[{"author": "Daniel Gillmor", "text": "<p>it looks like this is trying to give IETF a deadline of about 1 year to have a stable WG draft.  But this isn't a WG-forming BoF.  Is this timeline plausible?</p>", "time": "2026-07-22T07:13:52.000Z"}, {"author": "Aijun Wang", "text": "<p>This is the expection from 3GPP</p>", "time": "2026-07-22T07:14:35.000Z"}, {"author": "Philipp Tiesel", "text": "<p>We should treat all AI related deadlines with more than a grain of salt.</p>", "time": "2026-07-22T07:15:03.000Z"}, {"author": "Daniel Gillmor", "text": "<p>and it apparently envisions the network itself acting as an intermediary between the user agent and some sort of agent -- why would I choose my network operator to negotiate on my behalf for any service, let alone an AI-specific one?</p>", "time": "2026-07-22T07:16:36.000Z"}, {"author": "Aijun Wang", "text": "<p>The opeator can do some help for the communication</p>", "time": "2026-07-22T07:17:37.000Z"}, {"author": "Mike Blanche", "text": "<p>...by carrying the packets to your chosen AI platform...</p>", "time": "2026-07-22T07:18:09.000Z"}, {"author": "Lionel Morand", "text": "<p>services would be network services, e.g. connectivity services, mobility management, etc.</p>", "time": "2026-07-22T07:18:10.000Z"}, {"author": "Carolina Caeiro", "text": "<p>Did anyone define \"AI Agent Gateway-mediated collaboration\" yet?</p>", "time": "2026-07-22T07:18:14.000Z"}, {"author": "Mike Blanche", "text": "<p>@Carolina: no</p>", "time": "2026-07-22T07:18:36.000Z"}, {"author": "Linda Dunbar", "text": "<p>also need to consider AI agents carried by 5G/6G services</p>", "time": "2026-07-22T07:18:47.000Z"}, {"author": "Brian Carpenter", "text": "<p>@Daniel, the question should be: why do the operators want to do this for you?</p>", "time": "2026-07-22T07:18:55.000Z"}, {"author": "Orie Steele", "text": "<p>it might be helpful to provide examples of other NF activities that exist today, pre-agent.</p>", "time": "2026-07-22T07:19:19.000Z"}, {"author": "Daniel Gillmor", "text": "<p>@Brian: that's another good question, not a replacement for my question!</p>", "time": "2026-07-22T07:19:23.000Z"}, {"author": "Artur Hecker", "text": "<p>@Brian: because they can't use volume-based charging anymore? :-)</p>", "time": "2026-07-22T07:19:25.000Z"}, {"author": "Brian Carpenter", "text": "<p>@Artur: Bingo!</p>", "time": "2026-07-22T07:19:45.000Z"}, {"author": "Daniel Gillmor", "text": "<p>chairs: please use the timer for the speakers so that they're aware of the clock</p>", "time": "2026-07-22T07:20:10.000Z"}, {"author": "Joe Clarke", "text": "<p>We did.</p>", "time": "2026-07-22T07:20:23.000Z"}, {"author": "Daniel Gillmor", "text": "<p>OK, it's not visible remotely (or maybe i missed it).  sorry for the noise!</p>", "time": "2026-07-22T07:20:42.000Z"}, {"author": "Artur Hecker", "text": "<p>@Daniel: it could be required for compliance of the AI service, depends on the service and local legislation.</p>", "time": "2026-07-22T07:21:09.000Z"}, {"author": "Aaron Parecki", "text": "<p>the timer ran out a while ago</p>", "time": "2026-07-22T07:21:14.000Z"}, {"author": "Daniel Gillmor", "text": "<p>if it runs out, doesn't it show in the negative?</p>", "time": "2026-07-22T07:21:32.000Z"}, {"author": "Tommy Jensen", "text": "<p>It should still show blinking 0:00 in red</p>", "time": "2026-07-22T07:21:39.000Z"}, {"author": "Tommy Jensen", "text": "<p>ok this timer shows up</p>", "time": "2026-07-22T07:21:56.000Z"}, {"author": "Daniel Gillmor", "text": "<p><span aria-label=\"+1\" class=\"emoji emoji-1f44d\" role=\"img\" title=\"+1\">:+1:</span></p>", "time": "2026-07-22T07:22:04.000Z"}, {"author": "Joe Clarke", "text": "<p>It did.  She couldn't see the comfort monitor clearly, though.</p>", "time": "2026-07-22T07:22:11.000Z"}, {"author": "Brian Carpenter", "text": "<p>Once again I wonder why people use tiny fonts (illegible on my screen)</p>", "time": "2026-07-22T07:23:30.000Z"}, {"author": "Andrew Newton", "text": "<p>Yes, the presenters video monitor is below stage height... anybody with bifocals or progressives would have difficulty reading it.</p>", "time": "2026-07-22T07:23:32.000Z"}, {"author": "Pete Resnick", "text": "<p>@Brian Carpenter : Very much a non-native English speaker thing - Make sure you have everything you want to say on the slide so everyone can follow along.</p>", "time": "2026-07-22T07:25:15.000Z"}, {"author": "Pete Resnick", "text": "<p>Who is the cognizant AD for this session?