Skip to content
focoit.com
Back to the lab

AI engineering · Developer tools

AI Engineering Playbook

The rules I run AI coding agents under, across my own repositories: what blocks a commit and what only advises.

Diagram of the playbook loop: spec, plan, execute, report-only review, then a deterministic gate.

Planning, executing and reviewing are separate seats, and the model that executes never comes from the reviewer's family. Reviewers report; they don't fix. The final word belongs to a deterministic gate, never a model.

Secrets block at commit. Confidentiality decides where code may go. Every non-trivial change starts with a backlog item and a short spec, approved before code is written. Model choices live in one registry, so swapping a model can't quietly break the rules.

This site is built under it.