The Scope Creep Automated reviews of new AI-related working group submissions

Review · August 21, 2026

DAWN Problem Statement

draft-akhavain-moussa-dawn-problem-statement -05View on Datatracker →

DAWN Problem Statement (draft-akhavain-moussa-dawn-problem-statement-05)

Problem Being Solved

The draft identifies a gap in the AI ecosystem: there is no standardized mechanism for discovering AI entities (agents, tasks, workloads, models, datasets, compute resources) across organizational and administrative boundaries. Current approaches either rely on static configuration or platform-proprietary directories that only function within small, closed environments — neither scales to the open, cross-domain, dynamic deployments that multi-agent AI systems require.

How This Spec Fits In

This is a problem statement document for the DAWN (Discovery for Agentic Workloads and Networks) working group. It is informational rather than standards-track, meaning it characterizes a problem space rather than defining a protocol. It draws on draft-farrel-dawn-terminology for vocabulary and gestures toward a companion privacy document (draft-iannone-dawn-privacy-considerations). It does not extend or depend on OAuth or any authentication RFC — its scope begins where authentication ends, at the discovery layer.

The document does not create new protocol mechanisms. Its purpose is to motivate and scope a future solution space, making it a prerequisite artifact rather than an actionable specification.

High-Level Summary

The spec argues that AI ecosystem participants — agents, tasks, compute workloads, models, and data sources — need to find each other dynamically, at scale, and across organizational lines. Today's approaches (MCP agent cards, A2A name lookups, DNS variants) are described as too static and too siloed.

The document frames discovery as one stage in a broader entity lifecycle (registration → identity → authentication → authorization → discovery → communication → monitoring) and focuses exclusively on the discovery stage. It defines functional requirements across several axes: query granularity (broad capability search vs. specific named lookup), minimum discoverable information (capabilities, protocols, policies, pricing, status), cross-domain operation, handling of rapidly changing attributes, and human-operator-initiated discovery for audit and compliance.

A taxonomy of five entity types (AI Agent, Software Service, Compute Workload, Network Function, Application Endpoint) is introduced, classified by identity binding, control ownership, responsible party, rate of attribute change, and cross-domain visibility scope.

Nine candidate use case patterns are outlined: static pre-configured discovery, capability-oriented discovery, chained/parallel/sequential discovery, resource-oriented, closed-system, open cross-domain, interactive multi-round, reversed (entity advertises needs), and operational/audit discovery.

The security framing is brief but appropriate: discovery information is a high-value attack target, and cross-domain trust requires both authenticating the discovery source and verifying the registered entity separately.

Interoperability Considerations

This is a problem statement, not a solution spec, so interoperability of implementations is not yet in scope. That said, the document does set expectations for the eventual solution space, and there are some observations worth noting.

Risks of over-scoping future work: The taxonomy in Section 4.6 is detailed — five dimensions, five entity types — and the functional requirements span nine subsections. The risk is that a future protocol tries to address every row in every table, producing an over-engineered specification when a narrower subset might achieve the core interoperability goal. A stronger problem statement would distinguish "must have" requirements from "nice to have" ones.

Useful constraint: The document explicitly assumes trust is established before discovery. This is a welcome scoping decision that avoids conflating discovery with authentication, and aligns with how OAuth separates concerns. It should be preserved and strengthened in future solution work.

Gap analysis deferred: Section 6 lists existing approaches (DNS, DNS-SD, Webfinger, MCP, A2A, AGNTCY) and calls for a "formal gap analysis" — but that analysis is not provided in the document. For a -05 revision, this feels like a gap. The problem statement would be more compelling if it showed concretely where each existing mechanism falls short against the stated requirements.

Writing Quality

The document is generally readable and uses appropriate IETF language. The structure is sensible for a problem statement. The entity taxonomy table and the use-case list in Section 12 are clear and well-organized.

However, there are some areas where the writing is less precise than expected at revision -05:

  • Several sections read more like a survey or brainstorm than a focused problem statement. The operational considerations section (Section 10) in particular covers five subsections of monitoring, audit, and policy visibility — material that is relevant but could be substantially compressed since it doesn't change the core problem being articulated.
  • The claim that existing tools "work only in small, closed environments" is made but not substantiated. At revision 5, this assertion should be backed by concrete evidence or references, not just asserted.
  • The transition from "here is the problem" to "here are nine use case patterns" (Section 12) happens late in the document and feels appended rather than integrated. If these use cases motivate the requirements in Section 5, they should appear earlier.

The document could likely be reduced by 20–30% without losing substance, which would make it easier to review and align the working group around a focused problem scope.

Author Background

Arashmid Akhavain (Huawei Technologies Canada) — No published RFCs. Has 5 active Internet-Drafts. A newer IETF contributor currently building participation.

Hesham Moussa (Huawei Technologies Canada) — No published RFCs. Has 5 active Internet-Drafts. Also a newer IETF contributor at a similar stage.

Daniel King (Old Dog Consulting) — 22 published RFCs spanning 2006–2026, 17 active drafts. A highly experienced and prolific IETF participant with deep background in routing and network architecture. His co-authorship lends significant credibility and IETF process experience to what would otherwise be a document from two first-time RFC authors.