The muscle of dissecting problems (and what's left when AI does the typing)

If there's one thing we do more than anything else as programmers, it's dissecting problems to find solutions. The more you train it, the better you get. I call it a muscle, because it wastes away when all you do is consume tutorials, and it gets stronger when you face a real problem, even one that isn't yours.

The habit I've always had

I've always dissected companies, business models and problems that cross my path. A product that annoyed me. An app I used and kept thinking about how I'd build it. A KYC service: how much it costs, where it gets stuck, who profits. Any company: how does it really make money? A new café in town: how much did they invest and what's the payback?

It used to be desk research on Google. Opening tabs, cross-checking sources, taking notes. It worked. It was slow. And, honestly, I did it less often than I should have, because the friction was high.

What changed with AI

Today I open a session in Claude or ChatGPT. Sometimes by voice. I get the idea out just to see what comes of it. I don't need a perfect brief. I need curiosity and a willingness to iterate.

This detail matters: the idea has to get out of your head. While it stays in there, it's just noise. You repeat the same mental loop without testing any assumption. In the session, the idea becomes an object. You can push it, cut it, negotiate constraints.

The solution doesn't always need to be software. Sometimes the exercise points to operations, to pricing, to a business format you'd never think of if you stayed stuck on "how would I code this."

The concrete case: a 400m² lot

Near my parents' house there's a parking lot. I looked at the lot on Google Maps, measured about 400m², and went to ask Claude whether a parking lot would work there.

How many spaces fit? How much does it cost to cover them? What's the monthly net? What's the payback? What does the Pouso Alegre master plan say?

There was no right or wrong. The value was in the exercise.

The whole session went like that: a generic prompt at the start, with it making assumptions on its own. When you don't say anything, it leans toward the most expensive scenario: covered spaces, high investment, tight net. Then you negotiate. Concrete or just gravel? Where do you leave a permeable area? What if it's 18 spaces? What about motorcycle spaces?

I asked for diagrams of the lot layouts. It looked up the land-use rule. We iterated until the monthly net landed somewhere worth discussing. At the end: total investment and payback, not counting the lot itself.

This works for anything. Send a photo. Show it on the map. Ask for a hypothesis. Adjust the constraint. Repeat.

Four triggers to work the muscle

1. A product that annoyed you, or one you loved. Dissect why.
2. An app you used: how would you build it?
3. A company: how does it make money?
4. A physical business in your city: how much did they invest and what's the payback?

These are entry points. The point isn't to become a parking lot expert. It's to keep an active stance: instead of consuming ready-made answers, you investigate.

Why this matters for people who code

With AI typing and creating code for us, what's left is exactly solving the problem: investigating solutions, raising hypotheses and choosing the most viable one. Solving at a high level. Delegating the execution.

In Simple SaaS we already worked this way: create hypotheses before building. Today the kit is the same, except Google now comes with a lot more AI. The speed of desk research changed. The muscle didn't.

If all you do is ask for code, the muscle wastes away. If you use the session for dissection (business, product, constraints, trade-offs), you train what the tool still doesn't do in your place: judgment.

With AI typing the code, what's left is the muscle of dissecting problems. Train it, even on problems that aren't yours.

In short

Dissecting other people's problems trains you to solve your own. It used to take Google and patience. Now it takes an open session, voice if you like, and iteration until the hypothesis holds up.

The 400m² lot didn't become my business. It became training: assumption, constraint, number, master plan, payback. No absolute right or wrong, but with the mental relief of getting the idea out of my head and testing it.

Keep that muscle sharp. When execution gets cheap, what sets you apart is once again choosing what's worth solving.

---

Watch the full video at O músculo que todo dev precisa treinar (e quase ninguém treina) (in Portuguese). There I show the whole Claude session on the 400m² lot, the everyday triggers and why this changes what's left for developers.