14:57:58 RRSAgent has joined #webmachinelearning 14:58:02 logging to https://www.w3.org/2026/01/22-webmachinelearning-irc 14:58:02 RRSAgent, make logs Public 14:58:03 please title this meeting ("meeting: ..."), anssik 14:58:05 Meeting: WebML CG Teleconference – 22 January 2026 14:58:12 Chair: Anssi 14:58:16 Agenda: https://github.com/webmachinelearning/meetings/blob/main/telcons/2026-01-22-cg-agenda.md 14:58:20 Scribe: Anssi 14:58:24 scribeNick: anssik 14:58:32 domfarolino has joined #webmachinelearning 14:58:32 Present+ Anssi_Kostiainen 14:58:47 RRSAgent, draft minutes 14:58:49 I have made the request to generate https://www.w3.org/2026/01/22-webmachinelearning-minutes.html anssik 15:02:07 Mark_Foltz has joined #webmachinelearning 15:02:34 bvandersloot has joined #webmachinelearning 15:03:36 present+ Vasilii_Trofimchuk 15:03:36 Mark_Foltz has joined #webmachinelearning 15:03:46 BenGreenstein has joined #webmachinelearning 15:03:49 brwalder has joined #webmachinelearning 15:04:54 Ehsan has joined #webmachinelearning 15:04:56 Victor has joined #webmachinelearning 15:05:12 present+ 15:05:20 Present+ Ben_Greenstein 15:05:34 Present+ Ben_VanderSloot 15:05:44 present+ 15:05:45 Present+ Tarek_Ziade 15:05:53 EmLauber has joined #webmachinelearning 15:06:00 present+ Mike_Wasserman 15:06:02 Present+ Brandon_Walderman 15:06:07 Present+ Ehsan_Toreini 15:06:12 Present+ Emily_Lauber 15:06:19 Present+ Iris_Johnson 15:06:28 Present+ Kasper_Kulikowski 15:06:33 Present+ Mark_Foltz 15:06:56 Present+ Markus_Tavenrath 15:07:05 Present+ Sathish 15:07:21 Present+ Vasilii_Trofimchuk 15:07:27 Present+ Victor_Huang 15:07:30 who's here 15:08:18 Anssi: welcome back to our telcons 15:08:23 Anssi: first, I'd like to welcome our latest new participants: 15:08:29 Kasper_Kulikowski has joined #webmachinelearning 15:08:30 ... Léonie Watson from TetraLogical Services Ltd. 15:08:30 tarek has joined #webmachinelearning 15:08:48 ... Dominic Farolino and Francois Beaufort from Google 15:09:01 ... individual contributors Hao Wang, Teja Sunku and Paola Di Maio 15:09:06 ... welcome all to the WebML Community Group! 15:09:57 Topic: WebMCP API 15:10:03 gb, this is webmachinelearning/webmcp 15:10:05 anssik, OK. 15:10:27 Subtopic: WebMCP API explainer-to-spec transition 15:10:36 Anssi: issue #56 and PR #64 15:10:36 https://github.com/webmachinelearning/webmcp/issues/56 -> Issue 56 Is this a proposal or a group of proposals? (by 43081j) 15:10:36 https://github.com/webmachinelearning/webmcp/pull/64 -> Pull Request 64 Add initial spec draft and set up CI/CD (by anssiko) 15:10:43 ... today, we will execute the resolution to transition from the WebMCP explainer to a Community Group spec draft stage 15:10:47 -> Resolution https://www.w3.org/2025/12/11-webmachinelearning-minutes.html#6763 15:11:14 Anssi: we will also discuss practicalities and expectations for the Community Group spec draft stage 15:11:51 ... in PR #64 I have established the spec draft baseline and set up a CI/CD system to automate build, validation, checks, and publishing steps 15:11:59 ... this automation will make the lives of editors and contributors easier and ensure all spec builds are of consistent quality 15:12:17 ... spec contributors only need to edit the source file, index.bs, and upon successful merge, the built spec is deployed automatically 15:12:36 iris has joined #webmachinelearning 15:14:23 Dominic: I'd be willing to edit the spec, working with Khus 15:14:54 Brandon: I will be interested in editing 15:15:06 Victor: I will be contributing 15:15:22 ... I will appoint additional editors as appropriate 15:15:38 ... contributions to the WebMCP spec happen via PRs and issues are welcome from all group participants 15:15:54 ... editors and contributors are expected to familiarize themselves with the spec-authoring tool Bikeshed we will use for this spec, it is a powerful and versatile tool with some quirks 15:15:59 ... I'll share some materials below that may be helpful on this journey: 15:16:02 -> Bikeshed Documentation https://speced.github.io/bikeshed/ 15:16:06 -> Introduction to Writing Specifications with Bikeshed https://dlaliberte.github.io/bikeshed-intro/ 15:16:10 -> Writing Procedural Specs for the Web Platform using Bikeshed https://garykac.github.io/procspec/ 15:16:30 Sathish has joined #webmachinelearning 15:16:35 Anssi: folks who have been involved with W3C spec editing in other groups are likely