Coordinating Agent To Agent List of efforts (CATALIST)
Wednesday 2026-03-18 09:00-11:00 Shenzhen

Co-Chairs: Yingzhen Qu, Adrian Farrel
Note-takers: Muhammad Usama Sardar, Laura Scalone, Laurent Ciavaglia

Materials: https://datatracker.ietf.org/meeting/125/session/catalist

Note taking: https://notes.ietf.org/notes-ietf-125-catalist

Meetecho:
https://meetings.conf.meetecho.com/ietf125/?group=catalist&short=catalist&item=1

Zulip: https://zulip.ietf.org/#narrow/stream/catalist

==Agenda==

1. Administrivia [5 minutes]
   Chairs
2. Purpose of BoF [5 minutes]
   Chairs/ADs

3. Summary of IETF efforts and their scope [60 minutes]
   Questions should be for clarification only
   3.a. AI Agent Protocols [10 minutes]
   Cullen Jennings
   3.b. Network architecture for the AI ecosystem [10 minutes]
   Arashmid Akhavian
   3.c. AI Agent Discovery [10 minutes]
   Nic Williams
   3.d. AI-Agent Network Requirements and Protocols (ANP) [10
   minutes]
   Kehan Yao
   3.e. Dynamic Multi-agents Secured Collaboration (DMSC) [10
   minutes]
   Aijun Wang
   3.f. Agent Communication Interdomain Protocols (ACIP) [10 minutes]

   Toerless Eckert
   4. Work outside the IETF [10 minutes]
   Suresh Krishnan
   5. Open discussion [30 minutes]
   6. Closing remarks and next steps [10 minutes]
   Chairs/ADs

============
==Notes==

1. Administrivia

2. Purpose of BoF

Adrian went through the slides here

• Not a WG forming BoF
• Discussion forum for Agent-to-agent communication
• Chairs emphasize that side meetings are not a formal part of the
  IETF.

Usama: Security topic should not have been removed. It is really
important.
Chairs: Encourage folks to talk about security in their presentations
and on the list.

1. Summary of IETF efforts and their scope
   • Clarifying questions only and discussion later on

3.a. AI Agent Protocols

• No one in queue

3.b. Network architecture for the AI ecosystem

• Side meetings since IETF 122
  No one in queue

3.c. AI Agent Discovery

• Been talking about drafts since IETF in Madrid

Ekr (chat): We already have a thing for "show me agents given the
domain name". That's what MCP does!

Balazs Nemethi (chat): MCP does not have discovery unless you know the
exact MCP server endpoint. How do you find the endpoint?

Colin Perkins (chat): Remind me why this is better than a well-known
URL that serves some rich agent description format?

• Eric Rescorla: I'm still lost about the use case. How do most people
  discover it?

  • Answer (Nic): The way we split the use case is any of the above.
    You could be looking for specifically customer service, you
    could be looking at what an agent making text to image is doing.

• Jim Reid: Don't expect too much from DNS at this point in time. DNS
  dispatch discussion. Focus on spinning own focused WG.

3.d. AI-Agent Network Requirements and Protocols (ANP)

No one in queue

3.e. Dynamic Multi-agents Secured Collaboration (DMSC)

Someone was on mic but not in queue.

3.f. Agent Communication Interdomain Protocols (ACIP)

• Rohan Mahy: It would be very useful for someone to explain what
  existing protocols are used for human users or non-UI server
  communications, and what they don't do (if anything) that is needed
  for AI.

1. Work outside the IETF

• Jim Reid: How will the IETF interact and cooperate with those
  organisatoins? Liaison statements seems not adapted for this.

  • Answer (Suresh): ISO and other SDOs are setting up the World AI
    Standards exchange database

• Arashmid Akhavain: Will that stay open source?

  • Answer (Suresh): Individual outreach with A2A, MCP community.
    Still work in progress.

• Arnaud Taddei: Experiment a bit more with ways of exchanging between
  SDOs. For SG-17: having to deal with member states to members
  states. Different than in IETF.

  • Answer: It's not easy for just the leadership to send something.
    IETF is a ground-up way to do things needing consensus.

• Grace Rachmany: Decentralized Identity Foundation (DIF) from Linux
  Foundation. We have a Trusted Agents Working Group. We would like to
  collaborate.

  • Answer: Send me a link and I will add it to these slides.

