BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Sabre//Sabre VObject 4.5.8//EN
CALSCALE:GREGORIAN
LAST-MODIFIED:20260219T071554Z
BEGIN:VTIMEZONE
TZID:Asia/Tokyo
X-MICROSOFT-CDO-TZID:20
BEGIN:STANDARD
DTSTART:20241116T233000
TZOFFSETFROM:+0900
TZOFFSETTO:+0900
TZNAME:JST
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
UID:f4cc8f18-a362-4f41-adaf-5088e5f38364
DTSTAMP:20260219T071554Z
SUMMARY:Semantics for the Agentic Web
DTSTART;TZID=Asia/Tokyo:20251112T083000
DTEND;TZID=Asia/Tokyo:20251112T093000
DESCRIPTION:https://www.w3.org/events/meetings/f4cc8f18-a362-4f41-adaf-5088
 e5f38364/\n\nThe concept of an "agentic web\," where AI agents act on a us
 er's behalf\, is a growing topic of discussion. Currently\, these agents o
 ften rely on brittle inference\, parsing visual labels or class names to u
 nderstand a website's functionality. This model is fragile\; simple A/B te
 sts or site redesigns can break an agent\, leading to high task failure ra
 tes. Keeping in mind that for many tasks the acceptable failure rate is 0.
  This model can also be compute-expensive\, as many implementations resolv
 e ambiguities using a combination of DOM Tree and pixel scraping\, greatly
  limiting the performance of agents who might have resource constrained (i
 .e. on-device) models.\n\nFor human-facing assistive technologies\, ARIA c
 losed gaps in semantic HTML\, creating a more robust experience for users 
 of AT. Now\, we face a similar question for machines.\n\nThis session is a
  general discussion to explore whether our existing semantic toolkit is su
 fficient for this potential use case.\n\n- Is ARIA\, which is designed for
  human accessibility\, the right tool to serve machine agents?  An interes
 ting point of reference is [this FAQ from ChatGPT](https://help.openai.com
 /en/articles/12627856-publishers-and-developers-faq#h_30e9aae450) which st
 ates "ChatGPT Atlas uses ARIA tags—the same labels and roles that suppor
 t screen readers—to interpret page structure and interactive elements." 
 \n- What happens when we conflate these two distinct purposes?\n- More imp
 ortantly\, are there semantic needs for agents that have no human-facing e
 quivalent?\n\nConsider hints that could improve agent reliability and safe
 ty\, such as:\n\n- Explicitly flagging a button as a destructive action (e
 .g.\, distinguishing "Archive" from "Delete Permanently").\n- Identifying 
 page content as User-Generated Content (UGC) to allow agents to apply extr
 a safeguards.\n- Signaling transient states\, like logged-in vs. logged-ou
 t status\, to help an agent plan a task.\n\nThis session is not about a sp
 ecific proposal. It's about the core problem. Does ARIA already solve for 
 all the hints a machine might need\, or should we be considering an additi
 onal semantic layer for machines at all?\n\nLet's come together to discuss
  the problem\, the risks of both action and inaction\, and the potential p
 aths forwards.\n\n**Goal(s):**\nTo collectively explore whether semantic H
 TML and ARIA is sufficient for AI agents or if a new\, machine-facing sema
 ntic layer is needed for a reliable agentic web.\n\nAgenda\n\n**Materials:
 **\n- [deck](https://docs.google.com/presentation/d/1GnFVx8J3j3ZYGGR-zd5P_
 3yP1DBvn5AyH4i4SgxMAoE/edit?slide=id.g3207132d5c2_0_70#slide=id.g3207132d5
 c2_0_70)\n- [Session proposal on GitHub](https://github.com/w3c/tpac2025-b
 reakouts/issues/48)
STATUS:CONFIRMED
CREATED:20251028T131245Z
LAST-MODIFIED:20260219T071554Z
SEQUENCE:4
ORGANIZER;CN=W3C Calendar;PARTSTAT=ACCEPTED;ROLE=NON-PARTICIPANT:mailto:nor
 eply@w3.org
LOCATION:Floor 4 - 401
CATEGORIES:TPAC 2025,Breakout Sessions
END:VEVENT
END:VCALENDAR