From browser actions to shared protocols and commerce: a separate, source-linked chronology.
32 selected events · October 2024–October 2026 Announcements, previews, rollout starts and documented availability remain distinct.
Show only
Showing 32 of 32 events.
Color categories are editorial filters. Availability badges describe the event’s historical status, not a live account or deployment test.
AnthropicExecution
Anthropic releases screenshot-based computer use in public API beta
Developer API beta
Claude 3.5 Sonnet's computer-use beta lets developers supply screenshots and execute model-generated cursor, click and keyboard actions. It launched on the Anthropic API, Amazon Bedrock and Google Cloud Vertex AI.
Claude 3.5 Sonnet's computer-use beta lets developers supply screenshots and execute model-generated cursor, click and keyboard actions. It launched on the Anthropic API, Amazon Bedrock and Google Cloud Vertex AI.
Availability at the event
Public developer beta at launch; experimental, cumbersome and error-prone. This was a model/tool capability requiring a software execution environment.
Experimental capability, not universal GUI reliability. OSWorld 14.9% screenshot-only and 22.0% extended-step results are vendor-reported benchmark evidence, not reproduced here.
Editorial interpretation
Computer use becomes a developer capability
Editorial interpretation: Establishes screenshot-to-action software as a concrete building block, with an experimental API beta and benchmark limits.
“Execution” is an analytical filter, not a source claim or an official historical period.
Date and documentary review
Dated announcement explicitly says available today
Date precision
day
Source publication date
2024-10-22
Evidence snapshot
2026-10-11
Independent documentary review
Verified with qualification
Primary announcement and primary research explanation directly opened; vendor-reported evaluation, no reproduced task test
Opening paragraphs; Teaching Claude to navigate computers, responsibly
Anthropic; MCP creators David Soria Parra and Justin Spahr-SummersInteroperability
Anthropic open-sources MCP and local Claude Desktop integration
Released local integration
Anthropic released the MCP specification, SDKs, sample servers and local MCP-server support in Claude Desktop, providing a shared client/server interface for tools and data.
Anthropic released the MCP specification, SDKs, sample servers and local MCP-server support in Claude Desktop, providing a shared client/server interface for tools and data.
Availability at the event
Open source and local Desktop support available; remote production deployment toolkits were future work.
Qualification
Launch predominantly enabled local connectors. Block/Apollo integration is publisher-reported, and other companies were described as working with MCP.
Early integrations are vendor-reported; no measured adoption or dependable autonomous capability is established.
Editorial interpretation
A shared interface for tools and data
Editorial interpretation: Explains how agents acquire reusable tool/data access; local support and future remote tooling remain explicit.
“Interoperability” is an analytical filter, not a source claim or an official historical period.
Date and documentary review
Contemporaneous announcement says 'Today' and lists released components.
Date precision
day
Source publication date
2024-11-25
Evidence snapshot
2026-10-11
Independent documentary review
Verified
Primary source text inspected; no runtime test or audited adoption dataset
OpenAI releases Operator as a U.S. Pro research preview
U.S. Pro research preview
Operator used a remote browser and Computer-Using Agent model to navigate sites by screenshots, typing, clicking and scrolling; examples included forms, groceries and bookings.
Operator used a remote browser and Computer-Using Agent model to navigate sites by screenshots, typing, clicking and scrolling; examples included forms, groceries and bookings.
Availability at the event
Research preview for ChatGPT Pro users in the U.S.; integrated into ChatGPT agent on July 17, 2025.
Qualification
Login, payment details and CAPTCHAs required user takeover. Announcement does not prove broad adoption or unattended transaction reliability. OpenAI's same-day Computer-Using Agent research post reported 38.1% on OSWorld (human 72.4%), 58.1% on WebArena (human 78.2%) and 87% on WebVoyager. These are vendor-reported benchmarks, not reproduced here.
User takeover for login/payment/CAPTCHAs and approval before consequential actions. Examples and partnerships do not measure broad adoption or unattended success.
Editorial interpretation
A consumer browser-action preview
Editorial interpretation: Shows the transition from developer primitives to a user-facing web agent, while preserving U.S. Pro restrictions and takeover requirements.
“Execution” is an analytical filter, not a source claim or an official historical period.
Date and documentary review
Dated launch post says starting today
Date precision
day
Source publication date
2025-01-23
Evidence snapshot
2026-10-11
Independent documentary review
Verified with qualification
Primary launch text directly opened; demonstrations and benchmark claims not independently reproduced
January 23 date; opening; How to use; July 17 update
Google Cloud and more than 50 supporting technology/service partnersInteroperability
Google introduces the draft Agent2Agent protocol
Draft specification
Google introduced open-source A2A for capability discovery through Agent Cards, task lifecycle updates and artifact exchange between client and remote agents, building on HTTP, SSE and JSON-RPC.
Google introduced open-source A2A for capability discovery through Agent Cards, task lifecycle updates and artifact exchange between client and remote agents, building on HTTP, SSE and JSON-RPC.
Availability at the event
Draft specification, code samples and contribution paths released; a production-ready version was planned later in 2025.
Qualification
Partner support is not deployment/adoption measurement. The hiring scenario is illustrative; it does not establish completed cross-vendor transactions.
More than 50 supporters and illustrative hiring scenario do not prove deployed interoperability or completed workflows.
Editorial interpretation
Agents get a draft coordination protocol
Editorial interpretation: Introduces agent-to-agent discovery, tasks and artifacts as a separate layer from tool access; pairs with the later stable A2A milestone.
“Interoperability” is an analytical filter, not a source claim or an official historical period.
Date and documentary review
Dated launch announcement; linked specification explicitly called draft.
Date precision
day
Source publication date
2025-04-09
Evidence snapshot
2026-10-11
Independent documentary review
Verified
Primary source text inspected; no runtime test or audited adoption dataset
Date; 'How A2A works'; 'The future of agent interoperability', especially draft and future production version.
Coinbase Developer PlatformCommerce and fulfillment
Coinbase launches x402 for HTTP-native stablecoin payments
Public protocol release
Coinbase introduced x402, an open payment protocol that uses HTTP 402 to quote a resource and a signed stablecoin payment payload to authorize access. A payment facilitator verifies and settles payment before the server returns the resource.
Coinbase introduced x402, an open payment protocol that uses HTTP 402 to quote a resource and a signed stablecoin payment payload to authorize access. A payment facilitator verifies and settles payment before the server returns the resource.
Availability at the event
Public protocol launch and developer integration route; launch source describes stablecoin rails and example applications.
Qualification
Examples and partner logos are not transaction-volume proof. Facilitator, wallet, funding, authorization and implementation security remain dependencies.
Verification establishes the protocol launch and described flow. Collaborator lists/examples are not audited volumes or generic purchase authorization.
Editorial interpretation
Machines can pay for digital resources
Editorial interpretation: Broadens commerce beyond shopping carts to HTTP-native resource access, without implying transaction volume or universal authorization.
“Commerce and fulfillment” is an analytical filter, not a source claim or an official historical period.
Date and documentary review
Dated primary article says Coinbase is launching x402.
Date precision
day
Source publication date
2025-05-06
Evidence snapshot
2026-10-11
Independent documentary review
Verified
Vendor primary protocol-launch article read; flow inspected; no SDK execution or adoption audit.
Date; TL;DR; 'How x402 Works'; web extract lines46–59,84–97
ChatGPT agent combines research, browser action and artifact work
Phased paid-plan rollout
ChatGPT agent combined Operator's website interaction and deep research with a virtual computer, visual/text browsers, terminal and API access; it could research, navigate sites and produce editable artifacts.
ChatGPT agent combined Operator's website interaction and deep research with a virtual computer, visual/text browsers, terminal and API access; it could research, navigate sites and produce editable artifacts.
Availability at the event
Pro by end of launch day; Plus/Team over following days; Enterprise/Education promised later. EEA and Switzerland were excluded at launch.
Qualification
Consequential actions need approval; prompt-injection and model-error risks remain. Vendor SpreadsheetBench score was 45.54% with XLSX versus human 71.33%.
This verifies launch documentation, not reproduced completion reliability. SpreadsheetBench 45.54% with XLSX vs 71.33% human is vendor-reported and uses differing evaluation environments; consequential actions require approval.
Editorial interpretation
Research, browser action and artifacts converge
Editorial interpretation: Provides a clear integrated-work-agent milestone and a dated phased rollout, while keeping approval and reliability limits visible.
“Execution” is an analytical filter, not a source claim or an official historical period.
Date and documentary review
Dated launch / start of phased rollout
Date precision
day
Source publication date
2025-07-17
Evidence snapshot
2026-10-11
Independent documentary review
Verified with qualification
Primary launch text directly opened; vendor evaluations/demos, no reproduced task test
A natural evolution; An agent that works; Availability; Limitations
Brave publishes Comet indirect prompt-injection proof of concept
Historical controlled security report
Brave described a controlled proof of concept in which webpage instructions redirected Comet from summarizing a Reddit post to extracting an email address and one-time code from authenticated sessions.
Brave described a controlled proof of concept in which webpage instructions redirected Comet from summarizing a Reddit post to extracting an email address and one-time code from authenticated sessions.
Availability at the event
Public vulnerability-research report. Disclosure timeline records fixes and retesting; an update says the attack class was not fully mitigated then.
Qualification
Historical report, not a statement that current Comet remains vulnerable. Do not present lab demonstration as exploitation at scale.
Report includes patch/retest chronology and later incomplete-mitigation update. Brave is a competing browser vendor; exploit not reproduced here. Avoid current vulnerability or population impact claims.
Editorial interpretation
A controlled attack demonstrates a trust-boundary risk
Editorial interpretation: Provides concrete historical counterevidence to effortless autonomy; label the competing-vendor report and unreplicated proof of concept.
“Execution” is an analytical filter, not a source claim or an official historical period.
Date and documentary review
Public disclosure/publication date; discovery occurred July 25
Date precision
day
Source publication date
2025-08-20
Evidence snapshot
2026-10-11
Independent documentary review
Verified with qualification
Original researchers' report directly opened; proof of concept not reproduced; no real-world victim evidence
Attack demonstration; Disclosure timeline; Research Motivation
MCP Registry working group; contributors from Anthropic, GitHub, Block and PulseMCPInteroperability
Official MCP Registry opens in preview
Registry preview
The official MCP Registry opened an API/catalog for public-server metadata, with an open-source OpenAPI specification intended to support downstream public and private subregistries.
