Writing

Notes from inside a governed agent fleet.

I run my own finance function on a multi-agent system I built and supervise. What follows is what that has actually taught me - about controls, verification, and the failure modes that show up when software is allowed to act.

White paper

Governed Autonomy: A Reference Architecture for Multi-Agent OpenClaw Fleets

Standing up an always-on agent with real tool access takes an afternoon. Running one safely for months is a different discipline. This paper sets out the pattern - constitutions loaded and integrity-checked at boot, hard action gates that fail closed, a deterministic conductor with a durable ledger, single-writer discipline over shared state, and a quarantine lane for anything a stranger could have authored - and maps each element onto configuration that already exists.

Read the paper →