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

Review · November 16, 2026

DAWN Use Case for Protocol-Layer AI Agents

draft-jadoon-dawn-layered-agent-discovery -00View on Datatracker →

The Problem

An organization runs several AI agents that each help with one part of the network stack. One tunes application media. One helps with transport adaptation. One helps with routing. One helps with packet scheduling.

Two of them might both advertise "traffic optimization". A discovering agent cannot tell which one works on transport and which one works on routing.

The draft asks one question (Section 4): what is the minimum public discovery information needed to tell these agents apart and contact them, inside one organization's discovery scope?

It does not answer the question. It is a use case and a problem statement, nothing more.

How It Fits In

This is an individual draft with a 5 minute slot in the DAWN session on Thursday. It defines no mechanism, no fields and no roles.

Which charter deliverable does it address? The Use Cases and Applicability deliverable, which the charter describes as "a document describing broad DAWN use case categories". This draft is one narrow use case, not a category. It is input to a use cases document, not a candidate to be that document.

The problem statement also touches the Discovery Architecture deliverable. The charter says that deliverable covers "requirements (including the minimum public discovery information)". The draft reads as a test of REQ-PROP-1 in draft-king-dawn-requirements. That requirement says the Minimum Discoverable Information (MDI) must include the entity's type, offered services, reachability, protocols and security parameters. The draft asks whether "offered services" is enough.

It stays inside the charter. It limits itself to initial discovery within one organization, which is the second of the charter's three deployment contexts. Capability exchange, agent selection, authorization and coordination are declared out of scope (Section 1).

How does it compare with the main use cases draft?

draft-kay-dawn-use-cases is the broad one. It defines four discovery categories and five entity classes.

This draft does not cite it. That is the biggest gap in the document.

The scenario fits inside the existing categories. Section 7.4.3 of draft-kay-dawn-use-cases-01 already has an example of an AI-assisted network operations system discovering telemetry sources, topology systems and control points within an organization. The Jadoon scenario is a more detailed version, with an agent as the discoverer.

The entity class sections of draft-kay-dawn-use-cases-01 are still TBD, including Section 8.1 "AI Agent" and Section 8.4 "Network Function". This scenario could be a worked example for either one.

So the two drafts do not compete. One is a catalog. The other is a single entry.

Summary

The scenario (Section 3) is an augmented reality maintenance application in an industrial facility. The organization runs an application agent that adapts AR content resolution, a transport agent that watches round-trip time and loss, Internet-layer agents for routing and forwarding, and network-operation agents for scheduling and resource management.

The transport agent acts as the Discovering Entity. It looks for agents that expose network-operation services. The discovery mechanism returns several that look alike.

Section 4 says it "does not presume that new mandatory MDI fields or a particular discovery architecture are required".

Interoperability

There is nothing to implement here. The useful question is whether the use case produces a requirement the working group does not already have.

Is the problem specific to protocol layers?

No. Two agents with the same vague service label is a problem in every domain. Two "payments" agents have it too. A more specific label fixes it with no new field.

The charter gives its own examples of minimum public discovery information: "type, reachability information, communication protocol, services". A precise entry under "services" may already be enough.

There is also a limit on the other side. The charter puts "exchange of capability information beyond the minimum public discovery information" out of scope, along with selection of an AI resource based on capabilities. A description of which protocol layer an agent works on starts to look like capability information used for selection. The draft does not say where it draws that line.

What is missing

  • No candidate requirement. The draft ends on a question. One sentence in the style of the requirements draft would give the working group something to accept or reject.
  • No link to the MDI work. draft-cui-dawn-mdi-model proposes an actual MDI data model. The draft does not say whether that model can already express the distinction.
  • Security Considerations is one sentence. These agents influence routing, forwarding and scheduling. A discovery answer that lists them is a map of the network's control points.

Writing Quality

The writing is clean and grammatical. The scope statements are clear.

It is also thin. The introduction takes a full page to motivate agents at each protocol layer, then the problem statement is one sentence. The ETSI Technology Radar paragraph in Section 1 supports no claim in the draft.

It cites draft-farrel-dawn-terminology-06. The current revision is -07.

Authors

From their Datatracker profiles:

  • Muhammad Awais Jadoon (InterDigital Europe). No published RFCs. 4 active drafts and 4 expired.
  • Sebastian Robitzsch (InterDigital Europe). No published RFCs. 3 active drafts, all co-authored with Jadoon.
  • Carlos J. Bernardos (Universidad Carlos III de Madrid). 18 RFCs, mostly on IP mobility, DHCPv6 and deterministic networking. 14 active drafts.

Questions for the Working Group

So what should the session decide in the review of proposals against the deliverables?

  • Should this scenario be folded into the TBD "Network Function" or "AI Agent" section of draft-kay-dawn-use-cases instead of standing alone?
  • Can the authors state in one sentence a new requirement that REQ-PROP-1 does not already cover?
  • Can the MDI model in draft-cui-dawn-mdi-model distinguish a transport agent from a routing agent today?
  • Is "protocol layer" part of the minimum public discovery information, or is it capability information that the charter puts out of scope?