Will Anthropic launch a Muse-like personal agent before 2027?

- [2026-10-14] Anthropic's Claude Cowork help page, updated today, says Cowork is on paid plans only (Pro, Max, Team, Enterprise). It runs tasks "in the cloud" on Anthropic's servers in an isolated environment, and scheduled tasks run there with no device online. It can open sites, click, type and fill forms, and use connectors. The page also says Cowork "is now just Claude", with chat and Cowork merging. It does not describe a VM dedicated to the user that persists between tasks, and it still says sessions are "isolated" and task-scoped. https://support.claude.com/en/articles/13345190-get-started-with-claude-cowork - [2026-10-06] On Pro and Max, new Cowork tasks now start in Anthropic's cloud, and the "only on your computer" option was removed. Tasks that need local files, the browser or computer use still need the Claude Desktop app open. Source: secondary coverage of the Anthropic help-center change, plus the help page above. https://cellcog.ai/blog/claude-cowork-cloud-only/ - [2026-10-03] The New Stack's piece says Anthropic's answer to Muse and OpenAI's Dots is "already inside Claude" (Cowork merged into Claude, Claude Tag in Slack since June). That is commentary, not an Anthropic statement that Cowork is a Muse-like agent. It also says Meta's Muse is the only one of these aimed at consumers (see the 2026-10-14 comparison below). https://thenewstack.io/claude-answer-to-dots-muse/ - [2026-09-16 to 2026-07-07] Cowork timeline from a secondary tracker: research preview 2026-01-12, scheduled tasks 2026-02-25, GA on desktop 2026-04-09, web and mobile beta 2026-07-07, Chrome side panel 2026-08-12, merge with chat from 2026-09-16. https://munderdiffl.in/blog/what-is-claude-cowork/ - [2026-09-23] Claude Code cloud sessions went GA. They run in task-scoped Anthropic-managed VMs for coding, and "routines" run on a schedule. Coding-only, so likely excluded by the single-task clause. https://www.agent37.com/blog/claude-code-cloud - [2026-09-17] Claude Code Projects launched in beta for select Pro and Max users. It uses parallel cloud threads that keep running after the laptop closes, and it is coding-focused. https://www.marktechpost.com/2026/09/17/anthropic-launches-claude-code-projects-in-beta-parallel-cloud-sessions-that-keep-running-after-you-close-your-laptop/ - [2026-04 (public beta)] Claude Managed Agents is an API for developers to run cloud-hosted agents. It is API and business oriented, so it is excluded by the criteria. https://thenewstack.io/with-claude-managed-agents-anthropic-wants-to-run-your-ai-agents-for-you/ - [2026-09-08] Meta's Muse launch post, the benchmark in the question, describes a dedicated "Muse Secure VM" per person that keeps working after the app is closed. It is proactive, can fill forms and send email, and is rolling out in the US (iOS, Android, web). It is free for most uses, with subscription plans. https://about.fb.com/news/2026/09/introducing-muse-personal-ai-agent/ - [2026-10-12] Meta launched the Muse Wonder models. Muse has about 9M downloads, with growth slowing, per the WSJ as relayed by roic.ai. OpenAI unveiled its always-on agent "Dot(s)" on 2026-09-29, with a similar cloud design (BigGo summary). Competitive pressure is rising. https://www.roic.ai/news/meta-unveils-muse-wonder-to-challenge-openai-and-anthropic-in-frontier-ai-race-10-12-2026 - [2026-10-14] A comparison newsletter groups Cowork with OpenAI Dots and Gemini Spark as "work agents" for people at a desk, distinct from consumer "ready-made assistants" like Muse and Siri. https://home.makeyourmindup.ai/p/amazon-locked-metas-new-ai-out-walmart - [2026-09-30] A TechDogs comparison says Claude Cowork is oriented to knowledge work and documents, less to personal errands. https://www.techdogs.com/td-articles/trending-stories/best-ai-personal-agents-in-2026-muse-copilot-chatgpt-and-claude-compared BOTTOM LINE: Anthropic already ships cloud-run, scheduled, form-filling, connector-using Cowork to paid plans. Whether it counts as "Muse-like" depends on whether a resolver treats its sessions as per-task environments (excluded) or as a persistent per-user agent. I found no Anthropic announcement of a dedicated persistent per-user VM agent, and no dated launch event before 2027. NOT FOUND: any Anthropic statement or credible report of a planned Muse-style persistent personal agent launch. Anthropic release notes or blog posts for October 2026. Any confirmation of how the question's resolver classifies Cowork.