</p>", "time": "2026-07-22T07:25:54.000Z"}, {"author": "Joe Clarke", "text": "<p>Charles Eckel</p>", "time": "2026-07-22T07:26:43.000Z"}, {"author": "Artur Hecker", "text": "<p>sorry, but it has nothing to do with \"native\", @Pete. I have seen bad slides from native speakers as well. Maybe proficiency has something to with it, but even this can be discussed.</p>", "time": "2026-07-22T07:28:49.000Z"}, {"author": "Carolina Caeiro", "text": "<p>+1 Artur</p>", "time": "2026-07-22T07:29:42.000Z"}, {"author": "Dhruv Dhody", "text": "<p>I wish there was more of what needs to be standardized for interoperation here</p>", "time": "2026-07-22T07:31:00.000Z"}, {"author": "Brian Carpenter", "text": "<p>Re his talk: in the late 1980s we built email gateways and file transfer gateways because of diverse protocols. They were nightmares. These would be worse, much worse.</p>", "time": "2026-07-22T07:31:14.000Z"}, {"author": "Philipp Tiesel", "text": "<p>It's funny to see how different the ideas what an agent gateway ist\u2026 based on my dayjob, I would have assumed it's a proxy that helps with authentication and authorization, does some rate limiting and emits billing information\u2026</p>", "time": "2026-07-22T07:33:16.000Z"}, {"author": "Em Lauber", "text": "<p>+1 @philipp</p>", "time": "2026-07-22T07:34:12.000Z"}, {"author": "Antoine Fressancourt", "text": "<p>You also have a routing to LLM provider aspect, as in VLLM's semantic router</p>", "time": "2026-07-22T07:34:18.000Z"}, {"author": "Philipp Tiesel", "text": "<p>@Antoine: Request routing like a really fancy forward proxy, not as a internet packet router.</p>", "time": "2026-07-22T07:36:32.000Z"}, {"author": "Roberta Robert", "text": "<p>About the usecases: why are we considering using agents for contexts where ethical/fault tolerant fluxes are in play? We still can't guarantee the behaviour expected from them, there's issues in sensor swarms... why are we doing this, again?</p>", "time": "2026-07-22T07:37:07.000Z"}, {"author": "Artur Hecker", "text": "<p>@Roberta: I agree very much.  But, if history has taught me anything is that as soon as some nonsense is available, people start using it for all kinds of things, including for those which could be perfectly done without that nonsense.</p>", "time": "2026-07-22T07:38:43.000Z"}, {"author": "Philipp Tiesel", "text": "<p>@Roberta: Because we can and everyone values speed over considerate action</p>", "time": "2026-07-22T07:38:46.000Z"}, {"author": "Artur Hecker", "text": "<p>do i need WiFi or internet access in my dishwasher? or for what exactly, since it cannot offload itself? did you try to buy one without recently?</p>", "time": "2026-07-22T07:40:36.000Z"}, {"author": "Marie-Jose Montpetit", "text": "<p>There is a risk that the agent gateway becomes a bottleneck and more importantly prevents autonomous (multi)agent operations where agents are combined based on what functionality is needed.</p>", "time": "2026-07-22T07:41:18.000Z"}, {"author": "Antoine Fressancourt", "text": "<p>@Philipp I agree, maybe the router naming comes from an expectation that they are perceived as having a more recognized role than proxies</p>", "time": "2026-07-22T07:41:31.000Z"}, {"author": "Artur Hecker", "text": "<p>Marie-Jose: some would call it a \"feature\" or \"security function\" :-)</p>", "time": "2026-07-22T07:41:59.000Z"}, {"author": "Linda Dunbar", "text": "<p>@Marie-Jose Montpetit there could be many Gateways. Multiple agents anchored to a gw</p>", "time": "2026-07-22T07:42:20.000Z"}, {"author": "Hesham Moussa", "text": "<p>@Linda, is the assumption that the 'many' gateways will be synced up so each has some sort of a global view?</p>", "time": "2026-07-22T07:43:43.000Z"}, {"author": "Linda Dunbar", "text": "<p>I believe the scope is NOT wide internet</p>", "time": "2026-07-22T07:44:53.000Z"}, {"author": "Antoine Fressancourt", "text": "<p>I get the bottleneck aspect of intermediation gateways, here it shows as a logical infrastructure, I am quite sure we can have distributed versions of the logic element. The question IMHO is whether there is a role and protocol work needed for this logical gateway in the IETF</p>", "time": "2026-07-22T07:46:03.000Z"}, {"author": "Bing Liu", "text": "<p>For scalable and trusted Agent interconnection, there would be bottleneck without gateways, e.g. group communication, multicast streaming etc.</p>", "time": "2026-07-22T07:46:38.000Z"}, {"author": "Antoine Fressancourt", "text": "<p>Bearing in mind that proxies are commonly deployed, especially in corporate environments, without clear standards</p>", "time": "2026-07-22T07:47:20.000Z"}, {"author": "Brian Carpenter", "text": "<p>@Antoine, you can only distribute if the gateway action is stateless</p>", "time": "2026-07-22T07:47:48.000Z"}, {"author": "Hesham Moussa", "text": "<p>@bing can we consider zoom meetings as group communication setup? I don't think we need gateways for that.