Enterprise agent registries — systems that catalog, discover, and manage AI agents through metadata, versioning, and invocation records — solve a genuine and important problem: organizations deploying agents at scale lose track of what exists, who owns it, and whether it duplicates existing capabilities. Agent registries address this. This paper argues that agent registries are not agent governance. The conflation of catalog management with governance infrastructure produces a dangerous gap: organizations that believe they have governed their agents because they have registered them have not governed their agents. They have inventoried them. Governance requires identity physics. None of these are properties of a registry. They are properties of a governance substrate. The correct architecture combines both layers: a registry for discovery and reuse, a substrate for identity, safety, and accountability. This paper explains the technical distinction, identifies the failure modes that emerge when the distinction is not made, and describes the correct combined architecture for SMB and mid-market agentic deployments.
Narnaiezzsshaa Truong (Fri,) studied this question.