Review · August 21, 2026
Requirements for the Discovery of Agents, Workloads, and Named Entities (DAWN)
Problem Statement
Modern distributed systems — including AI agent networks, cloud workloads, and software services — increasingly need to discover each other dynamically, across organizational and administrative boundaries, without pre-configured relationships. Today's discovery mechanisms are largely proprietary and platform-specific, creating fragmented ecosystems. This spec defines the requirements that any interoperable discovery mechanism for these entities must satisfy, without prescribing a particular protocol or solution.
How the Spec Fits In
This draft is not an OAuth document at all — it belongs to the emerging DAWN (Discovery of Agents, Workloads, and Named entities) effort at the IETF. It has no dependency on OAuth RFCs. It is a companion to related DAWN documents: a terminology draft (draft-farrel-dawn-terminology) and problem statement drafts covering AI agent discovery and general DAWN framing.
The spec is purely a requirements document — it defines what properties a future discovery protocol must exhibit, but does not specify any mechanism. It is explicitly informational and solution-neutral, functioning as a design charter for future protocol work within the DAWN space.
High-Level Summary
The draft lays out requirements across six functional areas for any system that allows agents, services, workloads, and other named entities to be discovered across organizational and network boundaries:
- Discovery actors and scenarios: Any entity type — human operator, AI agent, software service, or compute workload — must be able to discover both specific instances and general classes of entities.
- Entity classification: Discovered entities must be classifiable by type, function/capability, geography or jurisdiction, owning organization, and provenance.
- Entity properties: Discoverable information must include communication protocols, capability descriptions (including structured "capability cards"), security parameters (TLS versions, authentication methods), operational status, and responsible party. Properties must be categorized as mandatory vs. optional and static vs. dynamic.
- Trust and security: Discovery information must be cryptographically verifiable (signatures, certificates), resilient to poisoning attacks, and controllable by the publishing entity (access control by audience). The mechanism must not require a single global trust anchor and must avoid becoming a denial-of-service vector.
- Scalability and architecture: The mechanism must be fully decentralized — no single required registry — and must scale across large numbers of entities and administrative domains, including cloud, edge, and enterprise environments.
- Protocol and extensibility: The mechanism should build on existing IETF protocols, support a predictable discovery entry point, and be extensible to new entity types and property schemas without breaking existing implementations.
Interoperability Assessment
Significant Interoperability Concerns
The draft is explicitly a requirements document and acknowledges it does not specify any protocol. This is not unusual for early-stage work, but several requirements are stated at a level of abstraction that will require substantial follow-on specification to produce interoperable implementations. For example:
- REQ-PROP-9 requires support for "capability cards or equivalent structured, machine-readable descriptions," but the format and schema for capability cards is left entirely to future work (or to the referenced terminology draft).
- REQ-PROTO-4 requires "a predictable, ubiquitous, interoperable entry point for discovery" — this is a critical interoperability requirement but is stated without any normative guidance on what that entry point looks like (a well-known URI? DNS? Something else?).
- REQ-SEC-5 ("operate across trust boundaries without a single global trust anchor") is a meaningful constraint but provides no guidance on how trust federation is to be established between administrative domains.
In short, this document cannot by itself produce interoperable implementations. It is intended as a requirements charter, not a specification. The interoperability gap is expected — the question for the working group is whether the requirements are specific enough to constrain the solution space usefully, or so broad that any mechanism could claim compliance.
Requirements That May Be Premature
Several requirements concern operational and policy matters that may be inappropriate to mandate at the protocol level:
- REQ-SEC-7 (auditability of publication and discovery operations) is an operational concern that varies widely by deployment context. Mandating this at the protocol level could over-constrain implementations.
- REQ-ARCH-5 (support cloud, edge, and enterprise environments) is a SHOULD that reads more like a design aspiration than a testable requirement.
- REQ-CLASS-3 through REQ-CLASS-5 (geographic/jurisdictional, organizational, and provenance associations) are reasonable to support but are all SHOULD-level — their inclusion as named requirements adds length without adding normative weight.
Writing Quality
The writing quality is competent and follows standard IETF conventions — RFC 2119 keyword usage is appropriate, tables are used effectively to organize requirements, and scope boundaries are clearly stated. The out-of-scope list is a useful tool that is well-executed here, making it clear that trust, authentication, capability negotiation, and agent communication are all explicitly deferred.
The spec is somewhat long for a requirements document, primarily because each requirement is restated across multiple sections with a mixture of table entries and prose. The tabular format for requirements is appropriate, but some of the prose elaboration in the surrounding text could be trimmed.
The terminology dependency on a separate draft (draft-farrel-dawn-terminology) is notable — normative reliance on an unstable I-D introduces risk if the terminology draft evolves or is abandoned. The relationship between the two documents is not entirely clear from a normative standpoint.
Author Background
Daniel King ([email protected]) King has published 22 RFCs and has 17 active Internet-Drafts, with IETF contributions spanning from 2006 through 2026 across routing, optical networking, PCE, and now AI/agent discovery. He is a highly experienced IETF author who is clearly very active in new work at the moment, given the volume of active drafts.
Adrian Farrel ([email protected]) Farrel has published 83 RFCs and has 9 active Internet-Drafts. He has served as a working group chair and reviewer across multiple areas and has been contributing since 2003. He is one of the more prolific and senior IETF contributors in the routing/transport space. His involvement lends substantial credibility to the draft's procedural quality.
Both authors are from Old Dog Consulting (UK) and are well-known figures in the IETF community. Neither is a newcomer. Their involvement in the DAWN space — Farrel is also the author of the DAWN terminology draft that this document normatively depends on — suggests a coordinated effort to bootstrap a new working group area around agent and workload discovery.