yaaf is my side project: a way to hand real work to a team of AI agents and have it actually get finished.
I’ve spent the last year writing about why agents break in production. They mark their own work done. They invent busywork and call it progress. They loop with each other at full price, and they can’t tell you when they’ll be finished. Eventually writing about it wasn’t enough, so I built the thing I kept describing.
You describe a project in a paragraph. yaaf drafts the team (a CEO, specialists, a budget and a delivery date), lets you rehearse the whole thing for free on scripted agents, then runs it for days or weeks. You’re the chairman. You get a daily summary and only the decisions that need you.
The idea underneath is simple: the rules live in the server, not in the prompt. A prompt is a suggestion, and a clever model talks its way around one. In yaaf, “done” means the platform checked it, every dollar is capped, and the date can’t quietly slip. An agent that breaks a rule doesn’t get a stern reminder. It gets a 403.
It’s young. The first commit was in August 2026, built in the open with Claude Code. It’s free, open source and self-hosted, works with Claude, GPT, Gemini and others, and plugs into Claude as an MCP server so you can run it from the chat window you already have open.
Visit yaaf.dplouffe.ca → · Source on GitHub
If you try it, I’d like to hear what breaks.