The official MCP Registry opened an API/catalog for public-server metadata, with an open-source OpenAPI specification intended to support downstream public and private subregistries.
Availability at the event
Public preview; no data-durability guarantee and possible breaking changes before general availability.
Qualification
Entries contain self-reported information. Moderation/denylisting is not a security certification; catalog publication does not prove safe installation or client integration.
Metadata is self-reported. Moderation and denylisting do not establish certification, safe installation or client compatibility.
Editorial interpretation
Tool discovery gets a shared catalog
Editorial interpretation: Makes distribution and discoverability intelligible as dependencies; preview status and self-reported metadata must stay visible.
“Interoperability” is an analytical filter, not a source claim or an official historical period.
Date and documentary review
Dated maintainers' post says registry is launching today in preview.
Date precision
day
Source publication date
2025-09-08
Evidence snapshot
2026-10-11
Independent documentary review
Verified with qualification
Primary source text inspected; no runtime test or audited adoption dataset
Date; opening; 'Public and private sub-registries'; 'Getting started' preview caveat.
Google; payments and technology collaboratorsCommerce and fulfillment
Google announces Agent Payments Protocol (AP2)
Public specification and reference materials
Google announced AP2, a payment-agnostic open protocol that can extend A2A and MCP. Cryptographically signed intent and cart mandates create an evidence chain for human-present purchases and delegated purchases. The primary source offers public specification, documentation and reference implementations.
Google announced AP2, a payment-agnostic open protocol that can extend A2A and MCP. Cryptographically signed intent and cart mandates create an evidence chain for human-present purchases and delegated purchases. The primary source offers public specification, documentation and reference implementations.
Availability at the event
Public protocol/specification and reference implementations; examples are proposed use cases; no general production checkout deployment proven.
Qualification
Support from more than 60 organizations is collaboration evidence, not deployment/adoption count. Cryptographic evidence does not eliminate fraud, mistaken intent or legal responsibility.
Sixty-plus collaborators do not establish sixty production integrations. Historical launch mandate terminology must be distinguished from later specification evolution; FIDO development is not a finished ratified standard.
Editorial interpretation
Payment intent and consent get a protocol
Editorial interpretation: Introduces the authorization-evidence layer, distinct from payment settlement or a deployed shopping service.
“Commerce and fulfillment” is an analytical filter, not a source claim or an official historical period.
Date and documentary review
Primary article display date September 16 and 'Today' announcement; protocol site also dates announcement September 16.
Date precision
day
Source publication date
2025-09-16
Evidence snapshot
2026-10-11
Independent documentary review
Verified
Primary announcement read including mandate flow and public implementation route; no code execution.
Display date; opening announcement; 'How it works: Establishing trust via mandates and verifiable credentials'; 'What’s next'
OpenAI; Stripe; Etsy; Shopify (planned)Commerce and fulfillment
ChatGPT launches U.S. Instant Checkout and open ACP
Limited public checkout
OpenAI launched Instant Checkout for U.S. ChatGPT Free, Plus and Pro users buying from U.S. Etsy sellers and published the Agentic Commerce Protocol with Stripe. Launch checkout supported single-item purchases; Shopify merchants, multi-item carts and additional regions were future plans.
OpenAI launched Instant Checkout for U.S. ChatGPT Free, Plus and Pro users buying from U.S. Etsy sellers and published the Agentic Commerce Protocol with Stripe. Launch checkout supported single-item purchases; Shopify merchants, multi-item carts and additional regions were future plans.
Availability at the event
Limited public availability: U.S. users/U.S. Etsy; user confirmation; single-item checkout. Shopify listed as coming soon.
Qualification
Do not report a million Shopify merchants live on launch day. Native checkout is an integration, not open-ended delegated purchasing. See C014 for later strategy change.
These are historical launch limits. C014 records later change. No checkout was executed during verification.
Editorial interpretation
In-chat checkout launches within a narrow scope
Editorial interpretation: Anchors commercial execution in U.S. Etsy single-item checkout and user confirmation; pairs with the later strategy revision.
“Commerce and fulfillment” is an analytical filter, not a source claim or an official historical period.
Date and documentary review
Dated primary launch says 'Starting today' and 'can now buy'.
OpenAI launches Atlas with paid-plan agent-mode preview on macOS
macOS launch; agent-mode preview
Atlas launched as a browser with ChatGPT, optional browser memories and agent mode using the user's browsing context. Browser access and action-agent access had different plan gates.
Atlas launched as a browser with ChatGPT, optional browser memories and agent mode using the user's browsing context. Browser access and action-agent access had different plan gates.
Availability at the event
Worldwide macOS browser launch for Free/Plus/Pro/Go; Business beta and admin-enabled Enterprise/Edu. Agent mode preview for Plus/Pro/Business. The current launch page says Atlas has since been deprecated.
Qualification
At launch the agent could not run browser code, download files, install extensions or access other computer apps/files; workflow mistakes remained possible.
Current page says deprecated. Agent at launch lacked code execution, downloads, extensions and other-app/filesystem access. Do not imply present availability or proven reliable transactions.
Editorial interpretation
Agent actions move into a browser product
Editorial interpretation: Documents the Atlas browser/agent-mode availability split; pairs with retirement so the chronology does not imply every product persists.
“Execution” is an analytical filter, not a source claim or an official historical period.
Date and documentary review
Dated launch says launching today
Date precision
day
Source publication date
2025-10-21
Evidence snapshot
2026-10-11
Independent documentary review
Verified with qualification
Primary launch text directly opened with current product-status banner; no reproduced task test
Opening availability paragraphs; Get work done for you; current deprecation banner
Linux Foundation; Anthropic, Block, OpenAI; supporting platform companiesGovernance
Agentic AI Foundation launches with MCP, goose and AGENTS.md
Foundation/project contributions announced
AAIF was formed under the Linux Foundation with founding contributions of MCP, Block's goose agent framework and OpenAI's AGENTS.md instruction convention.
AAIF was formed under the Linux Foundation with founding contributions of MCP, Block's goose agent framework and OpenAI's AGENTS.md instruction convention.
Availability at the event
Foundation and project contributions announced. Infrastructure already open source; stewardship change is the event.
Qualification
Claims of 'universal' standards and published-server/project totals are foundation/vendor assertions, not independent market-share measurements. Governance does not guarantee safe execution.
Open governance does not certify execution safety; universal-standard language and counts are institutional/vendor claims.
Editorial interpretation
Open projects gain a shared institutional home
Editorial interpretation: Groups MCP, goose and AGENTS.md without conflating their technical roles; introduces stewardship rather than another capability claim.
“Governance” is an analytical filter, not a source claim or an official historical period.
Date and documentary review
AAIF/Linux Foundation dated press release explicitly announces formation and founding project contributions.
Date precision
day
Source publication date
2025-12-09
Evidence snapshot
2026-10-11
Independent documentary review
Verified
Primary source text inspected; no runtime test or audited adoption dataset
Date; summary; formation paragraph and individual project descriptions.
Google; Shopify; Etsy; Wayfair; Target; WalmartCommerce and fulfillment
Google and retail partners launch Universal Commerce Protocol
Public protocol; checkout prospective
Google announced UCP, co-developed with retail/platform partners to provide a common commerce language from discovery and purchase through post-purchase support. It stated compatibility with A2A, AP2 and MCP; eligible U.S. native Google checkout would arrive soon.
Google announced UCP, co-developed with retail/platform partners to provide a common commerce language from discovery and purchase through post-purchase support. It stated compatibility with A2A, AP2 and MCP; eligible U.S. native Google checkout would arrive soon.
Availability at the event
Public protocol launch; native AI Mode/Gemini checkout prospective at announcement; seller remains merchant of record.
Qualification
Partner endorsements are not production deployments. Business Agent/Direct Offers are distinct features with their own scopes. Do not call every retailer UCP-ready.
Verify protocol launch, not all future full-lifecycle functions deployed. Business Agent Jan 12 and Direct Offers pilot are distinct products, not evidence of UCP merchant adoption.
Editorial interpretation
Retailers and platforms define a commerce lifecycle
Editorial interpretation: Introduces UCP as an open protocol spanning commerce stages, retaining the distinction between the protocol launch and prospective native checkout.
“Commerce and fulfillment” is an analytical filter, not a source claim or an official historical period.
Date and documentary review
Dated primary blog says 'Today' launching UCP.
Date precision
day
Source publication date
2026-01-11
Evidence snapshot
2026-10-11
Independent documentary review
Verified
Google primary launch and Shopify co-developer release read; protocol/checkout not executed.
'Unlock retail growth with a new open standard'; web extract lines43–52; Shopify Jan11 corroboration
MCP community; Anthropic, OpenAI, MCP-UI creators; Block and Microsoft clientsInteroperability
MCP Apps becomes an official interactive-UI extension
Official extension; client rollout differs
MCP Apps standardized tool-linked HTML/JavaScript interfaces rendered in sandboxed iframes within a conversation, with bidirectional host communication and SDK/examples.
MCP Apps standardized tool-linked HTML/JavaScript interfaces rendered in sandboxed iframes within a conversation, with bidirectional host communication and SDK/examples.
Availability at the event
Official extension/SDK released; Claude and Goose available that day, VS Code Insiders support, ChatGPT starting that week.
Qualification
Host support and restrictions vary. Sandboxing and optional consent are mitigations, not proof that arbitrary server-supplied UI is safe.
Host support/restrictions differ; iframe isolation and consent mechanisms are mitigations, not proof arbitrary server-supplied UI is safe.
Editorial interpretation
People can inspect and decide inside agent interfaces
Editorial interpretation: Shows why interactive UI matters for human participation, while retaining the client-by-client availability boundary.
“Interoperability” is an analytical filter, not a source claim or an official historical period.
Date and documentary review
Dated maintainers' release says extension is now live; precise client rollout qualifiers in 'Client Support'.
Date precision
day
Source publication date
2026-01-26
Evidence snapshot
2026-10-11
Independent documentary review
Verified with qualification
Primary source text inspected; no runtime test or audited adoption dataset
Date; opening; 'How It Works'; 'Security Model'; detailed 'Client Support'.
A2A community and Technical Steering CommitteeInteroperability
A2A v1.0 becomes the project's first stable specification
Stable specification and migration guidance
A2A v1.0 formalized multiple protocol bindings, version negotiation, multi-tenancy, signed Agent Cards and revised security flows for cross-agent coordination.
A2A v1.0 formalized multiple protocol bindings, version negotiation, multi-tenancy, signed Agent Cards and revised security flows for cross-agent coordination.
