From Cubicle to Lab: My (Pseudo)Scientific Method for Building Startups with AI

If you scroll through LinkedIn, it seems like everyone has the magic recipe for building a billion-dollar AI startup. I don't have the absolute truth, and I'm not here to hand out advice. Everyone has to find their own way through an ecosystem that changes every other Tuesday.

All I can share is what works for me, and how I went from being one more part in the corporate machine to running my own testing lab.

From Consulting to the Lab

For years, my day-to-day was consulting work: getting placed at big companies to work on massive projects. You're surrounded by bureaucracy, huge teams, and processes where pushing a single comma to production takes three meetings and an approval committee. You're a gear in a very heavy clock.

When I decided to start building the projects that today make up Oihu, I realized I couldn't — and didn't want to — replicate that monstrous model. I also didn't want to play the "visionary CEO" game, spending my days managing egos instead of building.

What I've done, thanks to orchestrating AI agents, is turn the way I work into a kind of scientific — or pseudo-scientific — method applied to software.

The Method: Hypothesis, Experiment, Data

Validating a business idea used to be an expensive act of faith. Now, my agents let me treat every idea as a low-cost lab experiment.

The process I follow is boringly analytical:

1. The hypothesis: I spot a problem. A friction point in a legal validation flow, say, or an unmet need in the audio world.
2. Designing the experiment: instead of writing a five-year business plan, I sit down with my agents and design the minimum architecture needed to put that idea to the test.
3. Fast execution: thanks to AI, what would take a large corporation six months to build, we code, iterate, and ship in a matter of weeks or days.
4. Collecting data: we launch. If people use it, the hypothesis holds and we keep investing effort. If not, we close the experiment, write down what we learned, and move on.

There Are No Magic Formulas

The real revolution AI brings isn't that it does the work for you, or that it guarantees success. The revolution is that it has drastically lowered the cost of being wrong.

I can formulate ten hypotheses and build ten working prototypes for the same effort and money it used to take me to make a single PowerPoint deck for an investment round.

I don't believe in the gurus selling "the definitive method" for building a startup. There are no shortcuts. But if you have an analytical mind and treat Artificial Intelligence as your lab instruments, you can test reality at a speed that was science fiction just a couple of years ago.

I'm still in the lab, trying things out. Some explode, some compile. That's what this is about.

Ez dadila haria eten.