Agentic AI Governance: How Do We Govern What We Didn't Design?
ITSM, Tuesday, September 8th, 2026
Governing systems that construct responses rather than follow defined paths calls for observation over prescription.
Jon Stevens-Hall examines the governance problem posed by autonomous systems that construct responses rather than follow predetermined paths. Drawing on complexity theory and the Cynefin framework, he argues that prescriptive rule-setting fails when the behavior space cannot be enumerated in advance.
The alternative proposed is continuous observation, including LLM-as-judge techniques, that allows creative freedom within defined boundaries while detecting when outputs drift outside them.
For service management teams asked to govern agents they did not specify, it is one of the more rigorous treatments available.