Availability at the event
First stable/production-ready specification and migration materials available, according to project; SDK support across programming languages remained ongoing work.
Qualification
Breaking interaction changes require migration. AgentCard supports advertising v0.3 and v1.0 together; do not assume wire-level compatibility without implementation checks.
Interaction protocol has breaking changes; AgentCards can advertise both v0.3/v1.0. Prefer SDK support across programming languages over ambiguous multilingual SDK wording.
Editorial interpretation
A2A reaches a stable specification
Editorial interpretation: Closes the draft-to-stable arc begun by P002; project-characterized readiness and breaking interaction changes remain qualified.
“Interoperability” is an analytical filter, not a source claim or an official historical period.
Date and documentary review
Dated A2A project release says community is announcing v1.0 today.
Date precision
day
Source publication date
2026-03-12
Evidence snapshot
2026-10-11
Independent documentary review
Verified with qualification
Primary source text inspected; no runtime test or audited adoption dataset
Metadata date; first paragraph; enterprise requirements; migration; closing SDK sentence.
Google adds the Google-Agent user agent for agents acting on user request
Rolling out over following weeks
Google added Google-Agent to its user-triggered fetchers. Agents hosted on Google infrastructure use it to browse and take actions when a user asks, and Google published a separate IP-range file for user-triggered agents.
Google added Google-Agent to its user-triggered fetchers. Agents hosted on Google infrastructure use it to browse and take actions when a user asks, and Google published a separate IP-range file for user-triggered agents.
Availability at the event
Changelog says the user agent was rolling out over the following weeks.
Qualification
Google says user-triggered fetchers generally ignore robots.txt. Documentation describes identity and IP ranges, not how often or where Google-Agent is used.
Documentation change; it does not measure agent traffic volume.
Editorial interpretation
Agents get their own identity in server logs
Editorial interpretation: Gives site owners a named, verifiable identity for agent traffic, distinct from Googlebot crawling.
“Interoperability” is an analytical filter, not a source claim or an official historical period.
Date and documentary review
Dated entry in Google's crawling documentation changelog
Date precision
day
Source publication date
2026-03-20
Evidence snapshot
2026-11-05
Independent documentary review
Added 5 Nov 2026; second review pending
Primary documentation opened; the IP-range file returned HTTP 200 on 5 November 2026; no traffic test
Changelog March 2026 entry; user-triggered fetchers page, Google-Agent section
OpenAI said initial Instant Checkout lacked desired flexibility and shifted focus to discovery with merchant-owned checkout. ACP expands to product feeds/promotions; deeper merchant ChatGPT apps remain possible.
Availability at the event
New shopping discovery rollout that week; merchant-owned checkout supported; deeper apps remain possible. It is a strategy shift, not proof all agentic commerce ended.
Qualification
The primary does not supply conversion data or declare all commerce apps shut down. Retain merchant-app exceptions.
Source does not establish conversion decline or permanent abandonment of every in-chat commerce path.
Editorial interpretation
Initial checkout is revised toward discovery and merchant checkout
Editorial interpretation: Keeps the commercial story falsifiable: the platform itself describes a limitation and strategy change, while merchant-app exceptions remain supported.
“Commerce and fulfillment” is an analytical filter, not a source claim or an official historical period.
Date and documentary review
Dated primary update states focus/checkout policy and rollout that week.
Date precision
day
Source publication date
2026-03-24
Evidence snapshot
2026-10-11
Independent documentary review
Verified
Primary product update read including merchant checkout and Walmart exception; no conversion data audit.
'The foundation for AI-native commerce'; 'Connecting ChatGPT users and merchants'; web extract lines50–65
Google documents experimental Web Bot Auth signing for Google-Agent
Experimental; subset of requests signed
Google documented how to verify Web Bot Auth signatures on some Google-Agent requests, using the Signature-Agent identity https://agent.bot.goog and a published key directory.
Google documented how to verify Web Bot Auth signatures on some Google-Agent requests, using the Signature-Agent identity https://agent.bot.goog and a published key directory.
Availability at the event
Experimental. Google says a subset of Google-Agent requests are signed and that site owners should keep IP, reverse DNS and user-agent checks as a fallback.
Qualification
Web Bot Auth is an IETF draft and may change. Not every Google user agent or Google-Agent request is signed.
Page has been updated since the changelog date; current wording may differ from the May version.
Editorial interpretation
Cryptographic identity for agents arrives in Google's documentation
Editorial interpretation: A large operator documents signed agent requests alongside IP and DNS checks.
“Interoperability” is an analytical filter, not a source claim or an official historical period.
Date and documentary review
Dated entry in Google's crawling documentation changelog
Date precision
day
Source publication date
2026-05-04
Evidence snapshot
2026-11-05
Independent documentary review
Added 5 Nov 2026; second review pending
Primary documentation opened; key directory returned a key set on 5 November 2026; no signature verified here
Google announces Gemini Spark, a 24/7 personal agent
Trusted testers; U.S. Ultra beta planned
Google introduced Gemini Spark as a cloud-based personal agent in the Gemini app that keeps working when your devices are off, connects to Workspace and new MCP partners, and asks before high-stakes actions such as spending money or sending email.
Google introduced Gemini Spark as a cloud-based personal agent in the Gemini app that keeps working when your devices are off, connects to Workspace and new MCP partners, and asks before high-stakes actions such as spending money or sending email.
Availability at the event
Trusted testers the week of the announcement; a beta for U.S. Google AI Ultra subscribers the following week. Operating your local browser was listed as a future capability.
Qualification
Capabilities are vendor-described. Several features were future plans at announcement.
Announcement, not documented general availability.
Editorial interpretation
Always-on agents become a mainstream product category
Editorial interpretation: Marks Google's entry into persistent delegated agents, with staged access.
“Execution” is an analytical filter, not a source claim or an official historical period.
Date and documentary review
Dated Google announcement at I/O
Date precision
day
Source publication date
2026-05-19
Evidence snapshot
2026-11-05
Independent documentary review
Added 5 Nov 2026; second review pending
Primary announcement opened; vendor description only; no product test
Google Chrome; participating website developersInteroperability
Chrome invites live testing through a WebMCP origin trial
Time-limited Chrome origin trial
Chrome announced WebMCP origin-trial access in Chrome 149, allowing participating sites to test structured agent tools with live users and gather feedback.
Chrome announced WebMCP origin-trial access in Chrome 149, allowing participating sites to test structured agent tools with live users and gather feedback.
The exact earliest trial activation day is not established here. Current docs warn about human-in-loop focus, direct-visit discovery and complex-interface overhead.
May 18 doc publication is not trial activation. Still proposed standard. Security guidance acknowledges prompt-injection risk and evolving consent design.
Editorial interpretation
Sites can trial explicit browser-agent tools
Editorial interpretation: Represents WebMCP through its live-testing invitation rather than treating a proposal as a finished standard or broad deployment.
“Interoperability” is an analytical filter, not a source claim or an official historical period.
Date and documentary review
Date of Chrome's explicit invitation to join an available origin trial in Chrome 149, not a claim of first platform activation.
Date precision
day
Source publication date
2026-06-09
Evidence snapshot
2026-10-11
Independent documentary review
Verified with qualification
Primary source text inspected; no runtime test or audited adoption dataset
Published date; opening Chrome 149 and time-limited origin-trial definition.
Google; industry launch partners; Linux Foundation AI Catalog working groupInteroperability
Google publishes Agentic Resource Discovery as an open specification
Open specification; native cloud support future
ARD defined domain-hosted capability catalogs, federated registries and verifiable publisher metadata for discovering resources including MCP servers, A2A agents and OpenAPI tools.
ARD defined domain-hosted capability catalogs, federated registries and verifiable publisher metadata for discovering resources including MCP servers, A2A agents and OpenAPI tools.
Availability at the event
Open specification and reference materials available; native Agent Platform support and authenticated publisher onboarding still future-tense.
Qualification
Domain ownership and signatures establish provenance, not harmlessness or task-level authorization. Demo discovery/execution does not prove a large deployed federated network.
Publisher/domain provenance is distinct from harmlessness, correctness and user authority. Runtime demo does not prove a deployed interoperable federation.
Editorial interpretation
Capability discovery extends across organizations
Editorial interpretation: Separates finding tools/agents and establishing publisher provenance from connecting to them or executing a task.
“Interoperability” is an analytical filter, not a source claim or an official historical period.
Date and documentary review
Dated Google Developers announcement says the specification is available now.
Date precision
day
Source publication date
2026-06-17
Evidence snapshot
2026-10-11
Independent documentary review
Verified with qualification
Primary source text inspected; no runtime test or audited adoption dataset
Enterprise-Managed Authorization for MCP reaches stable status
Stable extension; named implementation boundaries
EMA standardized organization-controlled MCP access via an identity provider and ID-JAG exchange, reducing separate per-server interactive authorization flows.
EMA standardized organization-controlled MCP access via an identity provider and ID-JAG exchange, reducing separate per-server interactive authorization flows.
Availability at the event
Stable extension; Okta first supported identity provider; Claude shared MCP layer and VS Code listed as clients; named SaaS servers support it, with Slack still adding support.
Qualification
Compatibility and feature completeness were not tested. Organization/server policy determines scope; no benchmark proves stronger containment against malicious instructions.
Compatibility untested. Central access policy is not per-action consent or a containment proof against malicious instructions.
Editorial interpretation
Enterprise access becomes a shared authorization layer
Editorial interpretation: Makes identity-provider/client/server compatibility a concrete dependency; stable EMA does not establish per-action consent or prompt-injection containment.
“Governance” is an analytical filter, not a source claim or an official historical period.
Date and documentary review
Maintainer's dated post says extension is now stable.
Date precision
day
Source publication date
2026-06-18
Evidence snapshot
2026-10-11
Independent documentary review
Verified with qualification
Primary source text inspected; no runtime test or audited adoption dataset
ChatGPT Work combined longer tasks, connected apps/files and artifact creation with scheduled or triggered work.
Availability at the event
Paid web/mobile rollout excluding Free/Go; Pro/Pro Lite/Enterprise/Edu first, Plus/Business later; all supported regions. Desktop supported local apps/files with permission. Enterprise/Edu: two-week preview, off by default; admin opt-out.
Qualification
Announcement does not establish reliability or adoption.
Add Enterprise/Edu two-week preview, off by default with admin opt-out. Local app/file access requires permission. Official release note supports availability documentation, not independently tested reliability/adoption.