familiar with Bikeshed and associated tooling 15:16:55 ... the WebMCP spec will be called a "Draft Community Group Report" in W3C's parlance 15:16:59 -> Types of documents W3C publishes https://www.w3.org/standards/types/#CG-DRAFT 15:17:28 Anssi: a partial patent policy applies to this type of specs to allow for an easy way innovate that is inclusive of all, including all individual contributors that join the group 15:17:51 ... formal wide reviews are not expected for Draft Community Group Reports 15:18:08 ... however, W3C TAG conducts and encourages early TAG design reviews, called "Early design/incubation review" to ensure cohesion within the whole at this earlier stage pre-standards track 15:18:12 ... my expectation is we will seek early review from the TAG when our early spec draft takes shape 15:19:13 -> https://webmachinelearning.github.io/webmcp/ 15:19:44 Subtopic: Agentic AI Foundation coordination 15:19:49 gb, this is webmachinelearning/charter 15:19:49 anssik, OK. 15:19:55 Anssi: issue #14 15:19:56 https://github.com/webmachinelearning/charter/issues/14 -> Issue 14 Agentic AI Foundation coordination (by anssiko) 15:20:32 ... as discussed earlier, we're actively looking to formalizing our coordination with newly Agentic AI Foundation for WebMCP to ensure cohesive development 15:20:47 ... I've worked with Vasilii from Block Inc., AAIF co-founder, on a coordination proposal 15:21:00 ... Vasilii shared the proposal in a GH comment in this issue, I'll summarize the key points 15:21:24 ... the goal is it keep this lightweight, flexible, let the two specs grow and develop 15:21:29 ... three key points of the proposal: 15:21:43 ... - scope - WebMCP to enumerate what MCP spec sections apply 15:22:15 ... - intent - document the intent is to keep WebMCP aligned with MCP where applicable 15:22:34 ... - engagement - agree on communication channels (GitHub, Discord, MCP Working Group meetings ...) 15:22:50 ... once we're settled, we will codify these expectations in the WebML CG's Charter that will be submitted for review 15:23:48 Vasilii: thanks for the overview, I'm thinking the engagement model can change with time, now starting with lightweight model 15:24:21 ... MCP moving fast, good to keep track of changes 15:25:07 q? 15:25:51 Dominic: do we expect AAIF folks to join? 15:26:32 Vasilii: I will invite AAIF folks to join the W3C WebML CG 15:26:32 q? 15:26:55 Subtopic: Declarative proposal to expose tools via HTML 15:26:59 gb, this is webmachinelearning/webmcp 15:26:59 anssik, OK. 15:27:06 Anssi: issue #22 and PR #26 15:27:06 https://github.com/webmachinelearning/webmcp/issues/22 -> Issue 22 Declarative API Equivalent (by EisenbergEffect) [Agenda+] 15:27:06 https://github.com/webmachinelearning/webmcp/pull/26 -> Pull Request 26 Declarative proposal to expose tools via HTML (by MiguelsPizza) [Agenda+] 15:27:13 -> Chromium Declarative WebMCP prototype https://github.com/chromium/chromium/commit/3df80580897d24fbb730c0ee411f25ffe04e766b 15:27:43 Anssi: the declarative proposal to expose tools via HTML has been identified as an important complementary feature to the programmatic WebMCP API by the group 15:28:02 ... we've received new feedback from various folks since our last discussion, to be discussed today 15:28:06 ... the feedback is split into the issue discussion and in the PR 15:28:12 ... we'll look at the recent comments in issue #22 first 15:28:34 -> https://github.com/webmachinelearning/webmcp/issues/22#issuecomment-3726418984 15:28:35 https://github.com/webmachinelearning/webmcp/issues/22 -> Issue 22 Declarative API Equivalent (by EisenbergEffect) [Agenda+] 15:28:41 ... Khushal clarified the two goals for the declarative proposal: 15:28:44 ... 1) Support for cross-document navigations, this can’t be done with the current imperative API 15:28:48 ... 