Free · one email · 7-part playbook
The AI SDLC Playbook
From a spec to production when AI agents write most of the code: spec, plan, build, review that reproduces bugs, gates, ship with a way back, and the lesson loop. Seven parts, the way I work.
What’s inside
- 1Spec firstThe agent builds what you wrote down. Write the outcome, the checks and the edges before any code.
- 2Plan in the repoBreak the spec into small steps, each with its proof, in a file every session reads.
- 3Build small, check as you goOne step at a time, the agent runs the checks itself, and nothing is committed on a red.
- 4Review that reproducesAI reviewers read the files, not just the diff, and a finding counts when it comes with the input that breaks it.
- 5Gates and evalsTests guard the code before it ships; evals guard the behaviour after. You need both.
- 6Ship with a way backEvery release can be undone in one move, and something watches it after it lands.
- 7The lesson loopEvery incident becomes a rule the next session reads. Agents propose lessons; a person accepts them.
Already subscribed? Use the same email and I'll send you straight in.