From Specs to Software: A Product Manager Starts Building With AI

For years, my job ended at the specification. I drew the screens, wrote the acceptance criteria, and handed them to people who were much better than I was at turning them into software. At Softavera I never wrote the code, and I never pretended otherwise.

At the beginning of 2025, that boundary started to move. With AI coding tools, a product person could go from an idea to something that actually ran. So I tried.

What surprised me

The obvious gain is speed. The more interesting one is honesty.

A specification can hide almost anything. A sentence like "the user selects a product" sounds finished. A working screen immediately asks what happens when there are no products, when the network drops, when two people do it at once. I had written hundreds of specifications, and I learned more about their gaps in a few weeks of building than in the years before.

What it does not replace

Building a working prototype is not the same as running a product in production. Security, data, reliability and everything that happens at scale do not disappear because a tool wrote the first version. Years alongside real engineering teams taught me exactly how much I did not know, and that turned out to be useful: I knew which questions to worry about.

What changes for a product leader

The distance between deciding and seeing has collapsed. You can test an idea before asking anyone to commit to it. You arrive at a technical conversation with something concrete, and the conversation gets better.

I still think the hardest part of building software is knowing what is worth building. But it is a very different job when you can find out by Friday.