The audacity of not knowing the rules

There's a Spanish artist who goes by C. Tangana. Born Antón Álvarez, he started as a rapper, reinvented himself multiple times, and in 2021 released El Madrileño — an album that nobody saw coming. No trap, no reggaeton, no attempt to copy American sounds. Instead: flamenco, bolero, copla, and a production sensibility that felt genuinely new. In a market dominated by imitation, he chose to sound like something that had never existed before.

As he says himself — "without being able to sing or stay in tune" — he managed to get a whole country to stop and listen.

A while back he gave a talk at the Fundación MOP with Santos Bacana about his creative process. His words hit differently when I applied them to building digital products with AI. Here's what I took away.

1. The audacity of ignorance

Antón talks about ignorance as, paradoxically, one of an artist's most sophisticated tools. When he decides to direct a film or compose with flamenco masters without being an expert in cinematography or harmony, what you see is someone who isn't afraid of what they'll find.

In software, the opposite tends to happen. When you try to build something new — especially with AI — the technical establishment shows up with its mantras:

"The code is bad."
"This won't be maintainable."
"We're generating technical debt from minute one."
"We're too dependent on the model."

Honestly: all of that already happens in the current production model anyway. The difference is that now, the fear of "doing it wrong" becomes the blocker. The key is to jump in, stay curious, and if it falls apart, let it fall apart. A broken thing that exists beats a perfect architecture that never ships.

2. The danger of "doing it right"

When you know exactly how something is supposed to be done, your brain takes the path of least resistance: what's already been proven. Knowledge gives you confidence, but it also boxes you in.

As Antón says, acting from a naïf or raw place allows the mistake to become a new aesthetic proposal. Maybe that's where the future of software lies: a phase where the standard matters less and radical personalization — even if the underlying code is "ugly" — becomes the norm.

The expert's trap isn't incompetence. It's over-competence. Knowing too many reasons not to do the thing.

3. AI as the dirty workshop

Antón describes his studio as a workshop: messy, full of worn-down tools and unfinished things. Brilliance appears in the friction, not in the efficiency.

He says he needs 9 hours of studio time "thrown in the bin" for hour 10 to produce something worth keeping. What AI gives us is a way to throw those 9 hours in the bin in 2 seconds. Not a perfect assembly line — a tool that lets you swing in the dark faster, and reach hour 10 before you run out of steam.

That reframe matters. If you treat AI as an efficiency machine, you miss the point. It's not there to eliminate the messy work — it's there to help you do more of it, faster.

4. The MVP: necessary mess

The Minimum Viable Product is the software equivalent of Antón's dirty workshop. Something you launch with the explicit purpose of learning, not of arriving.

The expert's trap: An engineer hates the MVP because they can see the seams and the shortcuts. It physically hurts.
The creator's reality: If you wait for the code to be a work of art, the window has already closed. As Santos Bacana puts it: "If it's bad, nobody's going to notice."

The MVP is your safety net. If it has duende — real utility, or genuine emotional resonance — users will ignore the rough edges. What matters is what it does for the person using it, not how it was built.

5. Consciously choosing the beginner's position

There's a difference between being ignorant and choosing the position of the beginner. Antón knew Spanish urban music was about to explode and decided to occupy the space. He didn't wait to be crowned anything — he acted like it until the rest of the world caught up.

This is more than fake-it-till-you-make-it. It's understanding that success is an anomaly inside a routine of failed attempts. Not a divine revelation at minute one. The result of enduring the frustration of swinging in the dark until the thread worth pulling finally shows up.

To reach project 11 — the one that works — you had to go through the 10 hours of waste first. The difference now is that AI lets you burn through those 10 hours faster, more consistently, and with more to show for it each time.

Knowledge as filter, not as brake

Antón tells a story about classical guitarists who said flamenco players were "playing wrong" because they didn't follow the rules. In AI-assisted development, people will tell you your way of building is a mess because it doesn't follow classical patterns.

But if that messy code carries a disruptive idea to the user — the way El Madrileño reached its audience — then what was "done wrong" becomes "done right."

The modern developer's role is shifting: less executor of every line, more someone who feels the system — detecting where there's a hallucination and where there's an opportunity. Moving from executor to curator.

The question isn't whether your code is beautiful. The question is whether it does something worth doing.