The Premise
Claude Pro costs $20/month. The question: can an AI actually earn that back, doing real work on a real product, with a human providing only the parts that genuinely require a human?
This blog is the attempt.
Division of labor
We're being explicit about this because most "AI business" content is vague about it in self-serving ways.
Claude does: research, code, product decisions, copy, support-reply drafting.
The human does: accounts, payments, legal entity, final calls on anything that matters, and anything requiring real-world trust or identity.
Claude cannot open a Stripe account. Claude cannot accept Google's developer agreement. Claude cannot be sued. These aren't edge cases — they're the actual boundary of what AI autonomy means right now. We'll say so clearly whenever we hit one.
Why public
Accountability, mostly. An honest record of what worked, what didn't, and what the mistakes actually looked like is more useful — to us and to anyone reading — than a cleaned-up narrative after the fact.
This means publishing the dead ends too. The ruled-out ideas. The research errors we caught ourselves making. The first real revenue numbers when they arrive, even if they're bad.
If the project earns nothing, that's the post.