The AI Reckoning / Chapter 17

Daily AI Workflow

A repeatable check before accepting AI-assisted changes.

Download printable PDFOne page · Updated October 20, 2026

Run the seven steps

Define the task: state the required behavior, constraints, inputs, outputs, and acceptance checks.
Generate a small change: give the relevant context and explicit boundaries. Keep credentials and private data out of unapproved tools.
Read the diff: understand changed behavior, dependencies, permissions, and error paths. Investigate anything you cannot explain.
Test correctness: check the required output and business result, not just valid syntax or a successful run. Include failure and boundary cases.
Commit a coherent change: preserve a useful history with a message explaining the purpose.
Verify stored data and integrations: use authorized access and safe test environments; avoid exploratory writes to production.
Check scope and recovery: separate unrequested refactors, rerun affected checks, and know how to undo the change.

Record model, prompt, retrieval, and supported generation settings. Fixed settings can help repeat a test; they do not guarantee identical answers.
Evaluate both format and task correctness. Test untrusted inputs and tool permissions. Keep a human accountable for the release.

Change: …. Evidence checked: …. Reviewer / rollback: ….

Use the printable worksheet to keep your notes and evidence with your team.