Six checks before AI coding: PRD first to stop drift

A familiar failure mode in AI-assisted development: because the code runs, developers keep adding the next feature, and the product drifts from what they first intended. Its answer is to make explicit, in words, the background that people would otherwise fill in through conversation — and to do so before implementation starts, since faster implementation widens any gap in understanding.

The six items are presented as a connected chain rather than a checklist. The PRD defines who and what is being solved and how success is measured; technical requirements capture constraints, quality, stack and the reasons for choices, with an ADR recording overall decisions. Screen flows and UI/UX policy then describe the experience, and data design turns the PRD's purpose into concrete types, keys and constraints. Scope and roadmap tie the purpose to the work, updated together as testing proceeds.

The point is the connection from purpose to experience to technology and data to implementation. Even a small fix should loop back to what the whole is aiming at.