Editorial interpretation
Work agents add longer and scheduled workflows
Editorial interpretation: Shows movement toward ongoing delegated work across connected apps and artifacts, while preserving staged tiers and Enterprise/Edu administration limits.
“Execution” is an analytical filter, not a source claim or an official historical period.
Date and documentary review
Dated release-note rollout entry
Date precision
day
Source publication date
2026-07-09
Evidence snapshot
2026-10-11
Independent documentary review
Verified with qualification
Official dated release note directly opened; no reproduced task test
July 9, 2026 — Introducing ChatGPT Work and new desktop app
OpenAI announced Atlas deprecation, scheduled to stop working August 9, 2026, as browser capabilities moved into ChatGPT and Codex.
Availability at the event
Retirement announcement; current official help article confirms retired. Exact operational shutdown date was not observed.
Qualification
Do not use August 9 as confirmed observed shutdown.
Event date must remain notice date. Do not call August 9 an independently observed completed shutdown; deprecation/consolidation does not establish why product failed.
Editorial interpretation
Atlas is retired and capabilities are consolidated
Editorial interpretation: Preserves a meaningful product reversal; use the July notice date and avoid inventing an observed August shutdown timestamp or a causal failure story.
“Execution” is an analytical filter, not a source claim or an official historical period.
Date and documentary review
Retirement notice date, not observed shutdown date
Date precision
day
Source publication date
2026-07-09
Evidence snapshot
2026-10-11
Independent documentary review
Verified with qualification
Official dated notice, current launch banner and dedicated retirement article directly opened
MCP lead maintainers and ecosystem implementersInteroperability
MCP July revision releases a stateless core and updated SDKs
Final specification and SDK updates
MCP's revision removed the initialize handshake and protocol session header, introduced self-describing requests, Multi Round-Trip Requests, header routing and cache hints, and moved Tasks into an extension.
MCP's revision removed the initialize handshake and protocol session header, introduced self-describing requests, Multi Round-Trip Requests, header routing and cache hints, and moved Tasks into an extension.
Availability at the event
Final specification and TypeScript, Python, Go and C# SDK updates released; Rust SDK support beta; older implementations require deliberate migration/compatibility paths.
Qualification
Stateless transport does not force application state to disappear. Breaking changes, deprecated features and extension negotiation complicate cross-version support; download counts were not used as adoption proof.
Transport statelessness does not remove application state. Claimed scaling improvements and download volume are not independently measured workflow success.
Editorial interpretation
MCP evolves for ordinary web infrastructure
Editorial interpretation: Shows infrastructure maturation and migration costs, without turning stateless transport or SDK downloads into proof of workflow reliability.
“Interoperability” is an analytical filter, not a source claim or an official historical period.
Date and documentary review
Dated maintainer post explicitly pushes final specification and SDK release.
Date precision
day
Source publication date
2026-07-28
Evidence snapshot
2026-10-11
Independent documentary review
Verified with qualification
Primary source text inspected; no runtime test or audited adoption dataset
Date; opening; 'No handshake or sessions'; 'Tasks'; 'Deprecations'; 'SDKs'.
Google connected Gemini Spark to Chrome for web tasks. With permission it can use logged-in accounts and saved passwords, and it hands sensitive actions such as payments back to the user.
Google connected Gemini Spark to Chrome for web tasks. With permission it can use logged-in accounts and saved passwords, and it hands sensitive actions such as payments back to the user.
Availability at the event
Chrome browsing initially rolling out in the U.S.; Spark access expanded to Google AI Pro subscribers in more than 160 additional countries from July 30, 2026.
Qualification
Google cites prompt-injection protections; these are not independently tested here.
Rollout start; completion and reliability not documented.
Editorial interpretation
A personal agent acts inside the browser people already use
Editorial interpretation: Agent actions move into a mainstream browser session, with user-permission and payment handoff stated.
“Execution” is an analytical filter, not a source claim or an official historical period.
Date and documentary review
Dated Google announcement
Date precision
day
Source publication date
2026-07-30
Evidence snapshot
2026-11-05
Independent documentary review
Added 5 Nov 2026; second review pending
Primary announcement opened; vendor description only; no product test
OpenAI; WebMCP implementersInteroperabilityAdded 5 Nov 2026
ChatGPT site tools implement the proposed WebMCP standard
Desktop app built-in browser only
OpenAI documented site tools, its implementation of the proposed WebMCP standard. Websites can register tools that ChatGPT Work and Codex use on the same live page and signed-in session, in the ChatGPT desktop app's built-in browser.
OpenAI documented site tools, its implementation of the proposed WebMCP standard. Websites can register tools that ChatGPT Work and Codex use on the same live page and signed-in session, in the ChatGPT desktop app's built-in browser.
Availability at the event
Site tools run only in the ChatGPT desktop app's built-in browser, for ChatGPT Work and Codex, with a supported model and a page that provides tools. Users can turn them off.
Qualification
WebMCP is a proposed standard. The exact launch date is taken from a community announcement because the documentation page is undated.
Date precision depends on the community post.
Editorial interpretation
A second major agent platform consumes WebMCP tools
Editorial interpretation: Shows WebMCP moving from a Chrome origin trial to use by a separate agent platform.
“Interoperability” is an analytical filter, not a source claim or an official historical period.
Date and documentary review
Date of OpenAI Developer Community announcement; the documentation page shows no date
Date precision
day
Source publication date
2026-08-25
Evidence snapshot
2026-11-05
Independent documentary review
Added 5 Nov 2026; date from community post
Documentation and community post opened; no tool registered or tested here
Site tools documentation; community announcement dated August 25, 2026
Meta introduced Muse, a personal agent with a dedicated cloud VM and browser, messaging through its app or WhatsApp, background tasks and user approvals for sensitive actions. The announcement includes Stripe Link checkout.
Meta introduced Muse, a personal agent with a dedicated cloud VM and browser, messaging through its app or WhatsApp, background tasks and user approvals for sensitive actions. The announcement includes Stripe Link checkout.
Availability at the event
Rolling out in the U.S. on iOS, Android and muse.ai; free for most needs with subscriptions. AI glasses, Shop Pay, 1Password and Confidential VM were future promises in this post.
Qualification
Claims of negotiation, safety and broad task completion are not independently reproduced. No adoption or end-to-end reliability metrics supplied.
Retain explicit updated-page caveat. World-first/safe/billions/negotiation claims are vendor assertions, not established comparative or deployed reliability evidence. Original day-one feature matrix not fully recovered.
Editorial interpretation
A new platform enters the personal-agent market
Editorial interpretation: Adds Meta and persistent consumer delegation to the player story; retain the September 30 updated-page caveat and avoid backdating every feature to September 8.
“Execution” is an analytical filter, not a source claim or an official historical period.
Date and documentary review
Initial publication/launch date; page also displays September 30 update date
Date precision
day
Source publication date
2026-09-08
Evidence snapshot
2026-10-11
Independent documentary review
Verified with qualification
Primary announcement directly opened; updated page, vendor claims and demos; no independently tested deployment
September 8/September 30 date pair; How It Works; Built to be Private; Looking Ahead
Tapestry; Coach; Kate Spade; GoogleCommerce and fulfillment
Coach and Kate Spade confirm live U.S. UCP checkout on Google
Merchant-reported live U.S. availability
Tapestry confirmed live U.S. Coach and Kate Spade purchases in Google Search, AI Mode and Gemini, using single-item UCP/Google Pay checkout with explicit consumer approval.
Tapestry confirmed live U.S. Coach and Kate Spade purchases in Google Search, AI Mode and Gemini, using single-item UCP/Google Pay checkout with explicit consumer approval.
Availability at the event
Live in the U.S. by announcement date; two named brands; single-item checkout; explicit consumer approval; geographic expansion planned.
Qualification
No sales volume or purchase reproduction. Single-item checkout does not demonstrate unrestricted purchasing autonomy.
This is independently checked merchant-reported availability, not independent sales/uptime/accuracy validation. Google's hundreds-of-thousands claim is not audited. Human confirmation remains essential to this reported flow.
Editorial interpretation
Named merchants report live, limited checkout
Editorial interpretation: Ends with stronger implementation evidence from two named brands, while preserving live-by-date, single-item, U.S. and approval restrictions.
“Commerce and fulfillment” is an analytical filter, not a source claim or an official historical period.
Date and documentary review
Dated merchant release says now purchasable and live in U.S.; individual first transaction date not stated.
Date precision
day; availability confirmed by date
Source publication date
2026-09-16
Evidence snapshot
2026-10-11
Independent documentary review
Verified with qualification
Merchant primary release and platform corroboration read independently by author; no hands-on purchase or transaction audit.
September 16 release; opening bullets; How It Works; Consumer Control; Availability and Rollout; corroborating Google September 16 commerce update.
OpenAI introduces dots, always-on agents in ChatGPT
Pro rollout; regional exclusions
OpenAI's documentation describes a dot as an always-on agent with its own cloud computer and browser that keeps working when your devices are off, sends sign-in requests through a private form, and asks for approval before actions that affect your accounts.
OpenAI's documentation describes a dot as an always-on agent with its own cloud computer and browser that keeps working when your devices are off, sends sign-in requests through a private form, and asks for approval before actions that affect your accounts.
Availability at the event
Rollout quoted in the community announcement: ChatGPT web, mobile and desktop for Pro users, excluding the EEA, Switzerland and the UK. Documentation now also lists Business Premium and Enterprise (off by default).
Qualification
UNVERIFIED against the openai.com announcement page itself. Plan list in the documentation may reflect later expansion.
Primary announcement not opened; treat date and launch wording as provisional.
Editorial interpretation
Persistent agents arrive in ChatGPT
Editorial interpretation: Adds OpenAI's persistent agent to the same month as Meta and Anthropic launches.
“Execution” is an analytical filter, not a source claim or an official historical period.
Date and documentary review
DevDay 2026 date, from OpenAI Developer Community announcement and press coverage; the openai.com post could not be opened for this check
Date precision
day
Source publication date
2026-09-29
Evidence snapshot
2026-11-05
Independent documentary review
Added 5 Nov 2026; primary announcement not opened
OpenAI documentation and community thread opened; openai.com announcement page did not load for this check
Anthropic introduced Claude Duet, an app where Claude works as a proactive agent with its own virtual machine and browser. It monitors connected apps, sends a morning brief and must get approval before sending messages, making payments, signing documents or deleting anything.
Anthropic introduced Claude Duet, an app where Claude works as a proactive agent with its own virtual machine and browser. It monitors connected apps, sends a morning brief and must get approval before sending messages, making payments, signing documents or deleting anything.
