Theory
Requirements and user stories are the foundation of any software project. The problem is that this phase is often slow, inconsistent, and information gets lost between product discussions and the backlog. As AI makes coding faster, the bottleneck shifts upstream: if you can generate code in minutes, weak or vague user stories become the real bottleneck.
AI is good at several things here. It can turn stakeholder interview notes or meeting transcripts into structured user stories, generate acceptance criteria from existing stories, evaluate stories against quality frameworks like INVEST, and help build user personas. It also bridges the communication gap between business needs and technical specifications.
But humans still need to stay in the loop. AI cannot replace actual user interviews or stakeholder conversations. It cannot validate whether a story makes real business sense, and it can produce plausible-sounding but wrong requirements. The real value is using AI to raise your baseline: generate a first draft, evaluate it, and then refine with human judgment. Think of it as a collaborator that handles the tedious drafting work while you focus on what actually matters to the people using your product.
Practice
Run this in a repository you already know, not a toy project. The point is to feel where the practice helps and where it gets in the way on code that has history.
Pick a real feature you are working on or planning. Then run this exercise.
Give an AI model (ChatGPT, Claude, etc.) a short description of your product and ask it to generate a user persona for your target user. Review the output and note anything surprising.
Next, ask the AI to write user stories for that persona using this format: "As a [role], I want [what] so that [why]." Pick one functional area to keep it scoped.
Then ask the AI to add acceptance criteria to each story, first as a simple list, then in Gherkin format (Given / When / Then).
Finally, paste the stories back and ask the AI to evaluate them against the INVEST criteria: Independent, Negotiable, Valuable, Estimable, Small, Testable. Ask it to flag only the criteria each story fails and suggest one concrete fix.
At the end, compare the output to what you would have written on your own. What did the AI catch that you missed? Where was it wrong? That reflection is the most useful part of the exercise.