- [2026-09-08] Meta launched Muse in the US on iOS, Android and muse.ai. It runs on a dedicated per-user cloud VM, keeps working after the app is closed, comes back when something changes, and can open a browser and fill out forms. It is free with paid tiers. Confirmed on Meta's newsroom page. https://about.fb.com/news/2026/09/introducing-muse-personal-ai-agent/ - [2026-09-29] OpenAI launched "dots" at DevDay. Each dot is an always-on agent with its own cloud computer and browser and 4,000+ app connections. It is available to Pro and Business Premium, with one agent per user initially, and was not available in the EU or UK at launch. Reported by Adweek and Trendingtopics. This is the second lab to ship a Muse-like product within about three weeks. https://www.adweek.com/media/openai-takes-on-metas-muse-with-dots-an-always-on-ai-agent/ - [2026-08-11 to 2026-09-29] A comparison piece says xAI (Grok Bot), Meta and OpenAI each launched an always-on agent with its own cloud computer in this window. It says Anthropic "folded its equivalent" into Claude. This is a secondary source, and its own framing is that the other labs shipped first. https://codeatic.com/blog/ai-agents-compared-dots-muse-grok-claude - [2026-10-03] The New Stack (published about 10 days ago) says Anthropic's "answer" is already in Claude. It describes Cowork as running scheduled jobs after a user closes their laptop. Cowork was merged into the main Claude app for Pro and Max, with Team and Free "coming soon". It also says Claude Tag has been in Slack for teams since June. It adds that a feature one lab ships tends to appear at rivals within weeks or months. I could not read the full article text (the fetch returned only the page boilerplate), so this rests on the search snippets. https://thenewstack.io/claude-answer-to-dots-muse/ - [2026-10-06] New Cowork tasks on Pro and Max now run in Anthropic's cloud by default, and the local-only option was removed. Scheduled tasks run in the cloud. Cowork has been available on web and mobile since 2026-07-07. Sources: Anthropic help-center text as quoted by secondary sites, and a Redreamality analysis. https://hackingdemand.com/blog/claude-cowork-cloud-october-6 - [2026-10-06] A critical qualification for resolution. Redreamality describes the new Cowork as giving each session its own sandbox that shares no state with others. The question excludes environments created per task or session. Some local-file tasks also need the desktop app open. Nothing I found says Cowork gets a persistent per-user computer. The resolver could still treat Cowork's cloud scheduling as "Muse-like". Whether it counts is the main ambiguity, and it is an interpretation issue, not a base-rate fact. https://redreamality.com/blog/claude-cowork-cloud-sandbox-where-agents-run/ - [2026-09-23] Claude Code cloud sessions went GA on 2026-09-23. They run in Anthropic-managed VMs and are described as "task-scoped by design", and they are coding-only. They would not qualify under the question's exclusions. https://www.agent37.com/blog/claude-code-cloud - [2026-10-13] Claude Managed Agents launched in April 2026. It is API/developer infrastructure that gives each session a virtual computer. It is an API product, which is excluded. https://every.to/source-code/why-we-handed-our-agent-s-infrastructure-to-anthropic - [2026-10-12] Meta launched the Muse Wonder models and said they will power the subscription Muse agent. The competitive pressure on Anthropic is rising, but the articles I found carry no report of an Anthropic launch plan or date for a dedicated persistent personal agent. https://qz.com/meta-muse-wonder-ai-model-launch-101226 - [Outside view] Cadence, from the dated sources above: Anthropic shipped Cowork in January 2026, scheduled tasks on 2026-02-25, computer use on 2026-03-23, cloud Cowork on 2026-07-07, and the cloud default on 2026-10-06. That is roughly one agent-related step every 1–2 months. Rival launches have tended to be matched within weeks to months (New Stack). A dedicated persistent personal agent from Anthropic by 12/31 would be one more step in this pattern. Anthropic's pattern is to fold features into Claude rather than ship a separately branded agent, so a release in that form would be easy to miss or to judge ambiguously. - [Outside view] Reference class: a lab launching a rival's new product category within about 3.5 months. Here OpenAI matched Meta in 3 weeks and xAI shipped Grok Bot in the same window. Anthropic's most direct analogue, Cowork, already exists and has been moving toward Muse-like behaviour. I could not find base-rate studies, expert estimates or other betting markets for this question. BOTTOM LINE: Competitors have matched Muse quickly, and Anthropic's Cowork is converging on a cloud-run, scheduled, connected-account agent. The outcome likely turns on whether the resolver counts Cowork, which appears to use per-session sandboxes and has no persistent dedicated computer, or whether Anthropic ships something explicitly persistent by 12/31. I found no report of an Anthropic launch plan. NOT FOUND: Any Anthropic statement or credible leak of a dedicated always-on personal agent with a persistent per-user cloud computer. Anthropic's primary help-center or newsroom pages on Cowork cloud (I only saw them quoted by secondary sites). Other Polymarket strikes or sibling markets for this event. Expert forecasts or historical base-rate data for this category.