Availability at the event
Research preview by waitlist, starting with U.S. Pro, Max and Ultra subscribers, on iOS, Android and claude.ai. Anthropic says it cannot yet show you its browser.
Qualification
Early version by Anthropic's own description; personal errands less developed. Some actions are off-limits even with approval.
Preview, not general availability.
Editorial interpretation
A third major lab ships a persistent agent within a month
Editorial interpretation: Completes a cluster of persistent-agent launches across OpenAI, Meta, Google and Anthropic.
“Execution” is an analytical filter, not a source claim or an official historical period.
Date and documentary review
Dated Anthropic announcement
Date precision
day
Source publication date
2026-10-21
Evidence snapshot
2026-11-05
Independent documentary review
Added 5 Nov 2026; second review pending
Primary announcement opened; vendor description only; no product test
Introducing Claude Duet; availability and approvals sections
Meta says more than 10 million people have set up Muse accounts
Company-stated figure via AP
The Associated Press reported that more than 10 million people in the U.S. and Canada had set up Muse accounts, and that about 25% of them use it on a given day, citing Meta.
The Associated Press reported that more than 10 million people in the U.S. and Canada had set up Muse accounts, and that about 25% of them use it on a given day, citing Meta.
Availability at the event
Figure covers U.S. and Canada accounts set up.
Qualification
A news report of a company statement, not an audited figure. Accounts set up are not active users. AP describes Muse as launched in August, while Meta's announcement in this chronology is dated 8 September 2026; that discrepancy is unresolved.
Secondary report; launch-month discrepancy with A016.
Editorial interpretation
First large usage figure for a persistent consumer agent
Editorial interpretation: Gives the persistent-agent cluster a dated, company-stated scale figure, kept separate from audited adoption.
“Execution” is an analytical filter, not a source claim or an official historical period.
Date and documentary review
Date of AP report on Meta's third-quarter results
Date precision
day
Source publication date
2026-10-21
Evidence snapshot
2026-11-05
Independent documentary review
Added 5 Nov 2026; Meta primary not checked
AP report opened (syndicated copy); Meta's own earnings materials not checked
32 separate Agentic Web events with historical availability and expandable documentary qualifications
Link
ExecutionDeveloper API beta
Anthropic
Anthropic releases screenshot-based computer use in public API beta
EvidenceA001
Documented fact
Claude 3.5 Sonnet's computer-use beta lets developers supply screenshots and execute model-generated cursor, click and keyboard actions. It launched on the Anthropic API, Amazon Bedrock and Google Cloud Vertex AI.
Availability at the event
Public developer beta at launch; experimental, cumbersome and error-prone. This was a model/tool capability requiring a software execution environment.
Experimental capability, not universal GUI reliability. OSWorld 14.9% screenshot-only and 22.0% extended-step results are vendor-reported benchmark evidence, not reproduced here.
Editorial interpretation
Computer use becomes a developer capability
Editorial interpretation: Establishes screenshot-to-action software as a concrete building block, with an experimental API beta and benchmark limits.
“Execution” is an analytical filter, not a source claim or an official historical period.
Date and documentary review
Dated announcement explicitly says available today
Date precision
day
Source publication date
2024-10-22
Evidence snapshot
2026-10-11
Independent documentary review
Verified with qualification
Primary announcement and primary research explanation directly opened; vendor-reported evaluation, no reproduced task test
Opening paragraphs; Teaching Claude to navigate computers, responsibly
Anthropic; MCP creators David Soria Parra and Justin Spahr-Summers
Anthropic open-sources MCP and local Claude Desktop integration
EvidenceP001
Documented fact
Anthropic released the MCP specification, SDKs, sample servers and local MCP-server support in Claude Desktop, providing a shared client/server interface for tools and data.
Availability at the event
Open source and local Desktop support available; remote production deployment toolkits were future work.
Qualification
Launch predominantly enabled local connectors. Block/Apollo integration is publisher-reported, and other companies were described as working with MCP.
Early integrations are vendor-reported; no measured adoption or dependable autonomous capability is established.
Editorial interpretation
A shared interface for tools and data
Editorial interpretation: Explains how agents acquire reusable tool/data access; local support and future remote tooling remain explicit.
“Interoperability” is an analytical filter, not a source claim or an official historical period.
Date and documentary review
Contemporaneous announcement says 'Today' and lists released components.
Date precision
day
Source publication date
2024-11-25
Evidence snapshot
2026-10-11
Independent documentary review
Verified
Primary source text inspected; no runtime test or audited adoption dataset
OpenAI releases Operator as a U.S. Pro research preview
EvidenceA002
Documented fact
Operator used a remote browser and Computer-Using Agent model to navigate sites by screenshots, typing, clicking and scrolling; examples included forms, groceries and bookings.
Availability at the event
Research preview for ChatGPT Pro users in the U.S.; integrated into ChatGPT agent on July 17, 2025.
Qualification
Login, payment details and CAPTCHAs required user takeover. Announcement does not prove broad adoption or unattended transaction reliability. OpenAI's same-day Computer-Using Agent research post reported 38.1% on OSWorld (human 72.4%), 58.1% on WebArena (human 78.2%) and 87% on WebVoyager. These are vendor-reported benchmarks, not reproduced here.
User takeover for login/payment/CAPTCHAs and approval before consequential actions. Examples and partnerships do not measure broad adoption or unattended success.
Editorial interpretation
A consumer browser-action preview
Editorial interpretation: Shows the transition from developer primitives to a user-facing web agent, while preserving U.S. Pro restrictions and takeover requirements.
“Execution” is an analytical filter, not a source claim or an official historical period.
Date and documentary review
Dated launch post says starting today
Date precision
day
Source publication date
2025-01-23
Evidence snapshot
2026-10-11
Independent documentary review
Verified with qualification
Primary launch text directly opened; demonstrations and benchmark claims not independently reproduced
January 23 date; opening; How to use; July 17 update
Google Cloud and more than 50 supporting technology/service partners
Google introduces the draft Agent2Agent protocol
EvidenceP002
Documented fact
Google introduced open-source A2A for capability discovery through Agent Cards, task lifecycle updates and artifact exchange between client and remote agents, building on HTTP, SSE and JSON-RPC.
Availability at the event
Draft specification, code samples and contribution paths released; a production-ready version was planned later in 2025.
Qualification
Partner support is not deployment/adoption measurement. The hiring scenario is illustrative; it does not establish completed cross-vendor transactions.
More than 50 supporters and illustrative hiring scenario do not prove deployed interoperability or completed workflows.
Editorial interpretation
Agents get a draft coordination protocol
Editorial interpretation: Introduces agent-to-agent discovery, tasks and artifacts as a separate layer from tool access; pairs with the later stable A2A milestone.
“Interoperability” is an analytical filter, not a source claim or an official historical period.
Date and documentary review
Dated launch announcement; linked specification explicitly called draft.
Date precision
day
Source publication date
2025-04-09
Evidence snapshot
2026-10-11
Independent documentary review
Verified
Primary source text inspected; no runtime test or audited adoption dataset
Date; 'How A2A works'; 'The future of agent interoperability', especially draft and future production version.
Coinbase launches x402 for HTTP-native stablecoin payments
EvidenceC002
Documented fact
Coinbase introduced x402, an open payment protocol that uses HTTP 402 to quote a resource and a signed stablecoin payment payload to authorize access. A payment facilitator verifies and settles payment before the server returns the resource.
Availability at the event
Public protocol launch and developer integration route; launch source describes stablecoin rails and example applications.
Qualification
Examples and partner logos are not transaction-volume proof. Facilitator, wallet, funding, authorization and implementation security remain dependencies.
Verification establishes the protocol launch and described flow. Collaborator lists/examples are not audited volumes or generic purchase authorization.
Editorial interpretation
Machines can pay for digital resources
Editorial interpretation: Broadens commerce beyond shopping carts to HTTP-native resource access, without implying transaction volume or universal authorization.
“Commerce and fulfillment” is an analytical filter, not a source claim or an official historical period.
Date and documentary review
Dated primary article says Coinbase is launching x402.
Date precision
day
Source publication date
2025-05-06
Evidence snapshot
2026-10-11
Independent documentary review
Verified
Vendor primary protocol-launch article read; flow inspected; no SDK execution or adoption audit.
Date; TL;DR; 'How x402 Works'; web extract lines46–59,84–97
ChatGPT agent combines research, browser action and artifact work
EvidenceA005
Documented fact
ChatGPT agent combined Operator's website interaction and deep research with a virtual computer, visual/text browsers, terminal and API access; it could research, navigate sites and produce editable artifacts.
Availability at the event
Pro by end of launch day; Plus/Team over following days; Enterprise/Education promised later. EEA and Switzerland were excluded at launch.
Qualification
Consequential actions need approval; prompt-injection and model-error risks remain. Vendor SpreadsheetBench score was 45.54% with XLSX versus human 71.33%.
This verifies launch documentation, not reproduced completion reliability. SpreadsheetBench 45.54% with XLSX vs 71.33% human is vendor-reported and uses differing evaluation environments; consequential actions require approval.
Editorial interpretation
Research, browser action and artifacts converge
Editorial interpretation: Provides a clear integrated-work-agent milestone and a dated phased rollout, while keeping approval and reliability limits visible.
“Execution” is an analytical filter, not a source claim or an official historical period.
Date and documentary review
Dated launch / start of phased rollout
Date precision
day
Source publication date
2025-07-17
Evidence snapshot
2026-10-11
Independent documentary review
Verified with qualification
Primary launch text directly opened; vendor evaluations/demos, no reproduced task test
A natural evolution; An agent that works; Availability; Limitations
Brave publishes Comet indirect prompt-injection proof of concept
EvidenceA006
Documented fact
Brave described a controlled proof of concept in which webpage instructions redirected Comet from summarizing a Reddit post to extracting an email address and one-time code from authenticated sessions.
Availability at the event
Public vulnerability-research report. Disclosure timeline records fixes and retesting; an update says the attack class was not fully mitigated then.
Qualification
Historical report, not a statement that current Comet remains vulnerable. Do not present lab demonstration as exploitation at scale.
Report includes patch/retest chronology and later incomplete-mitigation update. Brave is a competing browser vendor; exploit not reproduced here. Avoid current vulnerability or population impact claims.
Editorial interpretation
A controlled attack demonstrates a trust-boundary risk
Editorial interpretation: Provides concrete historical counterevidence to effortless autonomy; label the competing-vendor report and unreplicated proof of concept.
