Interview the idea before building
Use AI to surface missing decisions before it starts changing files.
By the end, you can ask the AI to identify uncertainties and return a reviewable first-build plan.
Step 1: Stop the guessing
AI can fill an unclear with features you never wanted. Like briefing a contractor, tell it to ask about users, the first task, stored data, access, and exclusions before starting work.
Step 2: Define one complete trip
A slice is a short trip from action to result: choose a book → send a practice → see its outcome. A styled screen with a dead button is not a complete trip. A small plan names the start, ending result, failure case, and the files or settings it needs.
Step 3: Approve the next trip
Ask one question at a time. Correct assumptions, then ask for a proposed plan and wait point. Compare extra features against the goal. Deadlines and visual style matter, but they cannot substitute for deciding who may do what.
Step 4: Use results to plan the next change
Try the agreed flow and one failure. what happened and which questions remain. Approve the next change only after you can explain the current result. If the AI edits unrelated parts, stop and a narrower proposal.
Do not write code yet. Interview me about the visitor’s task, the first workflow, needed data, permissions, and exclusions. Ask one question at a time. Propose one complete action-to-result slice and a failure check. Wait for my review. Illustrative before / after (behavior, not ready-to-run code): - Build eight features together + Propose one trip → review assumptions → build → check result
A plan is reviewable when you can tell what success and failure would look like, not when it contains the most technical detail.
Sources and further reading
You approved only a book catalog, but the AI starts adding chat and payments. Which instruction best restores the agreed scope?
Choose one answer.