I ship the prototype before the PRD

The standard order is requirements, then design, then build. It survives because it looks responsible on a Gantt chart. In practice it front-loads the most expensive kind of work — writing precise specifications for a thing nobody has seen yet — and defers the cheapest source of truth until the end.

When two people on a team disagree about a feature, they are almost never disagreeing about the words. They are disagreeing about a picture in their head. You can spend two days sharpening the document and still ship the wrong thing, because both of them read the same sentence and saw different products.

So I build the disagreement. Not the feature — the disagreement. The narrowest clickable thing that makes the two mental models visibly different. It usually takes an afternoon. It almost always ends the argument in the first ten minutes of the next call, and the PRD that follows is short, specific and uncontested.

This is not an argument against writing. The document still gets written, and it gets written better, because by then it is describing something real instead of proposing something imagined. Acceptance criteria written after a prototype are testable. Acceptance criteria written before one are wishes.

The failure mode to watch for is the prototype that becomes the product by accident. Prototypes are arguments, not architecture. Once the argument is settled, throw it away and build the thing properly — the two hours you spent are already paid back by the two weeks you did not spend building consensus in a comment thread.