Building Ciente made me realize the product is not the AI I’m building Ciente, a mobile-first investment platform for Brazilian investors. The idea started with a pretty practical frustration: people have investment apps, statements, bank portals, PDFs, advisors, and still often can’t answer basic questions about their own money. How much did I really make after taxes? What was just money I moved around? What did I pay in fees? Is this new option actually better, or just being sold better? At first, I thought the hard part would be connecting the pieces: Open Finance data, reports, an AI assistant, and eventually a marketplace of independent CVM-registered consultants. But the more I built, the more I saw that the product only works if the boundaries are clear. The AI cannot be the source of truth. It should explain, compare, and help the user ask better questions, but the actual financial facts need to come from deterministic calculations. That changed how I approached the whole product. A deposit is not yield. A sale proceeding is not performance. Taxes cannot be guessed by a language model. If the report is wrong, the AI becomes confidently wrong. So the boring parts became the important parts: classification rules, report versions, data isolation, auditability, and very explicit limits on what the AI can and cannot say. It also forced me to be more honest in the positioning. It would be easier to call Ciente a robo-advisor or say the AI gives recommendations, but that is not the current product. Right now the principle is: reports calculate the facts, AI explains them, and the person decides. If someone wants individualized professional advice, that belongs with a properly registered consultant unless a future regulated path is built for automated advice. The biggest lesson so far is that in finance, trust is not a brand word. It is a product constraint. It affects copy, architecture, logging, data storage, AI prompts, consultant flows, and even what features I choose not to build yet. A lot of the work has been removing things that would make the product sound more impressive but less honest. Ciente is still early, but building it has already changed how I think about startups in regulated spaces. The hard part is not adding AI to finance. The hard part is making the system useful without pretending it can safely decide for the user.

Ciente

A Ciente separa análise, explicação por IA e decisão.

A Ciente separa análise, explicação por IA e decisão.