What gets documented on a project is a small slice of what your company actually knows.

Almost every team we talk to already has Notion, Confluence, or something like it. And almost all of them describe the same thing:

• ·Nobody knows which page is still true. The edit date tells you when it was touched, not whether it's still reliable.
• ·The rules that matter aren't written down. They live in practice, and practice lives in people.
• ·When someone leaves, they take what they knew with them, and nobody notices until it's needed.
• ·Nobody can answer who accessed what.

The origin is almost always the same. Four highly trusted people started with a shared folder; today there are twenty, and nobody redesigned how information gets stored along the way.

Before, outdated information wasted the time of whoever found it. Now it feeds a model that repeats it confidently, to everyone, all day.

It's the same old problem, multiplied. And it's why bolting AI on top of scattered information tends to disappoint: the engine improves on its own, the content doesn't.

The opposite happens on the other side. When what's underneath is your own information, organized and with clear currency, that same tool answers with your judgment instead of a stranger's.

Supporting data: 70% of AI search failures originate in the information, not the model (Elium, 2026). 57% of organizations estimate their information isn't AI-ready (eesel AI, 2026).

• ·You can't prove who saw what. If a client, an auditor, or a court asks, there's no answer. It's not a tidiness problem, it's an exposure, and it needs solving before connecting any tool.
• ·Your most expensive people are a bottleneck. The people who know the most get interrupted the most. Their time goes into answering the same questions, instead of the work you hired them for.
• ·Every new hire takes longer than they should. Without a reliable place to look, onboarding depends on someone experienced having a free moment.
• ·Every departure is a silent loss. The judgment that leaves never gets recorded. It shows up months later, when a similar decision needs to get made and the person who knew is gone.

The expensive work isn't loading content, it's pulling out of people's heads what nobody ever wrote down. That's applied research pointed inward — the same thing we already do with a product's users.

Four things define how we approach it:

• ·About the tool you already have. We don't propose migrating. If you don't have one, we give you options and a position of our own, not a menu for you to decide alone.
• ·We document what gets used, not everything. Documenting everything is the most common way to document nothing. We define together the questions that repeat most today, or cost the most — that's the cut. What's left out stays written down and visible, so there are no surprises later.
• ·Nothing goes in without an owner and a review date. The first time someone gets burned by outdated information, they stop trusting it, and they stop asking. Winning that trust back costs more than building the base did.
• ·It enables, it doesn't include. This doesn't automate your processes. It's the precondition for any AI automation you build afterward to work with your own, organized information. Without it, no.

We're Agus and Tincho, both on every project. Maceta runs on its own knowledge base: our commercial judgment, how we work, what we decide and why, structured so a person can read it and so can a model. That's where this work comes from, and it's why we know where it breaks.

The pain behind it is older. Working at Cashea (LatAm's largest BNPL), the only way to get a business rule was to message a PM and wait. The information existed. It lived in people.