About

A public lab for AI-built software.

The Build Hypothesis is a YouTube channel and an ongoing experiment. Give several AI coding agents the same ambitious challenge, the same constraints and a fixed time budget. Let them build on their own. Then test what they made and show the evidence.

Why this exists

Everyone has an opinion about AI coding agents. Most of it comes from demos, screenshots and confident announcements. This project asks a narrower and more useful question:

What can an AI coding agent deliver when given a serious product specification, limited time and a fair opportunity to work autonomously?

That means asking whether the app actually works end to end, or only looks finished. How much help the agent needed. How it handled ambiguity, failing tests, deployment and real integrations. How usable, secure and maintainable the result is. And how the answers change as the challenges get harder.

How it's different

The experiments are built to be checked. Every spec is versioned and public, and every run leaves a repository. Every claim on this site is labeled with its evidence level, from what the spec asked for to what independent testing confirmed. When something fails, that's part of the story, not something to edit out.

The format works for professional developers, who can dig into architecture and test methodology, and for anyone curious about where AI-built software really stands.

Get in touch

Propose a challenge, or just follow along

There's no public email address yet. For now, the YouTube channel is the best place to reach the project and suggest the challenge you want to see the agents attempt. The source repositories are public on GitHub.

Watch the agents build, break and prove it.

Every experiment becomes an episode: the spec, the runs, the failures and the hands-on evaluation. Subscribe to see the results first.