“Execution” is an analytical filter, not a source claim or an official historical period.
Date and documentary review
Public disclosure/publication date; discovery occurred July 25
Date precision
day
Source publication date
2025-08-20
Evidence snapshot
2026-10-11
Independent documentary review
Verified with qualification
Original researchers' report directly opened; proof of concept not reproduced; no real-world victim evidence
Attack demonstration; Disclosure timeline; Research Motivation
MCP Registry working group; contributors from Anthropic, GitHub, Block and PulseMCP
Official MCP Registry opens in preview
EvidenceP007
Documented fact
The official MCP Registry opened an API/catalog for public-server metadata, with an open-source OpenAPI specification intended to support downstream public and private subregistries.
Availability at the event
Public preview; no data-durability guarantee and possible breaking changes before general availability.
Qualification
Entries contain self-reported information. Moderation/denylisting is not a security certification; catalog publication does not prove safe installation or client integration.
Metadata is self-reported. Moderation and denylisting do not establish certification, safe installation or client compatibility.
Editorial interpretation
Tool discovery gets a shared catalog
Editorial interpretation: Makes distribution and discoverability intelligible as dependencies; preview status and self-reported metadata must stay visible.
“Interoperability” is an analytical filter, not a source claim or an official historical period.
Date and documentary review
Dated maintainers' post says registry is launching today in preview.
Date precision
day
Source publication date
2025-09-08
Evidence snapshot
2026-10-11
Independent documentary review
Verified with qualification
Primary source text inspected; no runtime test or audited adoption dataset
Date; opening; 'Public and private sub-registries'; 'Getting started' preview caveat.
Commerce and fulfillmentPublic specification and reference materials
Google; payments and technology collaborators
Google announces Agent Payments Protocol (AP2)
EvidenceC005
Documented fact
Google announced AP2, a payment-agnostic open protocol that can extend A2A and MCP. Cryptographically signed intent and cart mandates create an evidence chain for human-present purchases and delegated purchases. The primary source offers public specification, documentation and reference implementations.
Availability at the event
Public protocol/specification and reference implementations; examples are proposed use cases; no general production checkout deployment proven.
Qualification
Support from more than 60 organizations is collaboration evidence, not deployment/adoption count. Cryptographic evidence does not eliminate fraud, mistaken intent or legal responsibility.
Sixty-plus collaborators do not establish sixty production integrations. Historical launch mandate terminology must be distinguished from later specification evolution; FIDO development is not a finished ratified standard.
Editorial interpretation
Payment intent and consent get a protocol
Editorial interpretation: Introduces the authorization-evidence layer, distinct from payment settlement or a deployed shopping service.
“Commerce and fulfillment” is an analytical filter, not a source claim or an official historical period.
Date and documentary review
Primary article display date September 16 and 'Today' announcement; protocol site also dates announcement September 16.
Date precision
day
Source publication date
2025-09-16
Evidence snapshot
2026-10-11
Independent documentary review
Verified
Primary announcement read including mandate flow and public implementation route; no code execution.
Display date; opening announcement; 'How it works: Establishing trust via mandates and verifiable credentials'; 'What’s next'
ChatGPT launches U.S. Instant Checkout and open ACP
EvidenceC007
Documented fact
OpenAI launched Instant Checkout for U.S. ChatGPT Free, Plus and Pro users buying from U.S. Etsy sellers and published the Agentic Commerce Protocol with Stripe. Launch checkout supported single-item purchases; Shopify merchants, multi-item carts and additional regions were future plans.
Availability at the event
Limited public availability: U.S. users/U.S. Etsy; user confirmation; single-item checkout. Shopify listed as coming soon.
Qualification
Do not report a million Shopify merchants live on launch day. Native checkout is an integration, not open-ended delegated purchasing. See C014 for later strategy change.
These are historical launch limits. C014 records later change. No checkout was executed during verification.
Editorial interpretation
In-chat checkout launches within a narrow scope
Editorial interpretation: Anchors commercial execution in U.S. Etsy single-item checkout and user confirmation; pairs with the later strategy revision.
“Commerce and fulfillment” is an analytical filter, not a source claim or an official historical period.
Date and documentary review
Dated primary launch says 'Starting today' and 'can now buy'.
OpenAI launches Atlas with paid-plan agent-mode preview on macOS
EvidenceA009
Documented fact
Atlas launched as a browser with ChatGPT, optional browser memories and agent mode using the user's browsing context. Browser access and action-agent access had different plan gates.
Availability at the event
Worldwide macOS browser launch for Free/Plus/Pro/Go; Business beta and admin-enabled Enterprise/Edu. Agent mode preview for Plus/Pro/Business. The current launch page says Atlas has since been deprecated.
Qualification
At launch the agent could not run browser code, download files, install extensions or access other computer apps/files; workflow mistakes remained possible.
Current page says deprecated. Agent at launch lacked code execution, downloads, extensions and other-app/filesystem access. Do not imply present availability or proven reliable transactions.
Editorial interpretation
Agent actions move into a browser product
Editorial interpretation: Documents the Atlas browser/agent-mode availability split; pairs with retirement so the chronology does not imply every product persists.
“Execution” is an analytical filter, not a source claim or an official historical period.
Date and documentary review
Dated launch says launching today
Date precision
day
Source publication date
2025-10-21
Evidence snapshot
2026-10-11
Independent documentary review
Verified with qualification
Primary launch text directly opened with current product-status banner; no reproduced task test
Opening availability paragraphs; Get work done for you; current deprecation banner
GovernanceFoundation/project contributions announced
Linux Foundation; Anthropic, Block, OpenAI; supporting platform companies
Agentic AI Foundation launches with MCP, goose and AGENTS.md
EvidenceP009
Documented fact
AAIF was formed under the Linux Foundation with founding contributions of MCP, Block's goose agent framework and OpenAI's AGENTS.md instruction convention.
Availability at the event
Foundation and project contributions announced. Infrastructure already open source; stewardship change is the event.
Qualification
Claims of 'universal' standards and published-server/project totals are foundation/vendor assertions, not independent market-share measurements. Governance does not guarantee safe execution.
Open governance does not certify execution safety; universal-standard language and counts are institutional/vendor claims.
Editorial interpretation
Open projects gain a shared institutional home
Editorial interpretation: Groups MCP, goose and AGENTS.md without conflating their technical roles; introduces stewardship rather than another capability claim.
“Governance” is an analytical filter, not a source claim or an official historical period.
Date and documentary review
AAIF/Linux Foundation dated press release explicitly announces formation and founding project contributions.
Date precision
day
Source publication date
2025-12-09
Evidence snapshot
2026-10-11
Independent documentary review
Verified
Primary source text inspected; no runtime test or audited adoption dataset
Date; summary; formation paragraph and individual project descriptions.
Commerce and fulfillmentPublic protocol; checkout prospective
Google; Shopify; Etsy; Wayfair; Target; Walmart
Google and retail partners launch Universal Commerce Protocol
EvidenceC011
Documented fact
Google announced UCP, co-developed with retail/platform partners to provide a common commerce language from discovery and purchase through post-purchase support. It stated compatibility with A2A, AP2 and MCP; eligible U.S. native Google checkout would arrive soon.
Availability at the event
Public protocol launch; native AI Mode/Gemini checkout prospective at announcement; seller remains merchant of record.
Qualification
Partner endorsements are not production deployments. Business Agent/Direct Offers are distinct features with their own scopes. Do not call every retailer UCP-ready.
Verify protocol launch, not all future full-lifecycle functions deployed. Business Agent Jan 12 and Direct Offers pilot are distinct products, not evidence of UCP merchant adoption.
Editorial interpretation
Retailers and platforms define a commerce lifecycle
Editorial interpretation: Introduces UCP as an open protocol spanning commerce stages, retaining the distinction between the protocol launch and prospective native checkout.
“Commerce and fulfillment” is an analytical filter, not a source claim or an official historical period.
Date and documentary review
Dated primary blog says 'Today' launching UCP.
Date precision
day
Source publication date
2026-01-11
Evidence snapshot
2026-10-11
Independent documentary review
Verified
Google primary launch and Shopify co-developer release read; protocol/checkout not executed.
'Unlock retail growth with a new open standard'; web extract lines43–52; Shopify Jan11 corroboration
MCP community; Anthropic, OpenAI, MCP-UI creators; Block and Microsoft clients
MCP Apps becomes an official interactive-UI extension
EvidenceP010
Documented fact
MCP Apps standardized tool-linked HTML/JavaScript interfaces rendered in sandboxed iframes within a conversation, with bidirectional host communication and SDK/examples.
Availability at the event
Official extension/SDK released; Claude and Goose available that day, VS Code Insiders support, ChatGPT starting that week.
Qualification
Host support and restrictions vary. Sandboxing and optional consent are mitigations, not proof that arbitrary server-supplied UI is safe.
Host support/restrictions differ; iframe isolation and consent mechanisms are mitigations, not proof arbitrary server-supplied UI is safe.
Editorial interpretation
People can inspect and decide inside agent interfaces
Editorial interpretation: Shows why interactive UI matters for human participation, while retaining the client-by-client availability boundary.
“Interoperability” is an analytical filter, not a source claim or an official historical period.
Date and documentary review
Dated maintainers' release says extension is now live; precise client rollout qualifiers in 'Client Support'.
Date precision
day
Source publication date
2026-01-26
Evidence snapshot
2026-10-11
Independent documentary review
Verified with qualification
Primary source text inspected; no runtime test or audited adoption dataset
Date; opening; 'How It Works'; 'Security Model'; detailed 'Client Support'.
InteroperabilityStable specification and migration guidance
A2A community and Technical Steering Committee
A2A v1.0 becomes the project's first stable specification
EvidenceP012
Documented fact
A2A v1.0 formalized multiple protocol bindings, version negotiation, multi-tenancy, signed Agent Cards and revised security flows for cross-agent coordination.
Availability at the event
First stable/production-ready specification and migration materials available, according to project; SDK support across programming languages remained ongoing work.
Qualification
Breaking interaction changes require migration. AgentCard supports advertising v0.3 and v1.0 together; do not assume wire-level compatibility without implementation checks.
Interaction protocol has breaking changes; AgentCards can advertise both v0.3/v1.0. Prefer SDK support across programming languages over ambiguous multilingual SDK wording.
Editorial interpretation
A2A reaches a stable specification
Editorial interpretation: Closes the draft-to-stable arc begun by P002; project-characterized readiness and breaking interaction changes remain qualified.
