How I use AI
As of September 27, 2026, this is how I use generative AI and LLM tools.
I have a paid Claude subscription, which I use to direct my questions, both on my mobile and through their website. I think of Claude in the same way I think of any other AI: as a thinking partner.
I think of personal websites like my own in the same way that the word personal itself suggests: as personal attempts at expression.
I once read Cris Dias talking about the meaning of the word essay (ensaio, in Portuguese). It comes from the French essais, or attempts. It was Michel de Montaigne who popularized the term around 1570. Cris points out that this is also why, in Brazil, we use the same word for a theatrical rehearsal: “this isn’t exactly ready yet, right?”.
I therefore declare that everything written here — whether it is a blog post or a note shared in my digital garden — is an essay. A big “So, here’s the thing: I was thinking that…”.
It carries my opinion, my personal view, my personal interpretation of things — never those of Claude or any other AI.
Ethan Mollick proposes an interesting way of deciding when it is worth using AI: the question should not be “is AI better than a human?”, but rather “is the best AI available right now better than the best human I can actually reach, for this specific problem, at this specific moment?”.
I like this formulation because it is very honest. The real comparison is never with some idealized expert, but with whoever is actually available. I don’t have an expert on call at 11pm when an idea comes out of nowhere for a note that I want to investigate and develop. That’s where Claude comes in: not as a replacement for the best available human, but as the best thinking partner I can reach at that moment.
While I write, I might ask questions such as “how many flags are there in the world that aren’t rectangular?” or “how does an astronaut sleep during a space mission?”.
And images?
Images are a little different.
I am not a professional illustrator, cartoonist, or visual artist. Sometimes, though, I have an idea for an image that I would like to put alongside something I have written. Generative AI gives me a way to turn that idea into an illustration when I don’t have the skills necessary to draw it myself.
Stephen Wolfson, writing for Creative Commons, described an experience that comes quite close to my own. He writes that he had little talent or training in the visual arts and had never been able to draw what he had in his head. With generative AI tools, however, he was able to turn his artistic vision into something concrete.
That’s a way of thinking about image generation that makes sense to me.
This does not mean that I consider myself an illustrator, nor that I think generative AI replaces illustrators, cartoonists, or other visual arts professionals. It simply means that, in the context of this personal website, I sometimes use image generation as another creative tool: I bring the idea, the concept, the composition, the words, and the judgment; the model helps me turn them into a visual artifact.
I also try to make it clear when an image has been generated with AI.
And programming?
I have been working with information technology, or things related to it, for more than two decades: managing IT projects, writing and reviewing business cases, brainstorming, and designing processes.
But I no longer write code by hand the way I used to (except for the occasional VBA macro to automate something that shouldn’t need to be automated in the first place). The logic stayed, though — and that’s what I rely on when I decide to do vibe coding.
Still, I don’t want to give the impression that I’m vibe coding all the time, because I’m not.
And, to be clear: this website is about essays, prose, and ideas — not about programming.
If that changes in the future, I’ll update this page, which will remain at this permanent URL: /en/how-i-use-ai.