Review · August 21, 2026
A Survey of AI Agent Discovery Mechanisms
Problem Statement
This draft addresses the fragmented landscape of AI agent discovery: there are numerous competing mechanisms (from IETF drafts, industry consortia like AAIF, 3GPP, and open-source projects) with no common understanding of what they cover, where they overlap, or what gaps remain. The document surveys all known mechanisms and conducts a structured gap analysis to guide the DAWN working group's future standardization work.
How the Spec Fits In
This is an informational survey document, not a protocol specification. It does not propose new mechanisms or extend existing RFCs. Its role is explicitly contributory to the DAWN working group's gap analysis deliverable. It references a broad set of existing and in-progress work — DNS-based drafts, A2A Agent Cards, MCP, ANP, 3GPP network-assisted discovery, and several registry/directory proposals — without normatively depending on any of them. It has no OAuth dependencies and is not in the OAuth working group.
High-Level Summary
The draft organizes the discovery problem around five functional components: Lookup (finding candidates), Name (stable identifier), Description (capability metadata), Trust (validating identity and metadata), and Interaction (post-discovery protocol). It then evaluates approximately a dozen existing mechanisms against these components and across five use case scenarios ranging from a curated bundle of tools within a single application to open internet-wide search.
The key findings are that no single mechanism covers all five use cases, pull-based self-description (A2A Agent Cards, api-catalog) is widely deployed but has no cross-domain reach, DNS-based approaches are decentralized but semantically limited, and registry/directory approaches gain cross-domain reach at the cost of centralization or federation complexity. Semantic capability-based discovery — finding an agent by what it can do rather than where it lives — remains almost entirely unstandardized.
The gap analysis surfaces eight specific problems the DAWN working group should address: absence of a standard for semantic capability discovery, no interoperable federation model between registries, no bridge between 3GPP network-layer and application-layer agent metadata, fragmented and incompatible metadata schemas, unresolved trust bootstrapping from DNS identity to interaction-level authorization, no lifecycle or change propagation for static well-known URI mechanisms, namespace pressure from uncoordinated well-known URI registrations, and the fact that a complete discovery system requires composing multiple mechanisms with no standard defining how they compose.
Interoperability and Necessity
This is a survey and gap analysis document, so the standard interoperability lens does not fully apply — it is not specifying a protocol. Evaluated as a gap analysis, it serves its stated purpose well. The comparison tables are clear, the use case taxonomy is useful for scoping future work, and the gap catalog gives the working group a concrete list of open problems.
One area worth noting: the draft surveys several mechanisms that appear to be single-implementer, academic, or otherwise low-maturity (IoA Framework, ADP by Walko Systems, AIDIP, the expired AgentDNS draft). Including these alongside well-deployed mechanisms like A2A and MCP could give a misleading impression of the real landscape. The document does acknowledge this in some cases (noting "expired draft" or "single-author, one implementation") but could be more consistent about flagging maturity levels.
The gap analysis is largely good but Gap 6.7 (well-known namespace pressure) reads as a procedural concern rather than a technical gap, and may be out of scope for a DAWN deliverable. The trust bootstrapping gap (6.5) is described at a high level but could benefit from more precision about which specific trust problem is in scope for DAWN versus other working groups like WIMSE or OAuth.
Writing Quality
The draft is well-written and reads as a genuine survey effort rather than advocacy for a particular approach. IETF conventions are largely followed. The comparison tables are a strong organizational choice and make the coverage gaps immediately visible.
The draft is appropriately scoped in length for a survey document. Section 4 (the mechanism descriptions) is the longest and could be tightened — some entries (particularly the lower-maturity ones) receive detail disproportionate to their likely standardization trajectory. The illustrative flow in Section 3.2 is a useful addition.
One terminological note: the draft uses "AI agent" narrowly to mean LLM-driven agentic loop systems. This definition, stated in the introduction, is narrower than how the term is used in some other DAWN documents and may need alignment with the DAWN terminology draft as that work matures.
Author Background
Jaime Jimenez (Ericsson) — 2 published RFCs, 7 active drafts, approximately a decade of IETF involvement. Serves as WG Chair for CoRE (Constrained RESTful Environments), which is relevant given this draft adapts CoRE's Resource Directory for agent use. Experienced contributor.
Jim Feng (Ericsson) — 0 published RFCs, 1 active draft (this document). New IETF contributor.
Jari Arkko (Ericsson) — 58 published RFCs, former IETF Chair. One of the most experienced contributors in the IETF. His involvement lends significant credibility to this work.
Mirja Kühlewind (Ericsson) — 20 published RFCs, 12 active drafts, former IAB Chair and Transport Area Director. Highly experienced and currently very active. Her breadth of involvement suggests this document will be carefully positioned within the broader IETF process.
Rajat Kandoi (Ericsson) — 0 published RFCs, 1 active draft (this document). New IETF contributor.
The author team is notably Ericsson-heavy (all five authors), which is worth keeping in mind when evaluating whether the survey coverage reflects a balanced perspective across the industry. The 3GPP mechanisms receive relatively detailed treatment, which may reflect Ericsson's strong involvement in that standards body.