I've watched BAs adopt AI tools over the last two years and noticed a pattern. The ones who use AI as a faster acceptance-criteria generator produce faster bad requirements. The ones who use AI as a thinking partner before they touch the AC produce something interesting. Same tools, opposite outcomes.
This piece is the workflow that produces the second outcome. It is not theoretical. It is what the BAs I respect actually do when they sit down with a new stakeholder request and an open Claude or ChatGPT window.
Three stages, in order. Each stage has a clear thing AI is good at and a clear thing it is not.
Stage 1: Stakeholder interrogation, with AI as opposition
The first hour after a stakeholder request lands is the highest-leverage time in the entire requirements process. It is also the time most BAs skip, because the request "feels clear" and the temptation is to start scoping.
