If you have thirteen tabs open about AI and the feeling that you're not making progress on any of them, this is for you.
Two years ago, at a LATAM fintech, I had thirteen Chrome tabs open at the same time. One with an AI course promising to "transform your career in 4 sessions." Another with a Twitter thread about the latest model that had dropped that week. Another with the documentation of a tool that was going to change everything. And so on. Thirteen tabs, zero output.
The pressure cascaded down from the C-level. The phrase of the moment kept landing on the Slack channel: "we have to start using AI in the team." Nobody knew what to do with that. Neither did I. But the thirteen tabs were still there.
If you've worked at a mid-sized or large company in the last two years, what follows probably sounds familiar.
The problem nobody tells you about
The pressure from above was constant. We have to use AI, we have to implement it, we have to build something. But below, in the team, nobody had any idea where to start. The company offered us courses: four one-hour sessions with content so generic it didn't apply to anything you actually had to solve. You'd come out of one of those with a certificate and the same lost feeling.
The serious consultancy, the one that could actually help, cost twenty thousand dollars and delivered a ninety-page PDF nobody on the team read. And the YouTube and Twitter tutorials were like drinking seawater: the more you read, the thirstier you got, because every week a new model came out, a new tool, a new "best practice" that invalidated the previous one.
Marketers, product managers, ops folks, analysts, early-stage founders — non-technical profiles with a real problem and nobody on the right side to solve it. That was the picture.
I had a concrete problem: I needed to cut down the time it took me to put together reports, I wanted to speed up content generation, I needed to improve team follow-up for product GTM. Three hours a week, every week. I knew AI could solve it. But I didn't know how. And there was nobody who could explain it to me — someone who understood the problem well, but wasn't technical — without throwing jargon at me or a four-hour module.
What I needed was simple: someone next to me, two or three hours, on my concrete problem, who would explain why this and not that. Who would leave me with a working solution and, above all, who would change the way I thought so the next time I could solve it on my own.
That person, that space, didn't exist.
The day I closed the thirteen tabs
One day I got tired. I closed the thirteen tabs, grabbed the most urgent problem I had, and sat down to solve it alone with AI. It took me an afternoon. When I finished, I had learned more in those three hours than in six months of consuming content about AI.
And right then I understood something:
AI isn't learned by watching it. It's learned by building with it, on top of something you actually care about.
It sounds obvious when you write it down. But the entire industry is doing exactly the opposite. Courses, podcasts, newsletters, certifications, conferences — most designed for you to know more about AI, not for you to use AI. It's as if the only way to learn to drive was to listen to eighty hours of podcast about defensive driving, without ever getting in a car.
What it took me two years to understand
But there's something else I learned over time, that took me two years to understand, and that is the real thesis of this article.
When I started building more things with AI, I noticed a pattern. Loose solutions — a prompt here, a custom GPT there, a one-off workflow — worked for two weeks and broke. The ones that held up had something different: a structure underneath. An architecture.
I ended up calling it a file system. Not for the computing metaphor, but because that's literally what it was: a system of files.
Folders with clients. Folders with roles. Folders with templates. Folders with criteria. Folders with history. Each file written in a format the AI could read and interpret, in markdown — a format that has existed for years and continues to be the standard for documents of this kind.
When you ask the AI "draft me an email for client X," the AI improvises with the little you gave it in the prompt. When the AI has a file system next to it — clients/X/context.md, clients/X/history.md, templates/closing-email.md, criteria/communication-tone.md — the AI doesn't improvise: it applies.
The difference is enormous.
The first version is unstable magic. The second is an operating system.
And here's the point I care about most: that file system ends up changing the way you think.
When you organize your work as a file system designed for the AI to read, you start structuring problems differently. You start separating context from criteria. Templates from decisions. Stable information from variable information. That mental discipline — which used to be a luxury of very organized people — is now the difference between AI amplifying you or just adding noise.
AI isn't learned as a tool. It's learned as a structure.
And understanding this changes everything you think about how to learn AI.
Because if what you have to learn is a structure, not a tool, then no four-hour course is going to teach it to you. No ninety-page PDF. No Twitter thread. The structure is only learned by building it: sitting next to someone who already went through it, on top of a problem that matters to you, watching how the pieces get pulled apart in real time — what's context, what's criteria, what's a template, what's a decision.
That was what I needed two years ago, at the fintech, with the thirteen tabs open. And two years later, looking around, it still didn't exist.
That's why I created Mate & Build
A couple of months ago I created Mate & Build. The idea is exactly what I would have liked to have when I was sitting with the thirteen tabs open.
Once a week, someone brings a real problem from their work. Something concrete. Something that's eating their hours. Something they keep postponing. We sit down, we talk, we understand what needs solving. And we build together in two or three hours.
The person leaves with three things: a working solution to their problem, a replicable kit (the prompts, the code, the documentation), and a base file system they can keep using to build the next thing.
Each build stays public on mateandbuild.com.ar. The documentation, the prompts, the exact times, the things that didn't work the first time. Everything open so anyone can replicate it.
We've done seven sessions, three published, the next one ships this week.
Teams from Rufus, PedidosYa, and others have come through the table. With Franco, a marketer at Rufus, we built a system that took copy production from 45 minutes to 5. With Manu, at PedidosYa, we built a workflow that turned entire days of email assembly into ten minutes. With Lupa, who came with three different problems at the same time, we solved them in parallel in a single session.
But the numbers, even though they look good, aren't the most interesting part. The most interesting part is the face of the person on the other side when something that didn't exist two hours ago suddenly works. That "I didn't know this was possible today" is what reminds me why I built this. It's not marketing — it's the real reaction of someone who went from watching AI to using it, and noticed the difference.
The invitation
If there's something I learned these two years — first as a frustrated client at a fintech, then as the creator of the format — it's that most people are learning AI from the wrong side.
Not because they're lazy. Not because they lack talent. Because they're trying to solve with content consumption a problem that only gets solved by production. Because they're treating AI as a tool when it's actually a structure. Because they're waiting for the perfect course when all that's needed is to sit down, one afternoon, with a real problem next to you.
If what I'm saying sounds familiar — if you have thirteen tabs open, pressure from above, and the feeling that every week a new model leaves you more lost — then this article is for you.
Not so you take a course. Not so you read more threads.
So that one regular day you open your calendar, grab the most concrete problem you have, and sit down to solve it. Alone, with a teammate, or with me at a table over mate.
What matters isn't how. What matters is that you start.
Learn by Doing. 🧉