A new Internet-Draft, AIDIP (AI Agent Discovery and Invocation Protocol), landed at IETF this year with a specific complaint: there's no standard way to find an AI agent or call it once you've found one. The draft's authors, from Tsinghua University and Beijing Zhongguancun Laboratory, describe the current state plainly. Frameworks like LangChain and AutoGen each define their own agent description format and invocation API, and they cannot interoperate. Building a multi-agent system today means writing custom glue code for every framework you want to touch, and that doesn't scale as the number of agent platforms grows. AIDIP's fix is a common metadata format covering an agent's capabilities, input/output schemas, supported languages, tags, and authentication methods, paired with a capability-based discovery mechanism and a single RESTful invocation interface. A registry — the draft calls it an Agent Registry or Discovery Service — maintains a directory of registered agents and answers queries by attribute or semantic search. An optional gateway can sit in front, routing client requests to the right agent so callers don't need direct knowledge of every endpoint. The latest revision (02, published in July) adds intent-based agent selection: instead of only searching by declared capability tags, a client can express what it's trying to accomplish and let the discovery layer match it to an agent, without changing the underlying discovery or invocation semantics. Security gets a real section rather than a footnote. All discovery and invocation traffic must run over TLS, and registries must enforce per-client entitlements so that both search results and invocation access respect scopes and permissions. Gateways forwarding requests are expected to authenticate themselves to agents, and agents are expected to maintain stable identifiers and use signed responses when integrity matters. It's worth being clear about status: this is an individual Internet-Draft, not an adopted IETF working group standard, and it carries no formal standing in the IETF process. It expires in January 2027 unless revised or picked up by a working group. But it's notable that it exists at all — a companion draft from the same lead author proposes DNS-native agent naming and resolution, and a separate survey draft (draft-jimenez-dawn-discovery-landscape) now catalogs AIDIP alongside identity and provenance proposals as part of a broader push to standardize how agents find each other before anyone gets to the harder question of whether they should be trusted.