Building was expensive. One wrong bet could waste a quarter. So we front-loaded certainty: weeks of research before greenlighting anything.

When a prototype takes hours, not months, you build to learn, not learn to build.

One-pagers, six-pagers, PRDs, design docs. The spec proved you understood the problem. Writing it was the thinking.

But a working prototype communicates intent at higher fidelity with fewer assumptions than any document can.

When every feature costs months of engineering, "no" is your most important word.

When prototyping costs almost nothing, "no" prevents learning. Instead of debating X vs Y for weeks, prototype both and let users decide.

Most treat AI like a search engine. Prompt, get answer, move on.

But when you capture team patterns as structured context, quality jumps. It goes from a tool to a teammate that compounds.

"What does the data say?" became the default answer to every product question. Data was the tiebreaker, the decision-maker, the shield.

Data is essential for validation. But it can't tell you what to build next. That takes a point of view. The best PMs use both, but they start with judgment.