PRD or Prototype: Which should start with when it comes to building products with AI?

Ravi Mehta argues that a product team should choose between a written requirements document and an AI-built prototype by the decision it needs to make next.

Ravi Mehta writes Top 1% Builder, a newsletter for product builders, and previously worked at TripAdvisor. A working prototype can now take less time to make than a product requirements document, the written plan for a feature. Mehta argues that the two should develop together, and that each prototype should be built to answer one named decision.

• He gave Magic Patterns, an AI prototyping tool, a requirements document that a TripAdvisor team had spent six weeks writing, and had a working prototype in about 20 minutes.
• That prototype also made choices he had not approved, such as removing a saved place without asking for confirmation.
• He names four kinds of prototype by the question each answers: which direction to take, what exactly to build, whether customers want it and whether it can work.
• He adds a formal step before launch that asks whether a finished piece of work is worth the customer's attention.

A product team can name the decision a prototype should settle before anyone builds it, as Mehta recommends, and read an AI-built prototype for choices nobody made.

Derived by Working Surface from the article; more in the Playbook. Source line: But speed does not turn generated decisions into product intent.

18 October 2026: as AI agents make code and prototypes cheap, teams are writing down which work needs a human check and which ideas have earned more effort.