</p>", "time": "2026-07-22T07:48:12.000Z"}, {"author": "Linda Dunbar", "text": "<p>@Antoine Fressancourt: Agent GW is similar to API GW</p>", "time": "2026-07-22T07:48:15.000Z"}, {"author": "Laurent Ciavaglia", "text": "<p>Folks, there are some interesting points being made in the MeetEcho chat, please bring them for discussion in Part 4  if you want them to be addressed live</p>", "time": "2026-07-22T07:48:40.000Z"}, {"author": "Antoine Fressancourt", "text": "<p>@Brian I disagree, you can distribute with session stickiness to an instance (e.g. CATS work)</p>", "time": "2026-07-22T07:48:55.000Z"}, {"author": "Brian Carpenter", "text": "<p>Yes, maybe, but it makes routing harder</p>", "time": "2026-07-22T07:49:22.000Z"}, {"author": "George Fletcher", "text": "<p>It seems like delegation and authorization are going to be very complex if the gateway is doing lots of transformations.</p>", "time": "2026-07-22T07:49:49.000Z"}, {"author": "Antoine Fressancourt", "text": "<p>@Brian my own experience with SIP proxies is that protocol developers can be very imaginative with session ID semantics to serve some purposes</p>", "time": "2026-07-22T07:50:28.000Z"}, {"author": "Harald Alvestrand", "text": "<p>Reminds me of email in the 90s. Gateways = pain.</p>", "time": "2026-07-22T07:50:46.000Z"}, {"author": "Philipp Tiesel", "text": "<p>@Brian: classic problem for API gateways: Someone brings a strict rate limiting requirement and everything dies on waiting for consistent rate limiting data</p>", "time": "2026-07-22T07:51:12.000Z"}, {"author": "Linda Dunbar", "text": "<p>@Hesham Moussa: zoom does have the anchor point. Multiple clients anchored to an Anchor</p>", "time": "2026-07-22T07:51:36.000Z"}, {"author": "Brian Carpenter", "text": "<p>@Harald, +1. Everyone likes standards, so let's invent a new one.</p>", "time": "2026-07-22T07:51:37.000Z"}, {"author": "Bing Liu", "text": "<p>@hesham Zoom/Wechat  or any group communication system would need a central point, maybe not called GW. For Agents across platform/orgs, there needs to be a third-party central point, which could be a GW of a carrier.</p>", "time": "2026-07-22T07:51:48.000Z"}, {"author": "Mike Blanche", "text": "<p>is it a requirement that your gateway is run by your telco/carrier?</p>", "time": "2026-07-22T07:53:35.000Z"}, {"author": "Hesham Moussa", "text": "<p>I agree, I just don't see why would agents which live on application layer will want to be anchored on GW of a carrier</p>", "time": "2026-07-22T07:53:41.000Z"}, {"author": "Antoine Fressancourt", "text": "<p>@George cross domain authentication / delegation would actually be easier with gateways sitting at domain borders and adapting token semantics</p>", "time": "2026-07-22T07:53:57.000Z"}, {"author": "Linda Dunbar", "text": "<p>@Mike Blanche: not necessarily</p>", "time": "2026-07-22T07:54:02.000Z"}, {"author": "Chris Hood", "text": "<p>here.</p>", "time": "2026-07-22T07:54:25.000Z"}, {"author": "Bing Liu", "text": "<p>@Mike it is a possible instance, not a requirement</p>", "time": "2026-07-22T07:54:36.000Z"}, {"author": "Antoine Fressancourt", "text": "<p>(e.g. Token exchange)</p>", "time": "2026-07-22T07:54:38.000Z"}, {"author": "Philipp Tiesel", "text": "<p>\u2026but a telco providing agent access to its services is an interesting use case</p>", "time": "2026-07-22T07:54:50.000Z"}, {"author": "Brian Carpenter", "text": "<p>AI protocols are a perfect example of <a href=\"https://xkcd.com/927/\">https://xkcd.com/927/</a> and the gateway may well end up as #15</p>", "time": "2026-07-22T07:55:03.000Z"}, {"author": "Sam Betts", "text": "<p>In LF Agntcy project we've been working on a protocol and implementation called SLIM which is a decentralised group messaging system based on MLS purpose built for Agents. </p>\n<p>We can carry semantic protocols like A2A over the top, or simply direct messaging with group broadcast or point to point requests.</p>", "time": "2026-07-22T07:56:01.000Z"}, {"author": "Bing Liu", "text": "<p>the essential is who can be a trusted third-party</p>", "time": "2026-07-22T07:56:17.000Z"}, {"author": "Luigi Iannone", "text": "<p>@Brian: But gateways will not replace existing protocols</p>", "time": "2026-07-22T07:56:22.000Z"}, {"author": "Lionel Morand", "text": "<p>We could discuss the need for GW. But... there are there. Please refer to the benchmark provided by the agentgateway LF project. Now, we should concentrate on the main question: is there anything to be standardized by IETF on this area?