Build AI from LatAm. Don't apologize for the region — use it.

I started Darwin in late 2023, the moment GPT made it obvious that whatever we knew about software was about to be rewritten. Every founder I talk to in LatAm asks the same set of questions: what do you think about model costs, hiring, pricing, regulation. We don’t have all the answers — some of our early hypotheses were right, others embarrassingly wrong — but we have a way of looking at things, and that’s the part worth sharing.

These notes started life as the “AI Thoughts” page of our investor deck. I’m pulling them out of that PDF and writing them up here, in case they help any other founder.

The frame underneath all of them is one belief: build from LatAm, for the world, but understand what makes the region different and use those differences as the edge.

People hear “build from LatAm” and assume it means accepting a smaller TAM. It does not. Look at what happened with each prior wave. eCommerce in LatAm produced MercadoLibre — bigger and better than its US counterparts on key metrics. Fintech produced Nubank — a bank with more customers than any single bank in the United States. Food delivery produced iFood — operationally tighter than Uber Eats in the same geography.

Atlantico, a Brazilian fund, ran the numbers and called the pattern “Latam does it best.” Same wave, better results in our region, because the constraints that look like disadvantages from California — fragmented markets, harder distribution, weaker infrastructure — force LatAm operators into behaviors that turn out to be the right ones.

The same will happen in AI. The only mistake is to wait for a US version to copy.

So if you’re a founder building AI from LatAm: build for the world, but use what your region teaches you. The dynamics are different. Don’t apologize for them. Lean on them.

▮