2) Developer ergonomics, make it trivial to expose existing actions without necessitating client-side scripting 15:29:20 Domenic: semantic HTML elements to register themselves as tools, slap attributes on form elements 15:29:54 ... the first take of this is in PR #26 15:30:15 ... Ben of Mozill has comments on PR that the agent might actuate the form itself 15:30:30 ... PR #26 is the baseline for the Chromium prototype 15:30:52 ... our goal is to resolve the group wants to work on the declarative proposal 15:31:21 q? 15:32:10 Dominic: Khushal's comment notes form submission has two cases: cross-document and same document interaction 15:32:51 ... cross-document schedules page navigation, this is the more complex case 15:33:10 q+ 15:33:32 ... we will prototype to find answers to the open design issues 15:34:39 ack bvandersloot 15:35:02 q+ 15:35:55 BenV: declarative approach may better address some issues from Mozilla's side re WebMCP 15:36:00 ack brwalder 15:36:23 Brandon: on our side, form submission is an important consideration, support declarative approach in addition to imperative API 15:36:44 ... I support mapping the forms, but the way of retrieving information still needs some work 15:36:52 jason_mcghee has joined #webmachinelearning 15:37:04 ... the proposal is to ask link elements to do things the link elements don't normally do e.g. respond with JSON objects 15:37:18 ... it encourages different patterns than I think we want 15:37:40 ... the premise is to augment existing pages for humans to be useful for agent 15:38:03 ... humans look at pages, but for agents, we need to add a link to the page that is not useful for humans 15:38:12 ... I support form submission approach 15:38:30 Dominic: from Chrome side we are not considering links yet in our design 15:38:31 q? 15:38:40 Present+ Jason_McGhee 15:39:10 (hello - apologies for tardiness / conflicting meetings) 15:39:36 RESOLUTION: Pursue declarative approach that complements imperative WebMCP API 15:40:19 Anssi: looking forward to further implementation experience 15:40:29 Subtopic: Define the API for in-page Agents to use a site's declared tools 15:40:33 Anssi: issue #51 15:40:34 https://github.com/webmachinelearning/webmcp/issues/51 -> Issue 51 Define the API for in-page Agents to use a site's declared tools (by khushalsagar) [Agenda+] 15:40:52 ... at TPAC we had a good discussion but we did not reach consensus so I wanted to revisit this topic with new information 15:41:06 ... since TPAC Khushal contributed an API sketch to make the ModelContextClient API proposal more concrete: 15:41:11 -> https://github.com/webmachinelearning/webmcp/issues/51#issuecomment-3598567325 15:41:11 https://github.com/webmachinelearning/webmcp/issues/51 -> Issue 51 Define the API for in-page Agents to use a site's declared tools (by khushalsagar) [Agenda+] 15:41:25 Anssi: I'd like us to review and discuss the proposed ModelContextClient interface 15:41:38 ... this extensible API allows listing current tools and executing them 15:41:47 ... key motivation is to avoid concurrency issues with multiple agents 15:43:17 Dominic: overall this is a good idea, I think it is the right direction, currently trying to avoid scope expansion so lower priority right now 15:43:19 q? 15:43:20 q+ 15:43:26 ack Victor 15:44:33 Victor: my thoughts, not sure if there's strong demand from 3P party agents for this feature 15:44:58 ... there must be interest from 3P sites to use this feature, need confirmed demand 15:45:13 Anssi: James shared a Salesforce UI use case and expectations wrt API in a comment 15:45:19 -> https://github.com/webmachinelearning/webmcp/issues/51#issuecomment-3599293112 15:46:15 Dominic: I can take an action to respond on this issue 15:46:44 +1 on some form of labelling to triage 15:46:50 ... we could label issues e.g. v1 or v2 15:46:54 q? 15:47:22 Subtopic: Add a callback to inform the site when tool invocation should be aborted 15:47:27 Anssi: issue #48 15:47:28 https://github.com/webmachinelearning/webmcp/issues/48 -> Issue 48 Add a callback to inform the site when tool invocation should be aborted (by khushalsagar) [Agenda+] 15:47:35 ... this issue is about how to inform the site if the user cancels the tool invocation that is in flight 15:47:42 ... it seems we're converging to use AbortSignal for the job 15:48:16 Brandon: precedent for AbortSignal is Fetch API 15:48:34 Dominic: should resolve this to use AbortSignal 15:48:52 +1 15:49:06 RESOLUTION: Use AbortSignal to signal cancellation (issue #48) 15:49:17 Subtopic: Should concurrent tool execution be allowed? 