</p>", "time": "2026-07-22T07:56:26.000Z"}, {"author": "Tommy Jensen", "text": "<p>everything on this slide seems to be applicable to enterprise scenarios where the owners of the data and the human the agent represents are different, rather than the same</p>", "time": "2026-07-22T07:56:41.000Z"}, {"author": "Tommy Jensen", "text": "<p>or there's a link I am missing in terms of the value, quite possibly on the middle column (token protection)</p>", "time": "2026-07-22T07:57:09.000Z"}, {"author": "Artur Hecker", "text": "<p>Brian: the good things about the standards is that you can choose the one you like (A. Tannenbaum)</p>", "time": "2026-07-22T07:57:31.000Z"}, {"author": "Antoine Fressancourt", "text": "<p>@Lionel Then it goes down to where do we need proxy-compatible protocol adaptations such as the Via header in SIP so gateways can have an anchor to operate on ongoing exchanges through them</p>", "time": "2026-07-22T07:58:18.000Z"}, {"author": "Brian Carpenter", "text": "<p>@Luigi, true, but the gateway will need to define its own internal lingua franca</p>", "time": "2026-07-22T07:58:36.000Z"}, {"author": "George Fletcher", "text": "<p>@Antoine it adds a super trusted component into the path as the GW will need to \u201cmint\u201d new tokens and mechanisms for the transformations unless the GW reaches out to the originating entity for approval. This is the complexity that needs to be accounted for in the trusted delegation chains.</p>", "time": "2026-07-22T07:59:17.000Z"}, {"author": "Luigi Iannone", "text": "<p>For what purpose? Synch? We can re-use existing technologies for that.</p>", "time": "2026-07-22T07:59:26.000Z"}, {"author": "Luigi Iannone", "text": "<p>The idea of gateway = mediator makes sense</p>", "time": "2026-07-22T07:59:47.000Z"}, {"author": "Chris Hood", "text": "<p>but what if agents don't need an IP address?</p>", "time": "2026-07-22T08:00:07.000Z"}, {"author": "Antoine Fressancourt", "text": "<p>@George As lionel mentionned, gateways will be there doing this thing in the shadows, so having protocol adaptations to address some proxy needs can go with the ability to control how proxy behave, and see what proxies are doing</p>", "time": "2026-07-22T08:00:51.000Z"}, {"author": "Artur Hecker", "text": "<p>Agreed with Lionel. Whether we like the gateways or not, they will there. An organization with 1000 internal agents will do hell letting them just go to the internet, this is not going to happen. Whether we standardize anything or not, this will be blocked, controlled, DMZed, mediated, scrutinized, filtered, etc.</p>", "time": "2026-07-22T08:01:05.000Z"}, {"author": "Chris Hood", "text": "<p>^ maybe</p>", "time": "2026-07-22T08:01:29.000Z"}, {"author": "Hesham Moussa", "text": "<p>@lionel, it depends on the nature of a GW...is it just a server? Or is it a new type of box that exposes new types of APIs...it is not still clear to me what are they</p>", "time": "2026-07-22T08:01:31.000Z"}, {"author": "Luigi Iannone", "text": "<p>@Artur: +1 you are right</p>", "time": "2026-07-22T08:02:08.000Z"}, {"author": "Bing Liu", "text": "<p>@Sam Hi Sam, the SLIM is an inspiring solution, who do you envision operating and maintaining the SLIM Node?</p>", "time": "2026-07-22T08:02:10.000Z"}, {"author": "Antoine Fressancourt", "text": "<p>I would rather have open standard mechanisms in the open than DPI based proprietary stuff I can't transparently observe</p>", "time": "2026-07-22T08:02:12.000Z"}, {"author": "Thomas Hardjono", "text": "<p>Enterprises concerned about their employees agents \u201cinteracting\u201d with unapproved external services (or external Agents) can always run an AI-driven filter for all the traffic.</p>", "time": "2026-07-22T08:02:29.000Z"}, {"author": "Artur Hecker", "text": "<p>... and will :-)</p>", "time": "2026-07-22T08:02:52.000Z"}, {"author": "Chris Hood", "text": "<p>the \"internet of agents\" vs. the \"intranet of agents.\"</p>", "time": "2026-07-22T08:03:02.000Z"}, {"author": "Thomas Hardjono", "text": "<p>Do these need a \"gateway\"?</p>", "time": "2026-07-22T08:03:34.000Z"}, {"author": "Charles Eckel", "text": "<p>One of the goals is to better understand for the stated use cases and requirements whether an agent gw is merely a convenient implementation option or an actual requirement.</p>", "time": "2026-07-22T08:03:40.000Z"}, {"author": "Tommy Jensen", "text": "<p>Notably, enterprises already deploy such \"AI Gateway\" services today, as the market is creating proprietary solutions to that, tldr using TLS termination.