How the work gets done · 6 of 7 posts
The AI Co-Pilot
How one person runs a complex home energy project with an AI co-pilot.
- Part 1 · 8 min read
The Co-Pilot Model
What makes a one-person home project manageable is not the model but the system around it: a knowledge file, a memory index, an issue tracker and session rituals.
- Part 2 · 10 min read
Never Accept First Output
The first draft of my first post had eighteen confident errors and still passed review. The fix was structural: a plan-review-execute-review-learn cycle with hard gates.
- Part 3 · 11 min read
The Knowledge That Compounds
Session five of a task should start further forward than session one. A plain-text memory system with a short index is what makes the lessons stick.
- Part 4 · 8 min read
Agents That Work While You Sleep
Overnight agent runs are only as good as the brief and the shared state. One wrong fact in the knowledge file went into every post drafted that night.
- Part 5 · 5 min read
The Ecosystem Problem
I asked an AI co-pilot to audit my own home project, which had sprawled to seven workstreams, and set a kill-or-commit date on one of them; then I let that date slide, and here is what happened.
- Part 6 · 7 min read
The Stack
The four layers under my home energy project, how one real piece of work moved through them, and why the hard line between the assistant and the house is now mostly a rule.
- Part 7 Coming
Reconcile, Don't Install
Why I keep only the parts of someone else's rules pack that earn their place.