I built the thing I now help you govern.
An AI program operations system, built in production: eight specialised agents on LLMs handling planning, risk, dependency tracking, and executive reporting. Alongside it, a reporting engine holding eight contributors across eleven initiatives to a nine-point quality rubric. Review loops dropped two to three times.
It went into program operations and executive reporting. Not the easy edges.
One design decision matters more than the rest. I built it as a single orchestrator with scoped skills, not a network of independent agents. Multi-agent demos better. Single orchestrator is far easier to audit, credential-scope, and gate. If you cannot explain who approved what, you do not have governance. You have a demo.
Most people selling AI governance have read the framework. I have had to run it, inside a company with real compliance requirements and real people who did not want another process.