</p>", "time": "2026-07-22T08:03:51.000Z"}, {"author": "Thomas Hardjono", "text": "<p>Tommy: +1</p>", "time": "2026-07-22T08:04:09.000Z"}, {"author": "Brian Carpenter", "text": "<p>gateway, firewall, protocol translator, all in one box? Good luck with that.</p>", "time": "2026-07-22T08:04:29.000Z"}, {"author": "Antoine Fressancourt", "text": "<p>@Tommy indeed, so is it a desirable situation, or do we think a standard would improve the situation?</p>", "time": "2026-07-22T08:04:40.000Z"}, {"author": "Luigi Iannone", "text": "<p>I wonder whether the name \"gateway\" is the most appropriate as it seems a bit misleading....</p>", "time": "2026-07-22T08:04:43.000Z"}, {"author": "Luigi Iannone", "text": "<p>Or reducing the scope..</p>", "time": "2026-07-22T08:05:22.000Z"}, {"author": "Artur Hecker", "text": "<p>Charles: I agree. The thing seem to become an architectural entity, we can then see from the IETF perspective, whether it should have standard protocols, e.g., for sync, for control, for discovery (cf. DAWN), for registration, for search, for negotiations/consensus, etc.</p>", "time": "2026-07-22T08:05:24.000Z"}, {"author": "Thomas Hardjono", "text": "<p>There\u2019s a huge business opportunity for new \u201cAI Firewall\u201d products.</p>", "time": "2026-07-22T08:05:34.000Z"}, {"author": "Tommy Jensen", "text": "<blockquote>\n<p>do we think a standard would improve the situation?</p>\n</blockquote>\n<p>Indeed the question, Antoine.</p>", "time": "2026-07-22T08:06:12.000Z"}, {"author": "Antoine Fressancourt", "text": "<p>My answer would lean towards \"yes\", but I think it was clear ;-)</p>", "time": "2026-07-22T08:06:54.000Z"}, {"author": "Tommy Jensen", "text": "<p>Very closely aligned with Charles' point (responsible AD, for those who may not be aware)</p>", "time": "2026-07-22T08:07:18.000Z"}, {"author": "Tommy Jensen", "text": "<p>(Charles is)</p>", "time": "2026-07-22T08:07:30.000Z"}, {"author": "Harald Alvestrand", "text": "<p>On this slide, \"network\" should be \"network of agents\", not \"network of of routers\".</p>", "time": "2026-07-22T08:08:26.000Z"}, {"author": "Hesham Moussa", "text": "<p>Agree with Charles... implementation trick to perhaps address potential scalability issues?</p>", "time": "2026-07-22T08:08:28.000Z"}, {"author": "Cullen Jennings", "text": "<p>So with HTTP Rest style API, lots of vendors wrote HTTP API gateways that provided high level functions that are similar in scope of what they do to the things talked about here. Interesting to look at what standardization they ended up needing.</p>", "time": "2026-07-22T08:08:37.000Z"}, {"author": "Artur Hecker", "text": "<p>well, SOCKS was standardized :-)</p>", "time": "2026-07-22T08:09:30.000Z"}, {"author": "Harald Alvestrand", "text": "<p>I ripped SOCKS support out of libwebrtc because I couldn't find anyone interested in using it.....</p>", "time": "2026-07-22T08:10:05.000Z"}, {"author": "Antoine Fressancourt", "text": "<p>I would think that standardization maybe more desirable for users than for potential gateway providers</p>", "time": "2026-07-22T08:10:36.000Z"}, {"author": "Artur Hecker", "text": "<p>Harald :-)))</p>", "time": "2026-07-22T08:10:55.000Z"}, {"author": "Roberta Robert", "text": "<p>I may be losing some context, but AI Firewalls are not working above 80% marks because it is usually using ai to try to sensor prompts, and the semantic capacities of the tech itself makes infeasible to predict all the possible patterns, which is an ongoing issue on how to protect and prevent bad usage. And this before thinking about anything related to auth fluxes.</p>", "time": "2026-07-22T08:11:00.000Z"}, {"author": "Lionel Morand", "text": "<p>@cullen +1</p>", "time": "2026-07-22T08:11:06.000Z"}, {"author": "Harald Alvestrand", "text": "<p>All providers that can imagine being the dominant player in a space will resist being constrained by standards....</p>", "time": "2026-07-22T08:11:22.000Z"}, {"author": "Mike Blanche", "text": "<p>@Antoine new entrants would prefer standardisation to make it easier for customers to switch to their gateway. incumbent providers, obviously not so much...</p>", "time": "2026-07-22T08:12:01.000Z"}, {"author": "Linda Dunbar", "text": "<p>@Cullen Jennings: That is a useful analogy: not every AI Agent Gateway function needs standardization. Standards are mainly needed for cross-vendor or cross-domain interoperability, such as identity, authorization context, provenance, and policy exchange.</p>", "time": "2026-07-22T08:12:26.000Z"}, {"author": "Linda Dunbar", "text": "<p>@Charles Eckel: Agreed. The key is to distinguish functions that can be implemented directly by agents or applications from those that require a gateway to enforce shared policy, trust, security, or interoperability across domains.