Building the RVM ecosystem — part 0 (the origin)

AI saved my career

From almost giving up on tech to two systems in production and a new job working with AI

· 6 min read

Yes, I thought about giving up on my career.

And I don’t mean a Sunday-night crisis. I mean the day I caught myself filling out the sign-up form to become a rideshare driver, thinking that maybe tech was over for me.

But to understand how I got there, I need to go back a little.

A truckload of money

At the end of 2025, I got a very good offer: working on a system that regulates sports-betting rules across several countries. High salary, big project, a nice name on the résumé.

But not everything is roses. What made me leave doesn’t matter here, but I left. I walked away from a truckload of money to keep my mental health. I would do it again.

And that is how, for the first time in many years, I found myself unemployed.

The confession I will deny to my grave

If you ask me this in person, I will deny it to my grave. But here it is: we in tech think we are rock stars. There is always an opening, always a recruiter in the inbox, always someone who needs a developer.

What we forget is that, to be in demand, you have to be at the cutting edge. And that was my mistake. I spent almost seven years at the same job and thought my life was set. I learned new things when I needed something, not because I needed to stay relevant. I let my mind settle.

When I went back to the market, the market had moved on without me.

Twenty-five “no”s

Rejection after rejection. Sometimes not even a rejection: just silence, which is very common in our field, but it hurts all the same.

Twenty days went by and the worry started to set in. After twenty-five interviews, I no longer believed in myself. Every “unfortunately, we decided to move forward with another candidate” seemed to confirm what I was already thinking: I had been left behind.

That is when I opened the rideshare sign-up form.

Thirty days

Before finishing that form, I remembered a system I had been thinking about for a while: a tool for small farmers to keep track of their crops, costs and sales, in a way they could actually use, on their phone.

And I set myself a goal: finish it in one month.

It was an absurd goal for one person. To hit it, I would have to move faster than I ever had. So I did something I had always looked at with some suspicion: I brought artificial intelligence in to work with me.

I started in the simplest possible way, with ChatGPT in the browser. I asked for a file, it handed it over. I copied, pasted, tested, came back with the error. Asked for the next one. It was manual labor, but it worked. And with every file that worked, something fell back into place: my confidence.

With confidence came the urge to improve the process. Copy and paste was no longer enough. I moved to the command line, and from there to Claude Code, an AI that didn’t just suggest code but read the whole project, ran the tests, found the bug and fixed it, while I decided what to build.

It was like trading a shovel for an excavator.

Before the deadline, the system was done. Today it is called Safra em Dia and it is far more robust than that first version: it traces coffee from the field to the sale, warns farmers before they sell below cost and accepts entries by voice message on their phone.

My father’s spreadsheet

A few days later, my father complained about the lack of control at his work: job sites, machines coming and going, crews moving between work fronts, all in a spreadsheet nobody could keep up to date.

I already knew the way. In a few days, I had another system live: Obra em Dia, which tracks machines and people on construction sites and blocks assigning anyone with expired documents before the problem reaches the site.

Two systems in production, built by one person, in a time frame I would have called impossible a year earlier.

The “yes”

Then, a few days later, came the “yes” I had been waiting for through twenty-five interviews.

I landed a new job. And not just any job: today I work with AI. The same tool I had reached for in a moment of desperation became the center of my profession.

I don’t think that was a coincidence. I didn’t walk into that interview saying “I know how to use AI”. I walked in showing two systems in production and explaining how I had built each one. AI didn’t get me the job. It gave me something to show.

What the new job taught me (and I took home)

The best part came afterwards. At work, I started spending my days with people who use AI every day, at scale, with a process. And I learned a lot, fast. Much of it I brought straight into my own environment:

  • Everything on Linux, even on Windows. I moved my entire environment to WSL, the Linux that runs inside Windows. AI agents work much better in a Linux terminal, and now my computer and my servers speak the same language: same scripts, same tools, same behavior. No more “works on my machine”.
  • AI with a defined role, not a chat that does everything. I split the work into roles: one agent plans, another codes, another reviews (ideally a different model, one that won’t just confirm the author’s logic), another tests in the browser. Each with its own permissions.
  • A single source of instructions. Every project has one file telling the AIs how to work on it, and that file works for any tool, whether Claude, Codex or another.
  • The machine verifies, I decide. Everything that can be checked automatically (tests, deploys, whether the system came up, the screenshot) runs without asking me. What is left for me is what only a person can judge. And production only gets changes with my sign-off.

What were two systems became an ecosystem: more than twenty repositories following the same standard, four servers and a “team” of AI agents that I orchestrate myself. I’ll explain how it all works in the next post.

If you are where I was

If you are at interview number twenty-five, with the rideshare sign-up open in another tab, I want to tell you three things:

  1. The market didn’t forget you, it just moved on. And you can catch up. Faster than it seems.
  2. Stop sending only résumés. Build something. A system in production speaks louder than any list of technologies. Today, with AI, one person can build what used to take a team.
  3. Use AI as a lever, not a crutch. It sped up my work in ways I never imagined, but deciding what to build, for whom and why was still on me. That is the part nobody automates.

I almost gave up. Good thing that, before finishing that sign-up form, I opened the terminal.

Rafael Veneroso Morici · rvmtech.com.br