YES means: Anthropic publicly launches, by 2026-12-31, a consumer-accessible agent that has a dedicated per-user cloud computer persisting between tasks, can start tasks on its own, and can act on websites or connected accounts. Alternatively, the resolver accepts an existing or updated Claude/Cowork product as meeting that definition. - [2026-10-14] The Cowork help page, which I fetched, says sessions and files are tied to the user's Claude account. It says scheduled tasks run with no device online and that cloud is the only option on Pro/Max. The page says Cowork "is now just Claude", with chat and Cowork merged. This is the main route to YES. A resolver could read the account-bound, always-available cloud agent as "dedicated and remains assigned to the user". A small change to the docs, such as a named per-user persistent workspace, would remove the per-session objection. https://support.claude.com/en/articles/15520349-use-claude-cowork-on-web-desktop-and-mobile - [2026-10-14] The same page says cloud Cowork runs on Pro and Max, in beta for Team, and in beta for Enterprise where an admin turns it on. The criteria accept paid-only availability. The "not sold only to businesses" clause is already met, because Pro and Max are consumer plans. https://support.claude.com/en/articles/15520349-use-claude-cowork-on-web-desktop-and-mobile - [2026-10-14] Cloud Cowork "projects" have their own files, instructions and memory. They are saved with the member's Claude account, per the Team/Enterprise help page. This is persistent per-user state, though it is not a persistent VM. https://support.claude.com/en/articles/13455879-use-claude-cowork-on-team-and-enterprise-plans - [2026-10-06 to 2026-10-14] Reverse-engineering of the local Cowork architecture reported that its Linux VM disk is re-formatted at every boot. The Cloud help text says sandboxes are created at session start and destroyed at the end. This is the strongest evidence for NO on the current product. Anthropic would need to change the architecture, or the resolver would need to look past it. https://cellcog.ai/blog/claude-cowork-cloud-only/ - [2026-06-23] Claude Tag launched as a beta for Team and Enterprise. It is an "always-on" Slack agent with a persistent identity, service accounts and an ambient mode that acts without being tagged. Anthropic's blog says it evolves Claude Code. It is multi-user and workplace-only. Reuters, via Let's Data Science, says Anthropic planned to expand it to other platforms. It shows Anthropic already has the engineering for an always-on, self-starting agent and could extend it to individuals. https://www.anthropic.com/news/introducing-claude-tag - [2026-10-07 to 2026-10-10] Anthropic's Claude release feed shows a heavy weekly cadence of consumer-surface features. These include Dashboards (a live dashboard that stays current) and Docs/Slides/Design moving to every plan, including Free. Claude Haiku 5.5 launched 2026-10-07 and is aimed at browser and computer use. I found no persistent-agent launch in October so far. https://releasebot.io/updates/anthropic/claude - [2026-10-07 to 2026-10-14] Platform notes show new SDK classes for the browser and computer-use tools. They also show dynamic workflows for Managed Agents (2026-10-09), which run server-side in the background, and OpenTelemetry export for Managed Agents sessions (2026-10-14). The Managed Agents API is excluded from resolution. Its long-running sessions that persist through disconnections show the infrastructure for a per-user persistent agent is in place. https://platform.claude.com/docs/en/release-notes/overview - [2026-10-03] The New Stack describes always-on agents with their own work identity as a category Anthropic already competes in. It says features spread to rivals within weeks or months. It is commentary, not an Anthropic commitment. https://thenewstack.io/claude-answer-to-dots-muse/ - [2026-09-23] Claude Code cloud sessions went GA, with $100 and $250 credits for Pro and Max. This shows Anthropic will spend to push users toward cloud agents. These are coding-only and probably excluded. https://www.agent37.com/blog/claude-code-cloud - [2026-10-14] The Cowork help page says that on desktop the built-in browser is the default, and a browser-driving task still needs the desktop app open. It also says the Chrome side panel runs as a Cowork session. That means website actions are partly not cloud-only. A fully cloud browser is the missing piece. A fully cloud browser is the specific prerequisite for a clean Muse-like launch. Anthropic is already rolling out the browser changes "this week". https://support.claude.com/en/articles/15520349-use-claude-cowork-on-web-desktop-and-mobile BOTTOM LINE: The most likely path to YES is a lenient reading of the existing cloud Cowork/Claude. It is account-bound, scheduled, runs with no device online and is on paid consumer plans, so one doc change or a cloud-browser update could make it clearly qualify. The second path is Anthropic extending Claude Tag or Managed Agents infrastructure into a consumer always-on agent within about 11 weeks, which would be easier to recognise but has no evidence of being planned. NOT FOUND: Any Anthropic statement, leak or job post about a persistent per-user cloud computer or a Muse-style consumer agent. Any Anthropic announcement dated after 2026-10-07 for a new always-on consumer agent. Any resolver ruling on whether Cowork counts.