</p>", "time": "2026-07-22T08:14:21.000Z"}, {"author": "Roberta Robert", "text": "<p>I am really enjoying this session, I think this has been well scoped, and the proposal about working towards collaboration with different agencies to bring a standard to this gateway part is really interesting to maintain an agnostic approach, not dependable on proprietary approaches.</p>", "time": "2026-07-22T08:16:21.000Z"}, {"author": "Jari Arkko", "text": "<p>In this discussion I think it might be useful to distinguish several different functions, clearly discovery is part it. Closely related to that is policy, as you can't randomly discover things without also knowing who is allowed to do what. Then there's the concept of connection routing which can be either direct or via some kind of on-path entity. Finally there's observability which is not linked only to on-path observation but also things that we discussed in AUDIT, e.g., in-band communication of traces.</p>", "time": "2026-07-22T08:17:24.000Z"}, {"author": "Sam Betts", "text": "<p>@bing SLIM is built so that it can be deployed within an organisation or cross organisation. But if people were willing to host public nodes there is no reason why we couldn't have an internet scale SLIM network.<br>\nAs I mentioned its decentralised / peer to peer so with a big enough network one or more members going down wouldn't impact the network.</p>", "time": "2026-07-22T08:18:53.000Z"}, {"author": "Tommy Jensen", "text": "<p>^^^ that makes me wonder how dawn and dmsc overlap (is discovery solely a gateway problem if the client only worries about finding/configuring a gateway?)</p>", "time": "2026-07-22T08:19:54.000Z"}, {"author": "Chris Hood", "text": "<p>short of slim, agtp proposes node based network discovery of agents.</p>", "time": "2026-07-22T08:19:57.000Z"}, {"author": "Brian Carpenter", "text": "<p>@Luigi, any suggestions?</p>", "time": "2026-07-22T08:20:03.000Z"}, {"author": "Bing Liu", "text": "<p>@Sam so it sounds like another \"BGP\"? personally I like the idea as a network guy:)</p>", "time": "2026-07-22T08:20:30.000Z"}, {"author": "Brian Carpenter", "text": "<p>(for a different name than gateway)</p>", "time": "2026-07-22T08:20:43.000Z"}, {"author": "Nic Williams", "text": "<p>@Tommy AFAIK \"entity\" to DAWN includes gateway so suppose the overlay is high</p>", "time": "2026-07-22T08:20:44.000Z"}, {"author": "Antoine Fressancourt", "text": "<p>@Jari I miss the authentication / authorization part in those functions, unless in your view it is linked with policy</p>", "time": "2026-07-22T08:21:00.000Z"}, {"author": "Harald Alvestrand", "text": "<p>Remember that routers used to be called gateways, too :-)</p>", "time": "2026-07-22T08:21:22.000Z"}, {"author": "Jari Arkko", "text": "<p>@antoine yes that is of course needed as well</p>", "time": "2026-07-22T08:21:42.000Z"}, {"author": "Luigi Iannone", "text": "<p>@Brian: \"Mediator\" can be an option, but I am not good at names, I think that once we have a better idea of what this \"thing\" should be we will have also a better idea of a name. My point is not use the name \"gateway\" to define what it is or should do as this would be a bias/limitation</p>", "time": "2026-07-22T08:22:45.000Z"}, {"author": "Lionel Morand", "text": "<p>@Tommy: yes there is some overlap for sure. The reason of the session is to try to identify specific parts that might not be already covered. And see if this should be added to the existing works or a new one should be launched</p>", "time": "2026-07-22T08:23:21.000Z"}, {"author": "Mike Blanche", "text": "<p>seem to be three different elements here:<br>\n1) requests into gateways<br>\n2) processing on gateways<br>\n3) communication to other entities (other agents, other gateways, non-AI systems and services)<br>\n?</p>", "time": "2026-07-22T08:23:38.000Z"}, {"author": "Daniel Huang", "text": "<p>How will the gateway protocol align with a2a protocol, or will it be a standalone protocol ?</p>", "time": "2026-07-22T08:23:45.000Z"}, {"author": "Artur Hecker", "text": "<p>@Jari, Antoine: probably more blocks are missing, filtering / security functionality is another different functional domain, etc.</p>", "time": "2026-07-22T08:24:29.000Z"}, {"author": "Artur Hecker", "text": "<p>Is Charles also the responsible AD for DAWN?</p>", "time": "2026-07-22T08:25:08.000Z"}, {"author": "Nic Williams", "text": "<p>No, Eric / INT for DAWN</p>", "time": "2026-07-22T08:25:17.000Z"}, {"author": "Artur Hecker", "text": "<p>ah yes, thanks, Nic.