AI agent governance

AI agent governance is the control layer over what AI agents actually run on. Where model governance asks "which model, with which guardrails," agent governance asks a more operational question: which skills, instructions, and memory does each running agent read — and who approved them?

The four functions

1. Inventory — every asset, across tools and machines, in one place
2. Approval — review before an asset version ships anywhere
3. Deployment — zero-touch propagation of approved versions
4. Audit — who changed what, when, and where it went

Why it became necessary

Agent behavior is configuration now. The configuration lives in files that multiply and diverge, and in platform-side assets a company cannot even list without counting them against the platform itself.

Related

• Agent approval workflow
• Agent asset inventory

This entry is part of the untactit glossary. Definitions draw on measured write-ups published on the blog.