Champion OS: I built an AI that tells me no

Every AI I've ever worked with thinks my ideas are wonderful. Ask Claude whether you should quit your job to build a course on typing speed and it will have the landing page drafted before you finish the sentence.

I hate that sycophantic nonsense. It’s there for dopamine hits and to keep me using Claude as if it’s me chatting with my buddy, and I know that’s not true. So I decided to flex Karpathy’s idea of “A Council”, which I had been doing for a while (it’s fun getting agents to argue with each other). This time, however, I wrapped it with process and fact-finding, and I absolutely love what I came up with.

I’m calling it Champion OS because it acts as your champion in any given situation. Like an agent works for a pro football player, this tool is working on your behalf to figure out the best way forward in any situation. Agent is so overused these days, so I went with champion, which sound a lot more fun.

It’s a free plugin for Claude Code. It opens with an interview, twenty or thirty minutes on your goal, your money, your values, and the pattern that keeps sabotaging you, then writes it all to a private dossier that never leaves your machine. From that point on you have a council that knows exactly who it's advising.

Bring it a business idea, a job offer, a negotiation, a draft, or a pull request, and it runs a three-round gauntlet: one round to stretch the idea as far as it can go, one to put it on trial (a prosecutor, a risk officer, a voice asking what it costs you in hours you don't have), and a final round that lands on a verdict from one to five with the strongest dissent preserved word for word. Verdicts stack up in a log the council reads next time, and you can grade its old calls against how things turned out, so the mirror gets more honest the longer you look into it.

Hope you like it! I had a ton of fun making it, and happy to hear your ideas.