Harness Engineering 2: Inspect and Improve
Israel Ayliffe reads 30 days of your real work and turns what it finds into a monthly loop.
You have been working for a month. Somewhere in that month you solved the same kind of problem four separate times without noticing, and each time you started over.
This session finds those. Israel Ayliffe takes the last 30 days of real AI coding work, runs it through the harness, and surfaces the patterns hiding inside it. Not opinions about what you should be doing. Evidence from what you actually did.
Then the second half turns those findings into something that keeps running: a monthly loop that reviews your work, proposes changes to the harness, and waits for you to approve them.
The rule never moves. The system may analyze on its own. It may recommend on its own. It may only change the harness after a human says yes.
This is Session 2 of a two-part series. Session 1 built the environment. This one makes it improve itself, on your terms.
Before you arrive:
Have your harness installed and running.
Bring 30 days of real sessions. Thin history means thin findings.
Same model and usage notes as Session 1.
Best if you attended Session 1 or already run a harness in Codex or Claude Code.