[agent2agent] Re: New I-D : A Dimensional Model for Characterizing AI Agent Protocol Proposals and Their Substrates

["Parisa Foroughi (Nokia)" <[e-mail]>]

Dear Chris,
Thanks for the feedback and for pointing me to AGTP. This is a bit long, but I hope it clarifies the positioning cleanly.
Agreed. AGTP is an interesting kind of substrate: it carries MCP, yet its verbs are agent semantics themselves, so it doubles as an agent-protocol surface. Into the substrate column it goes, and it earns its own dimensional signature.
By "dimensional signature" I mean that when AGTP carries MCP, it's playing the substrate role and sits in the transport column. But used directly i.e., as an agent issuing DELEGATE or DISCOVER over the wire, its own verbs are the agent-protocol surface, so that usage takes a value on each dimension: D2 from its cadence, D4 from DELEGATE, D5 from DISCOVER/Presence, and the independent-verifiability value from its signed identity headers. This doesn't strain the model; it's in fact its central move that the dimensions describe what an "exchange" does, the substrate column says what a "role" is bound to. AGTP occupies both and I think it makes it the cleanest dual-role case here.
On your three notes:

  1.
Agreed — and that's a companion model rather than a gap in this one. This draft models "exchanges"; the substrate column is an anchor, not a characterization of the substrate's internals. The effect you meant to surface is already in it: the substrate binding fixes and decides which dimensional values are reachable, so one that can't do server-initiated messages forces D1 = none. A substrate-side model, if you're interested, would complement this cleanly, and the two should ideally cross-reference.
  2.
If by "axis" you mean a new dimension, I don't think it lands as one. The distinctions those verbs carry are already dimensioned — QUERY vs EXECUTE is D2 plus operation nature,  DELEGATE is D4, DISCOVER is D5. What's genuinely new is where intent lives: a first-class wire primitive versus a field in the message body. That's a substrate-realization choice, the same split as D1,  real and consequential, but not a separate axis.
  3.
Two things here, and they separate cleanly. Where the chain is carried : AGTP headers versus an OAuth token in the body is substrate-realization; D4's value is the same either way, so it doesn't move D4. But "what the receiving agent can verify independently" is a different property, and it's not D4 at all — it's the independent-verifiability axis another contributor raised from the receipt side as well. You've arrived at it from the substrate side. Two routes to the same property are a strong signal, and it's under active consideration for exactly that reason.

In short, AGTP is worth adding in the next revision, both as a substrate entry and as a clean illustration of why reading the dimensions correctly means separating a protocol's role from its signature.
Looking forward to continuing in Vienna.
Best regards,
Parisa



________________________________
From: Chris Hood <[e-mail]>
Sent: Friday, July 10, 2026 5:57 PM
To: Parisa Foroughi (Nokia) <parisa.foroughi=[e-mail]>; [e-mail] <[e-mail]>
Subject: [agent2agent] Re: New I-D : A Dimensional Model for Characterizing AI Agent Protocol Proposals and Their Substrates

You don't often get email from [e-mail]. Learn why this is important<https://aka.ms/LearnAboutSenderIdentification>

CAUTION: This is an external email. Please be very careful when clicking links or opening attachments. See the URL nok.it/ext for additional information.



Parisa,

Thanks for sharing. I'd like to draw your attention to the Agent Transfer Protocol (AGTP), a substrate that sits next to HTTP as a proposed option for agent architecture.

You highlight the agent protocol layer over the substrate layer, but AGTP is an agent-focused substrate. I recognize the comparison table isn't exhaustive, but felt it was worth flagging that at least one agent-focused substrate exists as an option.

AGTP places agent identity, scope, semantics, and trust on the wire. It supports endpoint binding and composes with other agent protocols. AGTP-MCP is working and available on GitHub as one example.

As for your model, here are three short notes:

1. Table 2's substrate column names which substrate each proposal binds to, but the substrates themselves have properties worth dimensioning. What a substrate provides (identity model, trust anchoring, session semantics) shapes what the agent protocol above can express.

2. A substrate-layer axis for method vocabulary would surface a real design choice. HTTP verbs were built for document retrieval; agent-native methods (QUERY, DISCOVER, DELEGATE, EXECUTE) express intent at the wire. That difference reshapes the layers above.

3. Authorization derivation (D4) sits at the agent-protocol layer in your model, but substrates can carry delegation lineage directly. Where the chain lives changes what the mediator can preserve and what the receiving agent can verify independently.

For Agent Transfer, the full suite of specs is available at agtp.io or by searching datatracker for agtp.

Chris


-------- Original message --------
From: "Parisa Foroughi (Nokia)" <parisa.foroughi=[e-mail]>
Date: 7/10/26 8:25 AM (GMT-08:00)
To: [e-mail]
Subject: [agent2agent] Re: New I-D : A Dimensional Model for Characterizing AI Agent Protocol Proposals and Their Substrates

Dear all,

Strangely, the link appears broken.

Please use the following instead:  https://datatracker.ietf.org/doc/draft-foroughi-agent-protocol-dimensions/

Regards,
Parisa

PS: Thanks, Adrian, for raising the problem.
________________________________
From: Parisa Foroughi (Nokia) <parisa.foroughi=[e-mail]>
Sent: Friday, July 10, 2026 10:55 AM
To: [e-mail] <[e-mail]>
Subject: [agent2agent] New I-D : A Dimensional Model for Characterizing AI Agent Protocol Proposals and Their Substrates

You don't often get email from parisa.foroughi=[e-mail]. Learn why this is important<https://aka.ms/LearnAboutSenderIdentification>
Dear IETFers,
I have published a new I-D proposing a dimensional model for characterizing agentic protocols and their substrates:
https://datatracker.ietf.org/doc/draft-foroughi-agent-protocol-dimensions/<https://teams.microsoft.com/l/message/19:785e0265-[e-mail]/1783546826122?context=%7B%22contextType%22%3A%22chat%22%7D>
The motivation is simple: help the community denoise the fast-growing agentic landscape and position concerns appropriately. Phrases like "we need a session protocol" or "we need cross-domain support" bundle several concerns across transport, identity, agent protocol, and orchestration layers. The conversation loses precision, and the technical work slowly derails. The draft is an analytical instrument that routes each concern to its proper layer — a dimensional choice at the agent protocol layer, a substrate-inherited property, or broader-scope work — so discussions become precise about where each concern actually lives.
I hope you find the model useful, and I would very much welcome feedback, especially on the dimensional value domains and on the comparison table, which is deliberately marked as a starting point. If any of this resonates and you would like to collaborate on sharpening it further, I would be glad to hear from you.
I will be attending the Vienna meeting — my first in person IETF meeting. Looking forward to discussions on the list or in Vienna.
Best regards,
Parisa