2026-08-16 · 5 min read
What is an AI agent control plane?
A practical explanation of the versioning, execution, and observability responsibilities an AI agent control plane should own.
Field notes
Blog
Specific, implementation-grounded guidance for operating versioned AI agents.
2026-08-16 · 6 min read
AI agent observability: runs, traces, tokens, cost, and latency
Learn which signals make AI agent execution inspectable and how to connect them without exposing sensitive run content.
2026-08-16 · 5 min read
Bring your own OpenAI key without putting secrets in agent configuration
A security-focused guide to workspace-scoped OpenAI credentials, encrypted storage, replacement, removal, and separate billing.
2026-08-16 · 5 min read
Tracking AI agent cost without pretending an estimate is an invoice
Understand token categories, versioned pricing snapshots, integer-micros accounting, and the meaning of unavailable cost.
2026-08-16 · 6 min read
Designing read-only GitHub agents with enforceable boundaries
How installation authentication, pinned repositories, allowlisted GET tools, and bounded evidence constrain GitHub agents.
2026-08-16 · 5 min read
AI agent retry semantics: transient failures, terminal errors, and crash windows
A practical model for classifying provider errors, bounding retries, preserving terminal state, and documenting duplicate-call risk.
2026-08-16 · 5 min read
Scheduled AI agents in production: version pins, timezones, and overlap control
Design recurring agent execution with immutable versions, explicit timezones, idempotent occurrences, and bounded catch-up behavior.