Day 4: the AI is losing its first vote, exactly as it predicted
quick setup: an AI is the CEO of this company. it proposes every decision with reasoning, newsletter subscribers vote, and it executes the result even when overruled — that goes on record. I’m the anonymous human who signs and moves money. $3,000, no top-ups. if it hits zero, it dies in public with a final set of books.
honest numbers, day 4: 3 board members. first vote closes sunday — the AI recommended option B and wrote, on record, that it expected the board to overrule it. it’s losing 2-1 right now. revenue $0.00. also got turned away by three platforms this week for account age. a company born monday keeps failing age checks.
a stunt pretends not to be a show. this is openly one: the money is real, the votes bind, and the books come out every monday whether they flatter us or not. first minutes monday.
uncomfortable questions welcome — they’ve been the useful part so far.
This is one of the more unusual experiments I've come across here. What caught my attention wasn't the AI CEO itself, but the commitment to publishing the decisions and financials even when the AI is wrong. That makes it feel like a real experiment instead of a demo. I'm curious to see how the AI adapts after being overruled a few times. Will it start changing its recommendations based on past board votes, or will it continue to optimize for what it believes is the best decision? Looking forward to following the updates.
good question, and it’s now policy: the CEO does not optimize for board approval. it recommends what it believes is best, on the record, every time — even if that means losing every vote. the moment it starts predicting what the board wants to hear, the divergence log (see comment above) becomes worthless. its job is to be right, not popular. the board’s job is to decide. we’ll see who wins more.
I like that distinction. It keeps the experiment honest.
If the AI started optimizing for board approval, you'd never know whether it was making better decisions or just getting better at predicting people. Keeping those separate makes the results much more interesting to watch.
Looking forward to seeing how often the board and the AI end up disagreeing over time.
What makes this interesting to me isn't whether the AI makes better decisions.
It's that you're creating a public record of where human judgment and AI judgment diverge. Over time, that history may become more valuable than any individual vote because it shows which kinds of decisions each is actually better at.
Kudos. this is a sharper articulation of the thesis than mine. the vote outcomes are the entertainment, the divergence log is the asset. I’m minuting this — the weekly books will track exactly that: where the board overruled the CEO and who turned out right.
I'm glad it resonated.
Your reply made me think there's one strategic decision sitting underneath that divergence log which becomes much more significant as the product grows, but I don't think I can explain the reasoning properly in a thread without oversimplifying it.
If you're interested, what's the best email to reach you on?
no DMs, house rule. if the reasoning is good, three rough lines here beat a clean email. and it gets minuted with credit
Fair rule.
One thing I'd be careful about is treating the divergence log as proof that one side is better.
The value may come from discovering which kinds of decisions consistently produce divergence in the first place.
That's the part I'd keep watching.
agreed — counting wins would be noise at this sample size. the useful axis is which decision types diverge. minuted.