Review · August 21, 2026
DAWN Terminology (draft-farrel-dawn-terminology-04)
Problem Statement
As distributed systems proliferate — including AI agents, cloud workloads, software services, and network functions — there is no consistent shared vocabulary for talking about how entities discover each other. This draft defines a common terminology for the DAWN (Discovery of Agents, Workloads, and Named Entities) working group, so that related DAWN documents can use terms consistently without each document needing to re-litigate definitions.
How This Spec Fits In
This document is not an OAuth or identity spec. It belongs to the IETF DAWN working group, which focuses on automated discovery mechanisms for distributed entities including AI agents. This draft is a foundational terminology document — analogous to a glossary RFC — that will be referenced by subsequent DAWN architecture and protocol documents. It does not define any protocol behaviors, wire formats, or new mechanisms. It is strictly a shared vocabulary document intended to serve as a stable reference for the rest of the DAWN document family.
High-Level Summary
The draft defines roughly 25 terms organized around four conceptual clusters: entities (what can be discovered), discovery (the lookup process itself), capabilities (what entities offer), and registration (how entities become discoverable). The scope is explicitly limited to the discovery lookup step — finding an entity and learning enough about it to initiate interaction — explicitly excluding registration, capability exchange/negotiation, and entity selection from the DAWN charter.
Key defined terms include:
- Entity: Any system component that communicates — including AI agents, tools, tasks, workloads, services, and network functions
- Agent / Workload: Specific subtypes of entity; agents act autonomously, workloads are units of compute
- Minimum Discoverable Information (MDI): The minimum data set an entity must publish to be discoverable — the "common header" concept
- Capability Card: A structured machine-readable description of what an entity can do (variants: agent card, task card, tool card, etc.)
- Discovery Scope: The bounded domain over which discovery operates, described across administrative, trust, geographic, and temporal dimensions
- Trust Indicator: Signed attestations, certificates, or similar mechanisms enabling a discoverer to assess entity trustworthiness
- Discoverable Object: The information object that gets registered and returned during discovery, potentially represented as a capability card
The draft also carefully distinguishes terms that could be confused: "Attributes" (the full internal set of properties an entity has) vs. "Properties" (the discoverable subset), and "Capability Exposure" (publishing capabilities to a registrar) vs. "Capability Exchange" (negotiating capabilities directly with another entity, which is out of scope).
Interoperability and Necessity
Is a terminology RFC necessary? Yes. The DAWN working group is producing multiple documents concurrently, and without a stable shared vocabulary, each document would define its own variant of terms like "agent", "workload", and "capability card." This draft does the necessary groundwork to prevent that fragmentation. The approach is standard practice for new IETF working groups dealing with a novel problem space.
Underspecified concepts that may cause problems downstream: The "Minimum Discoverable Information" (MDI) concept is defined only loosely — as a "minimum viable dataset" analogous to a common header. For the rest of the DAWN document family to be interoperable, MDI will need precise specification somewhere (likely in the architecture document). The terminology doc wisely flags this without trying to prematurely define the full content, but the gap is notable.
Subtle distinction that may not hold up: The distinction between "Attributes" (everything an entity has) and "Properties" (the discoverable subset) is logically clean but may prove difficult to maintain consistently across documents. The boundary between what is discoverable and what is private will likely vary by deployment context, and the current definitions don't give implementers guidance on how to make that determination.
Terms that feel premature: "Task" is explicitly flagged as a legacy term retained for continuity, with no clear semantic distinction from "workload" or "software service." This suggests the terminology is not fully stabilized and may evolve. Retaining a term with a note that it might be equivalent to another term invites confusion rather than resolving it.
Writing Quality
The writing is professional and consistent with typical IETF informational document style. Sentences are grammatical and well-formed throughout. The alphabetical ordering of definitions is standard for glossary documents, and the supplemental Figure 1 (showing the hierarchical relationship between terms) is a useful addition that helps readers understand term dependencies.
The length is appropriate for a terminology document — it defines a bounded set of terms without padding. The security and privacy considerations sections appropriately note that this document only defines terms while pointing forward to the substantive implications those terms will carry in future DAWN architecture documents.
One structural oddity: the document simultaneously recommends reading terms alphabetically (for reference) and in Figure 1 order (for comprehension). The dual-ordering recommendation slightly undercuts the clarity of the document structure, though it is a minor issue.
Author Backgrounds
Adrian Farrel (Old Dog Consulting, UK) — 83 published RFCs, IETF veteran with contributions dating to 2003, primarily in MPLS, GMPLS, PCE, and traffic engineering. Currently chairs three working groups (MPLS, DAWN, and CATS) and holds roles in two directorates. He is the DAWN WG chair, which explains his authorship of the foundational terminology document. Highly experienced, clearly authoritative on process.
Kehan Yao (China Mobile, China) — 0 published RFCs, but 14 currently active Internet-Drafts, making him highly prolific in early-stage work. His drafts cluster around CATS, AI/agentic networking, and compute-aware traffic steering. No shipped RFCs is notable given the volume of active drafts.
Roland Schott (Deutsche Telekom, Germany) — 3 published RFCs (RTCP extended reports in 2013–2014, ALTO in 2024), with 5 active drafts spanning AI networking and DNS-based discovery. Long-term IETF participant whose interests have evolved toward AI-era networking topics. Moderate experience.
Nic Williams (Infoblox, USA) — 0 published RFCs, 3 active drafts focused on DNS-based AI discovery. Emerging contributor with a background in DNS/DHCP and an academic focus on cybersecurity and AI. New to the IETF standards pipeline.
The author team is anchored by a highly experienced WG chair (Farrel) with a mix of more junior co-authors who represent the AI/networking constituency the DAWN group is targeting.