As a programmer, artificial intelligence has always fascinated me. I remember at my first job, during lunch breaks with coworkers, we used to fantasize about what the future of programming would look like. One idea that kept coming up was being able to talk to computers using natural language—programming without barely touching the keyboard. It felt like science fiction.
You could say that moment has arrived. Today it's possible, using a microphone and tools like Claude or ChatGPT, to specify tasks and request changes on the fly. This is called "Vibe Coding": programming by describing what you want instead of writing it yourself line by line.
But before we get carried away with enthusiasm, it's worth pausing. Expectations are far above reality, and you shouldn't listen to those who have financial interests in inflating those expectations, nor to those who share opinions without ever having built a software project from scratch.
My First Steps: Text Generation
Since the first GPT came out, I started experimenting with text generation. As a programmer, creating written content for clients and my own websites has always been the task I enjoy the least and—why not say it—the one where I feel least confident.
I started testing it, but the results left a lot to be desired. You always had to edit, fact-check, and fix the tone. GPT kept improving with each new version and ChatGPT was a major leap forward, but to this day I still read, review, and correct every piece of text it generates.
The Jump to Code
Later came the ability to generate code and, as an early adopter, I dove right in. Something similar to what I experienced with text happened: ChatGPT would respond with total confidence, assuring me its code was perfect and functional, but the reality was that most of the time you had to spend hours fixing it, giving new instructions, starting from scratch, and rewriting things to actually make them work.
Claude Code: A Before and After
In May 2025, Anthropic publicly launched Claude Code. At first I didn't have much faith in this tool, but after hearing and reading about the positive experiences of other developers, I started using it in December of that year.
The difference compared to what I had tried before was remarkable. Not perfect, but a real qualitative leap in both speed and quality of the generated code.
The Copilot Metaphor
Microsoft got it right when they named their AI "Copilot." The concept is clear: the user is the main pilot of the ship and the AI is the copilot. The pilot steers, can delegate tasks, but is ultimately responsible for the outcome.
Talking with other programmers and professionals in the digital ecosystem, this metaphor perfectly reflects the current reality. AI doesn't replace the human—at least not one hundred percent. What it does is perform at high speed the work that millions of programmers have done over and over again manually for years: writing a form, setting up a database structure, creating repetitive components. It does it much faster, but there always has to be a human at the wheel.
What I've Learned So Far
After months of integrating AI into my daily workflow, my conclusion is that we're looking at an extraordinary tool—but one that requires an experienced pilot. Blindly trusting it leads to serious problems, and that's exactly what I'll talk about in my next article: the real mistakes I've made and the lessons they've taught me.
AI has changed the way I work. Not quite the way we imagined during those lunch breaks, but in an equally fascinating way.
// no comments yet
// sign in to comment with your account