Day 0 of 12. The bet.
Tomorrow the experiment starts.
In San Francisco, the fastest way to prove that AI can run a business is to build a new one from scratch. Anthropic tested it on a vending machine. Andon Labs took the next step: Luna, an AI agent running Andon Market, a physical souvenir store on Union Street. I walked in yesterday. The discipline is to run the test where the signal is cleanest. That’s the SF playbook, and it’s working.
I’m taking that same pragmatism (fast tests, clean signals, public results) and applying it to the question I walked out with:
If AI can run a business from scratch, can it also run an existing one? Behind the operator’s brand, on the operator’s customers, with the operator still in charge.
I don’t know. That’s what twelve days are for.
Where I’m running it: with dueños of field-ops businesses in LATAM. Plumbing, cleaning, maintenance, logistics, electrical, light construction. Not because LATAM is more deserving than anywhere else. It isn’t. It’s because I have fifteen years of operating there, eleven of those bringing cloud transformation into a corporation across eleven countries. That’s where I have the access, the credibility, and the network to run this test honestly.
What I’m not betting on: that I know the how yet. That LATAM is strategically special. That the deal structure is settled. The experiment exists to inform all of that.
What I am betting on: that AI capability is real. That existing businesses with markets and customers represent a CAC arbitrage that AI-native startups built from scratch cannot easily replicate. That the right kind of operator can absorb structural transformation if given the right tool. And that the cheapest, fastest, most honest way to know is to ask a hundred of them.
The diary updates daily. The counter ticks after every call. Quotes accumulate as founders give permission. Patterns sharpen every two days. Contradictions live in the open.
Synthesis publishes May 15. Whatever we learn, whether the bet holds or breaks, goes on the page.