I completely agree with a recent post by Raphael Oliveira.

His point works because it hits something real: too many people have turned setup into a trophy. Workflow became the deliverable. Automation became performance. Someone builds a futuristic pipeline that reads email, checks the calendar, prioritizes tasks, turns on LEDs, plays saxophone, makes coffee, and still doesn't do the thing that actually needed to be done.

I laughed when I read it because it hurt. Not because I built some agent workflow that runs Wall Street out of Notion, but for a different reason: I am exactly the kind of person who keeps piling up half-finished projects.

I've always had this habit of saying, "when I have time, I'll finish it."

But the truth is simple and humiliating: we never have time.

So I decided to make the time, one way or another.

I got tired of using lack of time as an elegant excuse for projects I had abandoned halfway through.

And this is where AI came in, at least for me. Not as a productivity toy. Not as cosplay for some CTO from the future. Not as a perfect workflow.

It came in as leverage.

I still don't have the perfect setup. I'm still learning. There's still improvisation, friction, rework, and plenty of janky bits in the process. But now there is an objective difference between "I'll do it someday" and "this is out in the world."

Over the last three months, a lot of things actually got done.

---

At work

On the professional side, the win wasn't "look how beautiful my workflow is." It was actual delivery.

• I restructured two important application modules while we were still shipping new functionality. The result was better UX, better architecture, and fewer bugs.
• We put AI to work automating an extremely annoying manual step in the product: reading PDF documents that used to create a lot of friction for users.
• I finally managed to stand up a serious test suite for the main project, covering the backend and both frontends. We now have unit, integration, and E2E tests. We are already past the two-thousand-test mark, and that has started to make a very concrete difference in confidence and delivery speed.
• I documented the key application processes.
• I structured the Context Engineering architecture I've been using at work, which I explained in another article here on the blog in Portuguese.
• New features now ship with a more complete package: documentation, tests, and an API contract when it makes sense.
• I'm also modernizing a legacy client site, updating dependencies and preparing the ground for future evolution.

None of this is sexy. And that's great.

This is real work. The kind of work that improves the system, reduces friction, cuts rework, and lets the team breathe a little better afterward.

---

On side projects

If AI helped me gain traction at work, on side projects it helped me stop leaving everything in permanent draft mode.

https://josenaldo.com.br

On my personal site, I finally got my own house in order.

I updated dependencies, finished the content for every section, revised how skills are displayed, updated my resume, and organized the download flow. And maybe most importantly: the blog finally became what I wanted it to be.

It's becoming more technical, more useful to me, and closer to what I wanted it to be: a kind of external memory where I can organize my thinking and leave behind notes I know I'll want to revisit later.

I also put my career database online, my brag document. It isn't perfect. But it exists. And in this case, existing matters a lot more than endlessly idealizing the final version. I tracked my projects, courses, experience, skills, and achievements. That alone is already valuable to me, even if it is still far from the ideal format.

https://livropog.com.br

The POG Book finally made it out of the "I'll finish it someday" limbo.

I finished the entire first draft, cleaned up the site, and migrated the project to a newer version of Next.js. As part of that, I moved it from the Pages Router to the App Router. I also added EPUB and PDF versions and improved both AEO and SEO. It isn't the dream final product yet, but it stopped being a vague promise and became something real.

At last, we have a tome to guide us toward a brighter future, where Gambiarra will be recognized as the art form it truly is, and POG will become the highest form of expression for the Brazilian developer.

https://pogometro.com.br

If a book about POG was already fun, now we have an app to certify your project's POG.

The POGometer is a gamified system for measuring a project's level of POG, based on Gambiarra principles, techniques, and design-pattern-like hacks. It's a simple web app where you can paste the address of your GitHub project, let it inspect the code, extract relevant metrics, compare them against a database of projects and POG patterns, and get a diagnosis of your project's POG level.

It also has a project wall, a POG ranking, POG levels, scoring, and validation based on principles, techniques, and gambi design patterns. I still want to evolve it a lot, but it is no longer just a funny idea written down somewhere.

My wife's website, or a mysterious framework?

Maybe the project that excites me the most right now is my wife's website, because it is starting to become more than "a website."

Using everything I learned from client sites, my own sites, and the structure I've been refining across this ecosystem, I started turning the project into a reusable base.

We migrated the code to TypeScript.

The site now has i18n.

We separated content, configuration, collections, strings, theme, and images into a single structure so other people can spin up their own site more easily.

And now I'm integrating it with Sveltia CMS so you can still have an admin panel on top of a static site.

The ambition here is interesting: turn this project into a kind of low-cost headless CMS with an admin panel that can be hosted on GitHub Pages or GitHub + Vercel, so an individual person or a small NGO can have a fast, simple website with maintenance costs close to zero.

That is already far more interesting than letting another repository quietly rot.

---

The point isn't the workflow. It's what the workflow unlocks

So yes, I agree with Raphael.

If AI is only helping you stage a little theater of productivity, then it has turned into an expensive hobby.

But I also think there is another side to this conversation.

For people like me, who kept accumulating unfinished things and hiding behind "I'll do it when I have time," AI can be exactly the tool that reduces enough friction for you to finally do the work.

It didn't do the work for me. It didn't inject discipline into my bloodstream. It didn't install focus directly into my prefrontal cortex.

What it did was reduce the weight of starting, unblocking, reviewing, organizing context, keeping consistency, researching faster, documenting better, and closing gaps I used to keep postponing.

And honestly, that's fucking great.

---

I didn't build Skynet. I got my house in order

I didn't build a revolutionary SaaS. I didn't invent a new AI to lift humanity to a higher plane. I didn't build some command center that controls my schedule while I sip coffee in front of a neon dashboard.

What I did was mundane work.

I got my house in order. I pulled projects out of limbo. I documented systems. I wrote tests. I refactored modules. I published content. I turned promises into artifacts.

And honestly, I'm pretty satisfied when I look back and see that a lot of rooms are cleaner, more organized, and more livable than they were three months ago.

I still don't have the perfect workflow. But now I have tangible results.

And that, for me, is already more than enough reason to keep learning, refining, and using more AI.

Craudio, Gepeto, Carona, and Germano, thanks for the help. I'll get each of you a chocolate truffle by the end of the year.