1. Open discussion

• Linda Dunbar: Can the presenters highlight the key differences
  between "Agent" and regular Apps being used today? It would be
  useful to differentiate "Agents"

• Ted Hardie: Definition of the "new generation of applications"
  should not be based on the fact that they contain or are based on
  "AI" but similarly to other applications (e.g. IoT) and capture
  their charateristics (as applications) and how they will impact the
  underlying network functionalities

• Dan Druta: Nobody is going to do the Agent to agent just for the
  sake of AI. You are not going to find Agents on the internet. Let
  the market place to drive the requirement. Need some use cases.

• Paul Hoffman: Discovery is going to be super important here. DNS is
  for naming, but doesn't work for discovery.

• Roberta Robert: Definition of agents. Need discussion about agents'
  quality, what they're doing, the metrics that we need.

• Aijun Wang: Mentions definition of agent and embodied AI use cases.

• Mallory Knodel : Agree with Dan. Emphasize the importance on
  coordination.

• Peter Liu: Two communities: security and non-security. Need
  consolidation on use cases to help security experts step in and
  contribute to security requirements and threat modelling.

• Muhammad Jadoon: How consolidate agentic work within the IETF?

• Muhammad Usama Sardar: Security and non-security communities need to
  collaborate. Start with definition of AI and do problem statement.

• Daniel Huang: Need a holistic overview. Routing technology seems not
  well used.

• David Schinazi: IETF good at focused activities to solve a
  particular problem. Not effective a paradigm shifts.

• Zaheduzzaman Sarker: I see lots of similar things happening in
  different communities outside IETF on protocols where IETF also have
  developed solutions. Think IETF should better define and focus on
  use cases, defining requirements and putting all things in a kind of
  framework to see all things fit together.

• Jodi Ros Giralt: Start with use cases

• Lionel Morand: Protocol aspect study will finish in September 2027.
  3GPP will not develop any protocol solution. Good to have IETF to
  reference.

• Arnaud Taddei: Here we are not connecting machines (done).
  Connecting synthetic and human intelligences. More effort in
  udnerstanding the core requirements and have the right design. 2nd
  comment: why existing process are not enough. on SG17 side, wht is
  going to happen between now and IETF 126/July: 3 workshops, 1
  content week and 1 plenary. IETF going with side meetings since IETF

  123. Time to define a proper strategy. Look into what other
       groups/SDOs are doing and help progress the work of the IETF (e.g.
       ITU-T SG17)

• Haomian Zheng : New solution going to enhance the exisitng protocols
  rather than replace them. Sugget channeling the work to existing
  working groups rather than starting new working gruops. Coordinating
  work like today is necessary and useful, and happy to help.

• Yunefi Zhang: Happy to see so many people from different
  backgrounds. Several talks refer to embodied AI (vehicles and
  drones). Talking about interactions between phycial agents,and
  between physical and digital agents, is quite different from between
  digital and digital. We are missing the semantic interactoin between
  physical and digital.

• Jari Arkko: Need for stability and standards for industries that
  move a little slower. Need to determine areas IETF can contribute,
  modify existing protocols or become an expert in an area?

• Bing Liu: I would be useful to have a place for "incubation" of
  Agentic AI discussion.

• Peng Liu: A coordination group might be needed as well as a protocol
  analysis document, and we don't need to wait for the whole picture
  to be clear enough but just start with the clear one.

• Arashmid Akhavian: Nothing preventing groups of people willing of
  progressing focused work in existing WG or as individual work.

• Georgios Karagiannis: May be useful to have an interim meeting
  before next meeting to setup charters.

• Toerless Eckert: Push side meetings to interim meetings with proper
  agenda, context, and souces/materials.

• Jian Song: Need to think about the name of AI agent-to-agent. Stil a
  lot of other AI issues to be worked on.

• Roberta Robert: Where would AI agents security work be done in IETF?

• Cullen Jennings: Building useful standards is going to take time.
  Focus on developing solutions common to use cases.

• Xing Zhao: Putting all AI work in one group is not a good idea.
  Already many existing groups that can take on specific tasks.

1. Closing remarks and next steps
   • Andy Newton (ART AD): No clear answers to the BoF questions.
     Emerging themes. Possiblities for scoped, focused WG to find a
     place in IETF.

• Adrian: Should we have an ongoing coordination activity?

Poll:
Do you think the IETF should have an ongoing AI agent-to-agent
coordination activity?
yes: 100
no: 43
no op: 11