</p>", "time": "2026-07-22T08:26:21.000Z"}, {"author": "Antoine Fressancourt", "text": "<p>@Mike I would say that your points 1 and 3 are subject to standards, point 2 is more an implementation issue imho</p>", "time": "2026-07-22T08:30:45.000Z"}, {"author": "Arnaud Taddei", "text": "<p>I am really really suprised by this session</p>", "time": "2026-07-22T08:32:51.000Z"}, {"author": "Mike Blanche", "text": "<p>@Antoine. i'd agree. but a number of the presentations seemed to refer to various gateway functions in category 2 (e.g. logging, audit) as candidates for standardisation.</p>\n<p>as for categories 1 and 3, the next question would be what is already being standardised (both in IETF and elsewhere)</p>", "time": "2026-07-22T08:34:13.000Z"}, {"author": "Brian Carpenter", "text": "<p>The gateway \"protocol\"???</p>", "time": "2026-07-22T08:34:18.000Z"}, {"author": "Brian Carpenter", "text": "<p>Cullen is 1000% on target</p>", "time": "2026-07-22T08:35:39.000Z"}, {"author": "Chris Hood", "text": "<p>I will continue to argue, the larger question for all of us, across all of the functionality being discussed, primitives, and agents, is do we continue to build everything on http? </p>\n<p>The underlying substrate, transfer layer, can provide  what is needed as a standard to these applications.</p>", "time": "2026-07-22T08:36:27.000Z"}, {"author": "Harald Alvestrand", "text": "<p>The design error of putting location into resource identifiers in HTTP was an example of a design that made gateways much harder than they should be.</p>", "time": "2026-07-22T08:37:06.000Z"}, {"author": "Brian Carpenter", "text": "<p>@Chris, I feel the same but I fear that ship has already sailed</p>", "time": "2026-07-22T08:37:11.000Z"}, {"author": "Chris Hood", "text": "<p>@Brian -- it's available. We just need more support behind it.</p>", "time": "2026-07-22T08:37:45.000Z"}, {"author": "Harald Alvestrand", "text": "<p>So Cullen's point is valid, but the protocols can make it harder or easier to deploy the gateway-like function.</p>", "time": "2026-07-22T08:38:04.000Z"}, {"author": "Harald Alvestrand", "text": "<p>(SIP and the B2BUA....)</p>", "time": "2026-07-22T08:38:31.000Z"}, {"author": "Antoine Fressancourt", "text": "<p>@Chris I agree, and I would go further and question why agent communication protocols assumes they are human-like</p>", "time": "2026-07-22T08:38:33.000Z"}, {"author": "Brian Carpenter", "text": "<p>@Harald, if the protocols deliver different semantics for the app layer, translation changes from hard to impossible</p>", "time": "2026-07-22T08:39:19.000Z"}, {"author": "Antoine Fressancourt", "text": "<p>@Harald, Intersting B2BUA reference, most of the times I had to use those because protocols were not adapted to cope with a proxy behaviour</p>", "time": "2026-07-22T08:39:50.000Z"}, {"author": "Antoine Fressancourt", "text": "<p>@Brian translation can be transparent if exchanges use similar patterns at both ends of the gateway</p>", "time": "2026-07-22T08:40:49.000Z"}, {"author": "Sam Betts", "text": "<p>One of the main reasons we built SLIM on top of gRPC is that we know HTTP is a well understood and supported protocol through firewalls/gateways today. However we just use this as the underlying substrate for bidirectional streams between SLIM nodes.</p>\n<p>The agent protocols run on top so you only need to place a SLIM node at the boundary of a network like a router, to allow for agents located across network boundaries to communiate without any deployment location awareness. This separates the operations layer from the application layer</p>", "time": "2026-07-22T08:40:58.000Z"}, {"author": "Brian Carpenter", "text": "<p>Yes, IF. That's exactly the problem.</p>", "time": "2026-07-22T08:41:25.000Z"}, {"author": "Laurent Ciavaglia", "text": "<p>We will be polling the audience once the queue is processed.<br>\nWe may give additional time for final questions</p>", "time": "2026-07-22T08:41:34.000Z"}, {"author": "Antoine Fressancourt", "text": "<p>@Sam I think it is a pretty good argument for having standard gateways :-)</p>", "time": "2026-07-22T08:41:47.000Z"}, {"author": "Nicholas Jones", "text": "<p>@chis, certainly agree that needs to be discussed especially in the agent -&gt; gateway context, where intent and capability requirements need to be communicated between the peers as early as possible to allow security controls and functional controls like load-balancing which both should be evaluated before whatever layer 7</p>", "time": "2026-07-22T08:42:51.000Z"}, {"author": "Antoine Fressancourt", "text": "<p>(or at least standard mechanisms to ease proxy interaction in protocols)</p>", "time": "2026-07-22T08:42:57.000Z"}, {"author": "Sam Betts", "text": "<p>@antoine To me gateway implies that its a inbound thing I need to find and talk to. I'm not sure we're considering these gateways communicating with each other right?