A company brain answers questions about your company's documents, with the source cited

Because an uploaded file is still a file. It is the test almost every company has already run: upload the folders to an assistant, ask it three questions and find four things. First, that the citation is vague —«according to the contract»— and vague does not hold up a decision. Second, that two similar questions bring back answers that contradict each other, with no way to tell which one is right. Third, that the old version of a document wins with the same confidence as the current one, because nothing told the system which one rules. And fourth, the most expensive: when the answer is nowhere, the system builds it anyway. Making sure those four things do not happen is the work, and it takes most of the project. It is not a setting or a tool you install: it is the state the information has to be left in before it is worth asking anything.

The difference is what it read, and whether it can tell you where the answer came from. A general assistant answers from what it read on the internet up to its cutoff date, and when it finds nothing, it fills the gap. A company brain answers from your company's material, and every statement arrives with its origin pointed at. The consequence is not the quality of the writing, which is good in both cases. It is that an answer without a source cannot be taken to a committee, cannot be defended in front of a client and cannot be audited six months later. There are two more differences, and they weigh at the moment of hiring: your documents do not go out to train anyone else's model, and what the brain produces is delivered in an open format, readable without TintoAI.

Skills: the judgement of whoever knows, running on the brain's sources. The brain answers questions; skills do complete tasks and return a finished deliverable. The regulatory brain TintoAI operates has 10 years of experience put into skills that answer technical queries, classify risk, assemble case files and produce specialised content. Each one works with the judgement of the expert behind it, and none delivers anything without that expert approving it. Inside a company the arrangement is the same: each area builds its own on top of its own brain, and on top of those connect the tools that area already works with. What is promised is measurable and written down before starting: hours per process, cost per process and how many steps stop being done by hand. That baseline is what later decides whether the brain was worth it.

A person, always. The brain delivers the answer worked out and with its citation; whoever knows the subject confirms it, adjusts it from experience and approves it. That step is not a leftover to be removed over time: it is part of the design, and it is what separates this from buying automation with nobody answering for it. The reason is practical. An answer with its source beside it is reviewed in seconds, because reviewing stops being research all over again and becomes checking a citation. Without a source, verifying costs the same as doing the work again, which is exactly why companies abandon their first attempts with artificial intelligence. The brain also warns when a source has gone months without review, instead of answering confidently from stale data.

What leaves with that person is the judgement, not the files. The documents stay; what disappears is the why: what was looked at before deciding, which exception was accepted and on what argument, which of two paths was discarded and for what reason. That is rarely written down in a folder — it lives in the head of whoever decided and in scattered emails. A company brain captures that judgement while the person is still there: the decisions they made, with the document behind them and the date. When they leave, whoever arrives does not start from zero or repeat an analysis already done; they ask and receive the earlier decision with its reason and its origin. Knowledge stops being a dependency on one individual and becomes an asset of the company, one that can also be audited.

No case repeats itself exactly, but none has to start from zero. Every company piles up closed files, won quotes, handled complaints, audits passed. Today that material gets archived: kept just in case, never consulted, and when a similar one arrives it is solved again from the beginning because nobody remembers where the previous one ended up. Inside a company brain that archive becomes the part that answers most. Faced with a new case it brings the precedents that resemble it —what was decided in each, on what argument and how it ended, with its source beside it— and with that most of the work arrives already done. What is left is what genuinely sets the new case apart: its particulars. That is new material, the expert resolves it, and the brain takes it in, so that the next similar one arrives with more of the road already walked.

With the map, never with the brain. No company starts by building the central brain. You start with the knowledge map, which answers which documents exist, where they live, who maintains them, how often they change and which ones have already expired. That map is worth having on its own even if the company goes no further, and it is what says what is worth including and what is not, because putting everything in is the fastest way to make a brain answer badly. With the map in hand you pick the area that repeats the most questions and build its brain. Once two or three areas are working, connecting them is a short step, because the hard part is done. Moving up a rung is not compulsory: each one is hired separately and decided with the previous one's measurement in view.