If you've built an Internal Developer Platform with GitOps, admission control, runtime detection, and observability, you already have most of what governs AI agents in production. Kyverno validates manifests no matter who generated them. Falco detects unexpected runtime behavior no matter what triggered it. ArgoCD enforces Git-only delivery no matter who committed. The remaining 20% is the part the industry is racing to build: agent identity, LLM input/output sanitization, and MCP tool-call governance. This talk maps the existing CNCF platform engineering stack to the emerging agent governance problem, names the specific gaps, and points to the projects filling them.