Build something in a day with AI by choosing one user, one painful job, and one complete path to a useful outcome. Do not define success as a full product. Define it as a working, reviewable experiment that answers one product question by the end of the day.
Write a one-page brief with the input, output, happy path, failure state, non-goals, and demo condition. Choose familiar, documented components and avoid optional integrations. Ask the agent to identify the riskiest assumption and build that slice first. Use fixtures or a small approved dataset. Keep work in version control, run the project’s checks after each milestone, and preserve a deployable state. If authentication, payments, or sensitive data are not essential to the question, stub them rather than implementing them hastily. If they are essential, reduce another part of scope and review them properly.
Timebox discovery, implementation, verification, and the final demo. Stop adding features once the core path works. Record bugs and follow-up ideas instead of hiding them in last-minute code. A rough interface is acceptable; unsafe data behavior or an unreproducible build is not.
End with a decision: continue, change direction, or discard the idea. Keep the brief, code, test results, and what you learned. AI helps compress the experiment, while strict scope turns one day into evidence rather than a fragile promise.
