I got rejected from the Anthropic hackathon. I built it anyway.

Anthropic ran a hackathon. $100k in API credits for the winner. 500 participants. A week to build something that uses Opus 4.7 and Claude Code.

I applied. Didn't get in.

The idea I had was Branch AI — a collaborative reasoning canvas that wraps the Claude Code CLI and captures extended thinking as a navigable, forkable tree. Every reasoning step becomes a node. You can branch at any point, inject new context, diff runs side by side.

The problem it was solving: Claude's extended thinking is powerful but opaque. You get a wall of reasoning you can't interact with. Branch makes it a structure you can walk through.

I thought it was a good idea. Anthropic apparently had enough applicants.

So I built it anyway.

Published branch-ai on npm on April 24th. Crossed 500 downloads in under two weeks.

---

I'm not writing this as a redemption arc. I'm writing it because rejection is just information about supply. It tells you the slot was full — not that the idea was wrong.

The thing worth building doesn't need permission to exist. It needs an npm account and a README.

If you've been waiting for someone to greenlight your idea: this is what not waiting looks like.