AI Coding Assistant Checklist: Safer Changes and Better Tests
AI-generated code is a proposed change, not proof that a feature works. A short review routine can catch many preventable problems.
Provide a reproducible problem
Give the language, framework, error message, expected behaviour and the smallest relevant code example. Remove passwords, API keys and personal data before sharing files.
Ask for a change plan
Request the files to be edited, the reason for each edit and an easy rollback. Prefer small diffs over full-project rewrites. If AI has not inspected your code or logs, its diagnosis is still a hypothesis.
Inspect security-sensitive logic
Review authentication, authorisation, inputs, database updates and secret handling. Do not put privileged credentials in browser JavaScript. Test denied access as well as successful access.
Test the user journey
Run the test suite, then manually check loading, forms, navigation, phone layout, accessibility and errors. A successful build is not the same as a working product.
Deploy with a rollback
Use a preview environment, review logs, compare behaviour with production and keep a known-good deployment. If an important regression appears, roll back before adding more changes.