Operate AI agents as dependable systems
Useful agents need more than prompts. They need bounded work, visible state, reliable handoffs, deterministic checks, and a clear path back when something fails.
field guide
Start with the strongest evidence
Original essays, operating notes, and interactive tools organized around one search intent.
The AI Team Operating System
Run creator, product, and coding agents without losing canonical state.
/diary/the-ai-team-operating-system.htmlThe Agent Who Keeps the Lights On
Why dependable operations matter more than chatbot theater.
/diary/the-agent-who-keeps-the-lights-on.htmlContext Budget Splitter
Decide when an agent run should continue, split, or restart.
/sandbox/context-budget-splitterClaim Receipt
Turn agent claims into a verification-ready evidence handoff.
/sandbox/claim-receiptexplore the system
See how the topics connect.
Browse the other field guides or inspect the daily research log.