AGENTPROTO

Summary

The AGENTPROTO (Agent Communication Protocols) Working Group forming BOF session was held to discuss the motivations, requirements, and potential charter for standardizing protocols to support communication between AI agents, tools, and human users. The session was chaired by Orie Steele and Leslie Daigle, with Charles Eckel serving as the responsible Area Director (AD).

The discussion centered on the proposed charter's focus: a standards-track session/context protocol, and an informational framework to coordinate with other working groups (e.g., OAuth, WIMSE). While there was overwhelming consensus that a problem exists and the IETF is the right place to solve it, the poll results indicated that the draft charter's scope and deliverables require significant refinement before a working group is formally established.

Key Discussion Points

Introduction and Scope Boundaries

Chairs' Introduction: Co-chairs Orie Steele and Leslie Daigle opened the session by outlining the agenda and introducing the agentproto presentation.
Scoping Boundaries: Orie Steele clarified that the BOF is not a venue to redesign or modify external specifications, such as the Linux Foundation's Agent-to-Agent (A2A) protocol or the Model Context Protocol (MCP). The proposed work must focus on IETF-specific interoperability gaps.

Motivations and Requirements

Presentation: Laurent Ciavaglia presented the industry shift from isolated, single-agent applications to decentralized agent networks operating at Internet scale across administrative and trust boundaries. He emphasized the risk of protocol fragmentation and proposed analyzing recurring interaction patterns (user-to-agent, agent-to-tool, and agent-to-agent) to derive standardized protocol requirements.
Industry Alignment: Aijun Wang questioned the precise technical gaps not addressed by current industry implementations. Laurent responded that the effort aims to design a generalized solution rather than deploying case-by-case, non-interoperable adapters.

Multimodal Communication and Session Binding

Presentation: Suresh Krishnan outlined the diverse transport requirements for multi-agent workflows, including real-time media (sub-100ms voice/video), interactive text, and bulk file transfers.
Binding Transports: Suresh argued that these disparate transport channels (potentially spanning WebRTC, Media over QUIC, or WebTransport) must be bound together logically so they can be managed, aligned, and terminated as a single unit.
Specificity: George Fletcher questioned whether this session binding was strictly AI-agent specific. Suresh clarified that while the mechanisms might be general, the immediate prioritization and requirements are driven by agentic use cases.

Session Protocol and Context Management

Presentation: Hisham Al-Khasib addressed the necessity of a protocol to manage stateful, multi-party contexts across long-lived, non-deterministic agent workflows. The protocol must robustly handle interruptions, pauses, reconnects, agent failures, and agent replacements.
Context vs. Session: Ted Hardie proposed reframing the "session protocol" as a "context propagation/signaling protocol." He noted that unlike point-to-point ephemeral protocols (such as SIP or WebRTC), an agent protocol must propagate multi-party context across multiple hops and trust boundaries over time.
Minimal Primitives: Brian Trammell supported focusing on a minimum viable set of primitives to manage the dynamicity of agent networks.
Security & Authorization Complexity: Stephen Farrell and Justin Richer expressed concerns about the immense complexity of managing and attenuating rights across trust boundaries. Stephen emphasized the need for a practical proof-of-concept before committing to a protocol, fearing that overly complex credential propagation might lead to over-authorization. Colin Perkins countered that securely partitioning and encrypting sections of context across hops is a classic, solvable security problem that the IETF is uniquely equipped to address compared to existing fragile ad-hoc solutions.
SIP Analogy: Jonathan Rosenberg compared the work to SIP but emphasized that agentic contexts are longer-lived and must persist across various devices and trust domains.

Protocol Framework and WG Coordination

Presentation: Daping Liu presented the proposed standards-track framework document, which is designed to identify the necessary building blocks for agentic systems and coordinate work across other working groups (e.g., WIMSE and OAuth) without duplicating their efforts.
Terminology & Framing Concerns: Arnaud Taddei argued that the term "AI Agent" lacks a stable, consensus-driven definition across the industry. He recommended dropping the framework document entirely to focus strictly on multimodal transport, citing overlapping work in ITU-T SG17.
Missing Elements: Pam Dingle highlighted several missing dimensions in the current charter, including orchestration definitions, cross-domain performance considerations, and session revocation mechanisms (e.g., how to safely revoke a delegated action if an agent goes offline).
Open Source Collaboration: Tong Feng and Aijun Wang asked if the proponents were actively coordinating with external open-source projects. Daping Liu clarified that core maintainers of specifications like MCP are actively participating via GitHub pull requests to ensure the proposed IETF work aligns with and reuses existing deployment patterns.

Decisions and Action Items

Session Polls

The following authoritative polls were conducted during the session:

Do you think that the problem statement is well-understood, solvable, and useful to solve?
- Yes: 124, No: 72, No_opinion: 14 (Total: 320)
Is the need for interoperability clear?
- Yes: 155, No: 30, No_opinion: 8 (Total: 318)
Do you think the IETF is the right place to do this work?
- Yes: 158, No: 30, No_opinion: 14 (Total: 315)
Do you think the initial scope from the charter we just discussed is correct?
- Yes: 38, No: 124, No_opinion: 40 (Total: 316)
Do you think the initial deliverables from the charter we just discussed are correct?
- Yes: 92, No: 58, No_opinion: 49 (Total: 316)
Do you think a WG should be formed on this topic, with a charter based on the draft charter we just discussed?
- Yes: 154, No: 51, No_opinion: 5 (Total: 316)
Are you willing to help write and/or review drafts in this potential WG?
- Yes: 137, No: 17, No_opinion: 40 (Total: 317)
Are you willing/planning to implement the protocols of the eventual WG?
- Yes: 42, No: 28, No_opinion: 40 (Total: 314)

Key Decisions

Charter Revision Required: While there is strong support to form a Working Group (Poll 6: 154 Yes vs. 51 No) and agreement that the problem is solvable and belongs in the IETF, the specific scope of the current draft charter is not supported (Poll 4: 38 Yes vs. 124 No). The charter must be extensively revised based on the feedback received.
Separation of Mailing Lists: To reduce noise and maintain focus, the discussions regarding charter drafting and IETF-specific protocol engineering will be moved off the general agent-to-agent mailing list.

Next Steps

Mailing List Creation: The AD (Charles Eckel) and chairs will establish a new, dedicated public mailing list (proposed name: agentproto) to host focused discussion on refining the charter and deliverables.
Charter Refinement via GitHub: Interested participants are encouraged to submit Pull Requests to the draft charter repository to address the feedback raised (e.g., reframing "session" to "context propagation," clarifying agent definitions, addressing cross-domain revocation, and scoping the framework).
Coordination with Other WGs: Proponents will continue to coordinate with OAuth and WIMSE regarding identity and authorization models. A dedicated discussion on agent authorization and behavioral policies is scheduled for the upcoming OAuth session.
WG Formation Path: Once the charter has been successfully refined on the new mailing list, the AD will determine if the group can be chartered directly or if a second BOF is required for the next IETF meeting.