</p>", "time": "2026-07-22T08:43:44.000Z"}, {"author": "Nic Williams", "text": "<p>I could see a world where gateways may refer to one another (potentially further inbound or to partner services) but I don't think that is GW to GW, more like forwarding</p>", "time": "2026-07-22T08:44:38.000Z"}, {"author": "Arnaud Taddei", "text": "<p>\"on n'est pas sorti de l'auberge\" with this session again. There are soooo many things missing in this discussion I don't know where to start</p>", "time": "2026-07-22T08:44:41.000Z"}, {"author": "Chris Hood", "text": "<p>HTTP is well understood, but not the best option for agents. We are building a frankenstien architecture for the sake of being fast.</p>\n<p>It's a problem. </p>\n<p>This is why I've spent so much time authoring AGTP as a means to build out a dedicated transfer protocol specifically for agents, specifically to solve these complex problems.  </p>\n<p>What applications, gateway, mcp, or others sit on top, benefit from a dedicated layer that's not http.</p>", "time": "2026-07-22T08:44:46.000Z"}, {"author": "Mike Blanche", "text": "<p>gw-to-gw was definitely on some of the slides</p>", "time": "2026-07-22T08:44:54.000Z"}, {"author": "Nicholas Jones", "text": "<p>@brian my feeling is the app layer should be invisble to the GW, so irrelevant, for privacy reasons, intent / credentials etc translated to a protocol understandable and enforceable by the gw, but actions private</p>", "time": "2026-07-22T08:44:54.000Z"}, {"author": "Antoine Fressancourt", "text": "<p>@Sam Inbound can be for a user or for another gateway. Gateway to gateway is typically used to build federations</p>", "time": "2026-07-22T08:45:52.000Z"}, {"author": "Arnaud Taddei", "text": "<p>What is for sure is that given how this started and frankly the lack of sound methodology, the end result IS going to be a Frankenstein architecture.</p>", "time": "2026-07-22T08:46:50.000Z"}, {"author": "Brian Carpenter", "text": "<p>@Nicholas, if the gateway has to _translate_ it needs to understand what it's translating. It needs to be trustworthy.</p>", "time": "2026-07-22T08:46:55.000Z"}, {"author": "Arnaud Taddei", "text": "<p>It ALREADY IS a Frankenstein architecture</p>", "time": "2026-07-22T08:47:01.000Z"}, {"author": "George Fletcher", "text": "<p>@Brian 100%</p>", "time": "2026-07-22T08:47:25.000Z"}, {"author": "Mike Blanche", "text": "<p>...and any other gw that gw talks with needs to understand too...</p>", "time": "2026-07-22T08:47:55.000Z"}, {"author": "Antoine Fressancourt", "text": "<p>@Brian sure</p>", "time": "2026-07-22T08:48:00.000Z"}, {"author": "Arnaud Taddei", "text": "<p>Yes @Brian and that was my point in my intervention ... not a ZINGLE point about how the control plane will do that ... I had a FULL workshop dedicated to that in AI for Good on 9th of July</p>", "time": "2026-07-22T08:48:02.000Z"}, {"author": "Arnaud Taddei", "text": "<p>Including with the leaders of some of the key companies involved here and it had been a specacular success</p>", "time": "2026-07-22T08:48:26.000Z"}, {"author": "Elaine Newton", "text": "<p>Seems like we can skip q2 given the answer to q1</p>", "time": "2026-07-22T08:50:04.000Z"}, {"author": "Chris Lemmons", "text": "<p>I would note that \"I will not contribute\" does not mean \"I do not think others should contribute\".</p>", "time": "2026-07-22T08:51:03.000Z"}, {"author": "Arnaud Taddei", "text": "<p>I would disagree there is a majortiy to work on this</p>", "time": "2026-07-22T08:51:18.000Z"}, {"author": "Artur Hecker", "text": "<p>Elaine, no why? :-) Maybe people in Q2 would like to refine the problem statement for Q1 to be better</p>", "time": "2026-07-22T08:51:22.000Z"}, {"author": "Chris Lemmons", "text": "<p>There exist problems that do not require my expertise. :)</p>", "time": "2026-07-22T08:51:36.000Z"}, {"author": "Nicholas Jones", "text": "<p>@brian maybe it could be similar to HTTP over TLS: various aspects of the inner protocol are represented in  the outer, eg: SNI, ALPN</p>", "time": "2026-07-22T08:51:58.000Z"}, {"author": "Artur Hecker", "text": "<p>Nicholas +1: yes, it's anyhow <em>always</em> the case, whether you like it or not, design it or not. You will have some arteefacts of the inner protocol on the \"wire\", e.g., the end points. Also, the interface to the \"inner\" domain might be way more verbatim than the interface to the outer domain (again, not unlike SOCKS), etc.</p>", "time": "2026-07-22T08:53:53.000Z"}]