The agent reference
Running AI agents with proof.
My companies run on AI agents around the clock. These pages are the rulebook they run under, the system that enforces it, and the ledger that shows what actually got done. All of it comes from production, not a demo.
Start here
The bar every agent task has to clear, the rules that enforce it, and the system that runs them.
- StandardThe Definition of Done for AI AgentsAn open standard (v1.0) for when an agent may call a task done, with a free checker that scores your logs.
- ReferenceAI Agent Governance: 12 RulesThe twelve rules my agents run under in production, and what each one caught.
- ReferenceWhat Is an Agent Control Plane?The nine parts of the control plane that runs my companies, and the five you need to start.
The receipts
What happened when I held my own agents to that bar, in numbers.
Use it
Check your own agents, or have a controlled operator built for one workflow.
- ToolThe Babysitting DiagnosticDrop in your Claude Code session logs and see how often you step in. Runs in your browser.
- ServiceControlled AI Operator SystemsScore one recurring workflow and see whether it is ready for an operator with approvals, caps and proof.
- TalkAgency AI, With ReceiptsThe talk for agencies: what running agents actually costs and how often "done" is true.
The live test all of this feeds is Company Autopilot: can a company keep running after I close the laptop? The marketing side runs on Kai CMO.
Get the next scorecard.
Real numbers from agents running a real company, including what broke. One email when there's something worth reading.