“Interoperability” is an analytical filter, not a source claim or an official historical period.
Date and documentary review
Dated A2A project release says community is announcing v1.0 today.
Date precision
day
Source publication date
2026-03-12
Evidence snapshot
2026-10-11
Independent documentary review
Verified with qualification
Primary source text inspected; no runtime test or audited adoption dataset
Metadata date; first paragraph; enterprise requirements; migration; closing SDK sentence.
Google adds the Google-Agent user agent for agents acting on user request
EvidenceP017
Documented fact
Google added Google-Agent to its user-triggered fetchers. Agents hosted on Google infrastructure use it to browse and take actions when a user asks, and Google published a separate IP-range file for user-triggered agents.
Availability at the event
Changelog says the user agent was rolling out over the following weeks.
Qualification
Google says user-triggered fetchers generally ignore robots.txt. Documentation describes identity and IP ranges, not how often or where Google-Agent is used.
Documentation change; it does not measure agent traffic volume.
Editorial interpretation
Agents get their own identity in server logs
Editorial interpretation: Gives site owners a named, verifiable identity for agent traffic, distinct from Googlebot crawling.
“Interoperability” is an analytical filter, not a source claim or an official historical period.
Date and documentary review
Dated entry in Google's crawling documentation changelog
Date precision
day
Source publication date
2026-03-20
Evidence snapshot
2026-11-05
Independent documentary review
Added 5 Nov 2026; second review pending
Primary documentation opened; the IP-range file returned HTTP 200 on 5 November 2026; no traffic test
Changelog March 2026 entry; user-triggered fetchers page, Google-Agent section
Commerce and fulfillmentStrategy change and weekly rollout
OpenAI; merchant partners; Shopify; Walmart
OpenAI shifts shopping toward discovery and merchant checkout
EvidenceC014
Documented fact
OpenAI said initial Instant Checkout lacked desired flexibility and shifted focus to discovery with merchant-owned checkout. ACP expands to product feeds/promotions; deeper merchant ChatGPT apps remain possible.
Availability at the event
New shopping discovery rollout that week; merchant-owned checkout supported; deeper apps remain possible. It is a strategy shift, not proof all agentic commerce ended.
Qualification
The primary does not supply conversion data or declare all commerce apps shut down. Retain merchant-app exceptions.
Source does not establish conversion decline or permanent abandonment of every in-chat commerce path.
Editorial interpretation
Initial checkout is revised toward discovery and merchant checkout
Editorial interpretation: Keeps the commercial story falsifiable: the platform itself describes a limitation and strategy change, while merchant-app exceptions remain supported.
“Commerce and fulfillment” is an analytical filter, not a source claim or an official historical period.
Date and documentary review
Dated primary update states focus/checkout policy and rollout that week.
Date precision
day
Source publication date
2026-03-24
Evidence snapshot
2026-10-11
Independent documentary review
Verified
Primary product update read including merchant checkout and Walmart exception; no conversion data audit.
'The foundation for AI-native commerce'; 'Connecting ChatGPT users and merchants'; web extract lines50–65
InteroperabilityExperimental; subset of requests signed
Google
Google documents experimental Web Bot Auth signing for Google-Agent
EvidenceP018
Documented fact
Google documented how to verify Web Bot Auth signatures on some Google-Agent requests, using the Signature-Agent identity https://agent.bot.goog and a published key directory.
Availability at the event
Experimental. Google says a subset of Google-Agent requests are signed and that site owners should keep IP, reverse DNS and user-agent checks as a fallback.
Qualification
Web Bot Auth is an IETF draft and may change. Not every Google user agent or Google-Agent request is signed.
Page has been updated since the changelog date; current wording may differ from the May version.
Editorial interpretation
Cryptographic identity for agents arrives in Google's documentation
Editorial interpretation: A large operator documents signed agent requests alongside IP and DNS checks.
“Interoperability” is an analytical filter, not a source claim or an official historical period.
Date and documentary review
Dated entry in Google's crawling documentation changelog
Date precision
day
Source publication date
2026-05-04
Evidence snapshot
2026-11-05
Independent documentary review
Added 5 Nov 2026; second review pending
Primary documentation opened; key directory returned a key set on 5 November 2026; no signature verified here
Google announces Gemini Spark, a 24/7 personal agent
EvidenceA017
Documented fact
Google introduced Gemini Spark as a cloud-based personal agent in the Gemini app that keeps working when your devices are off, connects to Workspace and new MCP partners, and asks before high-stakes actions such as spending money or sending email.
Availability at the event
Trusted testers the week of the announcement; a beta for U.S. Google AI Ultra subscribers the following week. Operating your local browser was listed as a future capability.
Qualification
Capabilities are vendor-described. Several features were future plans at announcement.
Announcement, not documented general availability.
Editorial interpretation
Always-on agents become a mainstream product category
Editorial interpretation: Marks Google's entry into persistent delegated agents, with staged access.
“Execution” is an analytical filter, not a source claim or an official historical period.
Date and documentary review
Dated Google announcement at I/O
Date precision
day
Source publication date
2026-05-19
Evidence snapshot
2026-11-05
Independent documentary review
Added 5 Nov 2026; second review pending
Primary announcement opened; vendor description only; no product test
Chrome invites live testing through a WebMCP origin trial
EvidenceP013
Documented fact
Chrome announced WebMCP origin-trial access in Chrome 149, allowing participating sites to test structured agent tools with live users and gather feedback.
The exact earliest trial activation day is not established here. Current docs warn about human-in-loop focus, direct-visit discovery and complex-interface overhead.
May 18 doc publication is not trial activation. Still proposed standard. Security guidance acknowledges prompt-injection risk and evolving consent design.
Editorial interpretation
Sites can trial explicit browser-agent tools
Editorial interpretation: Represents WebMCP through its live-testing invitation rather than treating a proposal as a finished standard or broad deployment.
“Interoperability” is an analytical filter, not a source claim or an official historical period.
Date and documentary review
Date of Chrome's explicit invitation to join an available origin trial in Chrome 149, not a claim of first platform activation.
Date precision
day
Source publication date
2026-06-09
Evidence snapshot
2026-10-11
Independent documentary review
Verified with qualification
Primary source text inspected; no runtime test or audited adoption dataset
Published date; opening Chrome 149 and time-limited origin-trial definition.
InteroperabilityOpen specification; native cloud support future
Google; industry launch partners; Linux Foundation AI Catalog working group
Google publishes Agentic Resource Discovery as an open specification
EvidenceP014
Documented fact
ARD defined domain-hosted capability catalogs, federated registries and verifiable publisher metadata for discovering resources including MCP servers, A2A agents and OpenAPI tools.
Availability at the event
Open specification and reference materials available; native Agent Platform support and authenticated publisher onboarding still future-tense.
Qualification
Domain ownership and signatures establish provenance, not harmlessness or task-level authorization. Demo discovery/execution does not prove a large deployed federated network.
Publisher/domain provenance is distinct from harmlessness, correctness and user authority. Runtime demo does not prove a deployed interoperable federation.
Editorial interpretation
Capability discovery extends across organizations
Editorial interpretation: Separates finding tools/agents and establishing publisher provenance from connecting to them or executing a task.
“Interoperability” is an analytical filter, not a source claim or an official historical period.
Date and documentary review
Dated Google Developers announcement says the specification is available now.
Date precision
day
Source publication date
2026-06-17
Evidence snapshot
2026-10-11
Independent documentary review
Verified with qualification
Primary source text inspected; no runtime test or audited adoption dataset
Enterprise-Managed Authorization for MCP reaches stable status
EvidenceP015
Documented fact
EMA standardized organization-controlled MCP access via an identity provider and ID-JAG exchange, reducing separate per-server interactive authorization flows.
Availability at the event
Stable extension; Okta first supported identity provider; Claude shared MCP layer and VS Code listed as clients; named SaaS servers support it, with Slack still adding support.
Qualification
Compatibility and feature completeness were not tested. Organization/server policy determines scope; no benchmark proves stronger containment against malicious instructions.
Compatibility untested. Central access policy is not per-action consent or a containment proof against malicious instructions.
Editorial interpretation
Enterprise access becomes a shared authorization layer
Editorial interpretation: Makes identity-provider/client/server compatibility a concrete dependency; stable EMA does not establish per-action consent or prompt-injection containment.
“Governance” is an analytical filter, not a source claim or an official historical period.
Date and documentary review
Maintainer's dated post says extension is now stable.
Date precision
day
Source publication date
2026-06-18
Evidence snapshot
2026-10-11
Independent documentary review
Verified with qualification
Primary source text inspected; no runtime test or audited adoption dataset
ChatGPT Work combined longer tasks, connected apps/files and artifact creation with scheduled or triggered work.
Availability at the event
Paid web/mobile rollout excluding Free/Go; Pro/Pro Lite/Enterprise/Edu first, Plus/Business later; all supported regions. Desktop supported local apps/files with permission. Enterprise/Edu: two-week preview, off by default; admin opt-out.
Qualification
Announcement does not establish reliability or adoption.
Add Enterprise/Edu two-week preview, off by default with admin opt-out. Local app/file access requires permission. Official release note supports availability documentation, not independently tested reliability/adoption.
Editorial interpretation
Work agents add longer and scheduled workflows
Editorial interpretation: Shows movement toward ongoing delegated work across connected apps and artifacts, while preserving staged tiers and Enterprise/Edu administration limits.
“Execution” is an analytical filter, not a source claim or an official historical period.
Date and documentary review
Dated release-note rollout entry
Date precision
day
Source publication date
2026-07-09
Evidence snapshot
2026-10-11
Independent documentary review
Verified with qualification
Official dated release note directly opened; no reproduced task test
July 9, 2026 — Introducing ChatGPT Work and new desktop app
ExecutionRetirement notice; current retired status confirmed
OpenAI
OpenAI announces Atlas retirement and migration to ChatGPT/Codex
EvidenceA015
Documented fact
OpenAI announced Atlas deprecation, scheduled to stop working August 9, 2026, as browser capabilities moved into ChatGPT and Codex.
Availability at the event
Retirement announcement; current official help article confirms retired. Exact operational shutdown date was not observed.
Qualification
Do not use August 9 as confirmed observed shutdown.
Event date must remain notice date. Do not call August 9 an independently observed completed shutdown; deprecation/consolidation does not establish why product failed.
Editorial interpretation
Atlas is retired and capabilities are consolidated
Editorial interpretation: Preserves a meaningful product reversal; use the July notice date and avoid inventing an observed August shutdown timestamp or a causal failure story.