15:49:21 Anssi: issue #47 15:49:22 https://github.com/webmachinelearning/webmcp/issues/47 -> Issue 47 Should concurrent tool execution be allowed? (by khushalsagar) [Agenda+] 15:49:32 ... Khushal notes tool execution is async, multiple tools can be invoked in parallel 15:49:46 ... James shared concurrent execution is important, and managing concurrency should be always left to userland code 15:50:02 ... promise-returning API ensures the tool author can choose to await for a tool call to finish before invoking a new tool call 15:50:04 q? 15:50:16 q+ 15:50:19 ack brwalder 15:50:48 Brandon: I think this is dependent on how we handle multiple clients, ref issue #51 15:50:49 https://github.com/webmachinelearning/webmcp/issues/51 -> Issue 51 Define the API for in-page Agents to use a site's declared tools (by khushalsagar) [Agenda+] 15:51:17 Brandon: first need to resolve issue #51 prior to resolving #47 15:51:46 Anssi: should we punt this for later? 15:51:57 [agreement heard] 15:52:16 Topic: Built-in AI APIs 15:52:25 Anssi: we have four built-in AI APIs in scope of the group: 15:52:32 -> https://github.com/webmachinelearning/prompt-api 15:52:36 -> https://webmachinelearning.github.io/writing-assistance-apis/ 15:52:42 -> https://webmachinelearning.github.io/translation-api/ 15:52:47 -> https://webmachinelearning.github.io/proofreader-api/ 15:53:03 Anssi: first, I want to welcome new editors on board and then discuss 2026 priorities 15:53:14 ... Translator and Language Detector APIs was the last API in the group's scope without active editors 15:53:25 ... I'm pleased to appoint Reilly Grant from Google and Ehsan Toreini from Samsung as the editors of this spec 15:53:44 ... Reilly is an experienced web spec author and browser engineer and Ehsan is a research engineer with privacy, cryptography and security focus, a great pair to work on this spec 15:53:49 ... thank both for volunteering to take on this role 15:55:36 Ehsan: thank you Anssi and hello everyone, I work in Samsung based in London UK, worked on trustworthiness of ML, privacy and security, HW security 15:55:42 ... nice to be part of the group! 15:56:38 Anssi: to help the group participants focus their attention across these API specs, I'd ask the editors to label high priority issues with "Agenda+" labels 15:56:48 ... we will give Agenda+ issues higher priority and more agenda time 15:57:09 ... also, if implementers have certain APIs and/or features higher priority for this year, it'd be good to know to ensure the group gives such specs appropriate attention 15:57:32 ... questions, comments? 15:57:43 q+ 15:58:06 ack msw 15:58:50 Mike: I can offer insight on Chrome plans for 2026, Prompt API is one of our biggest focus areas, for issue priorities, on Prompt API repo a bunch of issues was filed representing feedback received from the TAG 15:59:07 ... these issues benefit from input from this group 15:59:32 q? 15:59:37 Topic: Meeting scheduling fine-tuning 15:59:44 Anssi: I want to solicit feedback for the proposal to shift this CG call forward by one hour 15:59:59 ... this timeslot we have chosen is a compromize we've used to allow US West Coast, Europe and China and sometimes even Japan join 16:00:02 ... do we currently have active participants from China, Japan, or from APAC? 16:01:15 to 6am PST? 16:01:16 No conflicts 16:01:21 no conflicts 16:01:23 Mike: Kenji might join on occasion from Google Tokyo 16:01:56 jason_mcghee, I believe 8am PST 16:02:12 no conflicts 16:02:15 vasilii - thanks anssi immediately answered :) 16:02:20 +1 16:02:25 RESOLUTION: Shift the WebML CG call forward by one hour and re-evaluate schedule from time to time considering active participants' timezones" 16:02:53 thanks anssi 16:03:48 RRSAgent, draft minutes 16:03:49 I have made the request to generate https://www.w3.org/2026/01/22-webmachinelearning-minutes.html anssik 16:21:48 s/… I will/Anssi: I will 16:21:50 RRSAgent, draft minutes 16:21:52 I have made the request to generate https://www.w3.org/2026/01/22-webmachinelearning-minutes.html anssik 16:23:02 s/I will appoint/Anssi: I will appoint 16:23:04 RRSAgent, draft minutes 16:23:05 I have made the request to generate https://www.w3.org/2026/01/22-webmachinelearning-minutes.html anssik 16:24:53 s/with newly/with the newly founded 16:28:00 s/programmatic/imperative 16:28:47 s/Domenic/Dominic 16:29:09 s/of Mozill/of Mozilla 16:35:50 RRSAgent, draft minutes 16:35:51 I have made the request to generate https://www.w3.org/2026/01/22-webmachinelearning-minutes.html anssik 16:37:46 s/3P party/3rd party 16:41:50 s/thank both/thank you both 16:42:09 s/work in/work at 16:43:51 s/timezones"/timezones 16:43:54 RRSAgent, draft minutes 16:43:55 I have made the request to generate https://www.w3.org/2026/01/22-webmachinelearning-minutes.html anssik 18:30:58 Zakim has left #webmachinelearning