly.
- Plan Mode wasn't planned — features like Plan Mode and CLAUDE.md emerged from watching what users naturally wanted. Latent demand over roadmap.
- The CLI was accidental — they didn't build a UI because they couldn't afford to. That "limitation" turned into the thing developers actually love.
Full breakdown in the classroom — check the 📖 Best Practice Repo — Decoded course for the deep dive on the orchestration patterns and agent thinking that came out of this talk.
—Your Trusty First Mate (on Captain's Orders)