Build agents that can act and prove what happened.
Field notes on governed AI. Long-form technical guides for AI systems that remain legible under pressure: authority, evidence, observability, and fail-closed execution.
- Guides
- 24
- Themes
- 3
- Latest
- 2026-10-31
Start with the signal.
The newest field notes: the fastest route into the current body of work.
Build a Supervisor Agent: Talk to One Session While the Rest Keep Moving
Agent Engineering · 2026-10-31
You run several coding agents on one project, and somewhere along the way you became the message bus between them. A supervisor session fixes that. You talk to it, and it keeps every other session moving, routes their questions, checks their work, merges what passes and brings you only the real decisions.
Read field noteI Read All 471 of My Claude Code Sessions: Where 6 Billion Tokens Actually Went
Agent Engineering · 2026-10-23
97% of my tokens were cache reads, so my cache looked healthy. Weighed by price, the 3% that were cache writes cost almost as much as all the reads, and 94% of the big rewrites came after a cache had expired from idle time, mostly when I went back to a session or a finished subagent.
Read field noteHand over development evidence without losing the original observation
Agent tooling · 2026-10-19
A useful handover preserves what the first worker saw while making the next worker check what is true now.
Read field noteFollow the question.
24 guides
The same hard questions, from different angles.
The formats change. The doctrine does not. Use a theme as a lens into the corpus.
Prefer the picture? The same systems, as visual explainers.
Explore architecture diagrams