“Execution” is an analytical filter, not a source claim or an official historical period.
Date and documentary review
Retirement notice date, not observed shutdown date
Date precision
day
Source publication date
2026-07-09
Evidence snapshot
2026-10-11
Independent documentary review
Verified with qualification
Official dated notice, current launch banner and dedicated retirement article directly opened
InteroperabilityFinal specification and SDK updates
MCP lead maintainers and ecosystem implementers
MCP July revision releases a stateless core and updated SDKs
EvidenceP016
Documented fact
MCP's revision removed the initialize handshake and protocol session header, introduced self-describing requests, Multi Round-Trip Requests, header routing and cache hints, and moved Tasks into an extension.
Availability at the event
Final specification and TypeScript, Python, Go and C# SDK updates released; Rust SDK support beta; older implementations require deliberate migration/compatibility paths.
Qualification
Stateless transport does not force application state to disappear. Breaking changes, deprecated features and extension negotiation complicate cross-version support; download counts were not used as adoption proof.
Transport statelessness does not remove application state. Claimed scaling improvements and download volume are not independently measured workflow success.
Editorial interpretation
MCP evolves for ordinary web infrastructure
Editorial interpretation: Shows infrastructure maturation and migration costs, without turning stateless transport or SDK downloads into proof of workflow reliability.
“Interoperability” is an analytical filter, not a source claim or an official historical period.
Date and documentary review
Dated maintainer post explicitly pushes final specification and SDK release.
Date precision
day
Source publication date
2026-07-28
Evidence snapshot
2026-10-11
Independent documentary review
Verified with qualification
Primary source text inspected; no runtime test or audited adoption dataset
Date; opening; 'No handshake or sessions'; 'Tasks'; 'Deprecations'; 'SDKs'.
Google connected Gemini Spark to Chrome for web tasks. With permission it can use logged-in accounts and saved passwords, and it hands sensitive actions such as payments back to the user.
Availability at the event
Chrome browsing initially rolling out in the U.S.; Spark access expanded to Google AI Pro subscribers in more than 160 additional countries from July 30, 2026.
Qualification
Google cites prompt-injection protections; these are not independently tested here.
Rollout start; completion and reliability not documented.
Editorial interpretation
A personal agent acts inside the browser people already use
Editorial interpretation: Agent actions move into a mainstream browser session, with user-permission and payment handoff stated.
“Execution” is an analytical filter, not a source claim or an official historical period.
Date and documentary review
Dated Google announcement
Date precision
day
Source publication date
2026-07-30
Evidence snapshot
2026-11-05
Independent documentary review
Added 5 Nov 2026; second review pending
Primary announcement opened; vendor description only; no product test
ChatGPT site tools implement the proposed WebMCP standard
EvidenceP019
Documented fact
OpenAI documented site tools, its implementation of the proposed WebMCP standard. Websites can register tools that ChatGPT Work and Codex use on the same live page and signed-in session, in the ChatGPT desktop app's built-in browser.
Availability at the event
Site tools run only in the ChatGPT desktop app's built-in browser, for ChatGPT Work and Codex, with a supported model and a page that provides tools. Users can turn them off.
Qualification
WebMCP is a proposed standard. The exact launch date is taken from a community announcement because the documentation page is undated.
Date precision depends on the community post.
Editorial interpretation
A second major agent platform consumes WebMCP tools
Editorial interpretation: Shows WebMCP moving from a Chrome origin trial to use by a separate agent platform.
“Interoperability” is an analytical filter, not a source claim or an official historical period.
Date and documentary review
Date of OpenAI Developer Community announcement; the documentation page shows no date
Date precision
day
Source publication date
2026-08-25
Evidence snapshot
2026-11-05
Independent documentary review
Added 5 Nov 2026; date from community post
Documentation and community post opened; no tool registered or tested here
Site tools documentation; community announcement dated August 25, 2026
Meta introduced Muse, a personal agent with a dedicated cloud VM and browser, messaging through its app or WhatsApp, background tasks and user approvals for sensitive actions. The announcement includes Stripe Link checkout.
Availability at the event
Rolling out in the U.S. on iOS, Android and muse.ai; free for most needs with subscriptions. AI glasses, Shop Pay, 1Password and Confidential VM were future promises in this post.
Qualification
Claims of negotiation, safety and broad task completion are not independently reproduced. No adoption or end-to-end reliability metrics supplied.
Retain explicit updated-page caveat. World-first/safe/billions/negotiation claims are vendor assertions, not established comparative or deployed reliability evidence. Original day-one feature matrix not fully recovered.
Editorial interpretation
A new platform enters the personal-agent market
Editorial interpretation: Adds Meta and persistent consumer delegation to the player story; retain the September 30 updated-page caveat and avoid backdating every feature to September 8.
“Execution” is an analytical filter, not a source claim or an official historical period.
Date and documentary review
Initial publication/launch date; page also displays September 30 update date
Date precision
day
Source publication date
2026-09-08
Evidence snapshot
2026-10-11
Independent documentary review
Verified with qualification
Primary announcement directly opened; updated page, vendor claims and demos; no independently tested deployment
September 8/September 30 date pair; How It Works; Built to be Private; Looking Ahead
Commerce and fulfillmentMerchant-reported live U.S. availability
Tapestry; Coach; Kate Spade; Google
Coach and Kate Spade confirm live U.S. UCP checkout on Google
EvidenceC017
Documented fact
Tapestry confirmed live U.S. Coach and Kate Spade purchases in Google Search, AI Mode and Gemini, using single-item UCP/Google Pay checkout with explicit consumer approval.
Availability at the event
Live in the U.S. by announcement date; two named brands; single-item checkout; explicit consumer approval; geographic expansion planned.
Qualification
No sales volume or purchase reproduction. Single-item checkout does not demonstrate unrestricted purchasing autonomy.
This is independently checked merchant-reported availability, not independent sales/uptime/accuracy validation. Google's hundreds-of-thousands claim is not audited. Human confirmation remains essential to this reported flow.
Editorial interpretation
Named merchants report live, limited checkout
Editorial interpretation: Ends with stronger implementation evidence from two named brands, while preserving live-by-date, single-item, U.S. and approval restrictions.
“Commerce and fulfillment” is an analytical filter, not a source claim or an official historical period.
Date and documentary review
Dated merchant release says now purchasable and live in U.S.; individual first transaction date not stated.
Date precision
day; availability confirmed by date
Source publication date
2026-09-16
Evidence snapshot
2026-10-11
Independent documentary review
Verified with qualification
Merchant primary release and platform corroboration read independently by author; no hands-on purchase or transaction audit.
September 16 release; opening bullets; How It Works; Consumer Control; Availability and Rollout; corroborating Google September 16 commerce update.
OpenAI introduces dots, always-on agents in ChatGPT
EvidenceA019
Documented fact
OpenAI's documentation describes a dot as an always-on agent with its own cloud computer and browser that keeps working when your devices are off, sends sign-in requests through a private form, and asks for approval before actions that affect your accounts.
Availability at the event
Rollout quoted in the community announcement: ChatGPT web, mobile and desktop for Pro users, excluding the EEA, Switzerland and the UK. Documentation now also lists Business Premium and Enterprise (off by default).
Qualification
UNVERIFIED against the openai.com announcement page itself. Plan list in the documentation may reflect later expansion.
Primary announcement not opened; treat date and launch wording as provisional.
Editorial interpretation
Persistent agents arrive in ChatGPT
Editorial interpretation: Adds OpenAI's persistent agent to the same month as Meta and Anthropic launches.
“Execution” is an analytical filter, not a source claim or an official historical period.
Date and documentary review
DevDay 2026 date, from OpenAI Developer Community announcement and press coverage; the openai.com post could not be opened for this check
Date precision
day
Source publication date
2026-09-29
Evidence snapshot
2026-11-05
Independent documentary review
Added 5 Nov 2026; primary announcement not opened
OpenAI documentation and community thread opened; openai.com announcement page did not load for this check
Anthropic introduced Claude Duet, an app where Claude works as a proactive agent with its own virtual machine and browser. It monitors connected apps, sends a morning brief and must get approval before sending messages, making payments, signing documents or deleting anything.
Availability at the event
Research preview by waitlist, starting with U.S. Pro, Max and Ultra subscribers, on iOS, Android and claude.ai. Anthropic says it cannot yet show you its browser.
Qualification
Early version by Anthropic's own description; personal errands less developed. Some actions are off-limits even with approval.
Preview, not general availability.
Editorial interpretation
A third major lab ships a persistent agent within a month
Editorial interpretation: Completes a cluster of persistent-agent launches across OpenAI, Meta, Google and Anthropic.
“Execution” is an analytical filter, not a source claim or an official historical period.
Date and documentary review
Dated Anthropic announcement
Date precision
day
Source publication date
2026-10-21
Evidence snapshot
2026-11-05
Independent documentary review
Added 5 Nov 2026; second review pending
Primary announcement opened; vendor description only; no product test
Introducing Claude Duet; availability and approvals sections
Meta says more than 10 million people have set up Muse accounts
EvidenceA021
Documented fact
The Associated Press reported that more than 10 million people in the U.S. and Canada had set up Muse accounts, and that about 25% of them use it on a given day, citing Meta.
Availability at the event
Figure covers U.S. and Canada accounts set up.
Qualification
A news report of a company statement, not an audited figure. Accounts set up are not active users. AP describes Muse as launched in August, while Meta's announcement in this chronology is dated 8 September 2026; that discrepancy is unresolved.
Secondary report; launch-month discrepancy with A016.
Editorial interpretation
First large usage figure for a persistent consumer agent
Editorial interpretation: Gives the persistent-agent cluster a dated, company-stated scale figure, kept separate from audited adoption.
“Execution” is an analytical filter, not a source claim or an official historical period.
Date and documentary review
Date of AP report on Meta's third-quarter results
Date precision
day
Source publication date
2026-10-21
Evidence snapshot
2026-11-05
Independent documentary review
Added 5 Nov 2026; Meta primary not checked
AP report opened (syndicated copy); Meta's own earnings materials not checked
This selected chronology documents building blocks and commercial revisions. Source announcements and independent documentary review do not demonstrate audited adoption, safe unattended transactions, or reproduced task success.
Prepared 12 October 2026 from the independently reviewed 11 October documentary snapshot. Events marked “Added” were checked against their sources on 5 November 2026 and have not yet had a second independent review. Exact event dates, source publication dates and availability qualifications are shown separately in each record.