Most of the coverage of the Cloud Security Alliance's new Agentic AI IAM paper focuses on its headline pitch: decentralized identifiers and verifiable credentials for AI agents. But tucked inside the framework is a piece that belongs to a different problem entirely — discovery. The paper's authors propose an "Agent Naming Service (ANS) for secure and capability-aware discovery, dynamic fine-grained access control mechanisms, and critically, a unified global session management and policy enforcement layer for real-time" coordination across agents. That's the framework treating discovery as inseparable from identity: an agent can't be trusted to act until something can resolve who it is and what it's allowed to do, and that resolution step is where ANS sits. The framing matters because it's explicit about why existing web infrastructure doesn't do this job. The paper argues traditional IAM protocols — OAuth 2.1, OIDC, SAML — were built for human users or static machine identities, and their coarse-grained, single-entity model breaks down once you have agents spinning up and down, delegating tasks to each other, and needing to be looked up mid-transaction rather than provisioned once at deployment. ANS, as sketched, isn't just a lookup table. It's meant to answer a capability question, not just a location question: given a task, which agent has the right verified credential and behavioral scope to be trusted with it. That's a step beyond DNS's original job of mapping a name to an address — it's closer to a registry that encodes trust metadata directly into the resolution answer. The practical gap is that ANS remains a proposal inside an academic-style whitepaper, not a running protocol with adopters. CSA has published a related Agent Identity Governance Framework that maps these ideas onto existing standards like the AI Controls Matrix and OWASP's Agentic Security Initiative, which suggests the industry is converging on vocabulary even where it hasn't converged on implementation. Whether ANS becomes an actual deployed naming layer, or just a conceptual placeholder that gets absorbed into other registries, is the thing to watch. The paper is honest that this is early: it's building on a prior CSA publication and explicitly frames itself as expanding scope rather than closing the gap between agentic AI and workable IAM.