YES means: Anthropic launches, by 2026-12-31 11:59 PM ET, a publicly accessible consumer or prosumer AI agent that runs on a cloud computer dedicated to each user and persists between tasks. It must start tasks on its own and act on websites or connected accounts. Per-task or per-session environments, single-task products, API-only products and business-only products do not count. - [2026-10-14] Anthropic's own Cowork architecture page, updated today, is the strongest NO evidence. Cloud Cowork gives each session its own sandbox, created at session start and destroyed at its end, sharing no state with other sandboxes. The criteria exclude exactly that ("created for each task or session"). The sandbox holds only session-scoped tokens that expire within hours. Connector tokens never enter it, and connector calls run server-side. This describes a design that is not a persistent per-user computer. https://support.claude.com/en/articles/14479288-claude-cowork-architecture-overview - [2026-10-14] The same page says that when a cloud session needs local files or the browser, it goes through the Claude Desktop app. If the app is offline, the cloud session cannot reach the device. Browser-based form-filling therefore depends on the user's own machine being on, unlike Muse's per-user VM. It also says the unified chat-and-Cowork experience is rolling out gradually on Pro and Max. Team keeps the old split, and Enterprise is beta only. https://support.claude.com/en/articles/14479288-claude-cowork-architecture-overview - [2026-10-14, Enterprise unified-Claude help page, ~6 days old] Under the unified experience, each conversation that needs a sandbox gets its own Anthropic-managed VM. The VM stops after a few minutes idle, and its disk is saved as an encrypted snapshot. The sandbox expires 30 days after creation, and the conversation then continues in a fresh environment. That is a per-conversation environment, not a standing per-user computer. A lenient resolver could call the snapshot a persistent environment, but the 30-day expiry and per-conversation scoping cut against it. Scheduled-task runs "don't belong to a conversation." https://support.claude.com/en/articles/17444462-set-up-and-manage-the-new-unified-claude-experience-for-your-enterprise-organization - [2026-10-14] Anthropic's Help Center release notes (fetched; the page itself says it was updated over a week ago) list no personal-agent launch in October. The only October entry is Haiku 5.5 and API credits on 10-07. September shows Cowork merged into chat on 09-16, plus model launches and plugins. The 08-25 entry is cross-session memory, not a persistent computer. There is no "always-on agent" or "dedicated computer" entry in Aug–Oct. Anthropic's recent shipping is model releases (Opus 5.5, Sonnet 5.5, Haiku 5.5) and folding agent features into Claude. https://support.claude.com/en/articles/12138966-release-notes - [2026-10-10] The Releasebot trackers for Claude, Claude Code and Claude Platform list October items: Dashboards/Motion, Google Workspace, mods, Managed Agents workflows and Claude Tag fixes. None is a Muse-style agent. The agent infrastructure improvements (Managed Agents dynamic workflows on 10-09, OTel on 10-14) are API-only and excluded. https://releasebot.io/updates/anthropic/claude - [2026-10-08] Agent37 explains that Anthropic's cloud sessions are "task-scoped by design" and have no shell. It sells a separate "persistent instance" as the always-on alternative, which implies Anthropic does not offer one itself. The routines it describes are coding-only. https://www.agent37.com/blog/claude-code-cloud - [2026-10-03] The New Stack (via the daily.dev summary, 2026-10-09) frames Anthropic's strategy as shipping to developers and paying subscribers inside the existing app. It says Anthropic has not put Claude where Muse lives, such as WhatsApp. Claude still asks you to open the app or tag it in Slack. It also says Claude Tag in Slack has had its own service identity since June. That is a team tool, which is business-oriented. The positioning is that Cowork is the answer, not that a new consumer product is coming. https://thenewstack.io/claude-answer-to-dots-muse/ - [2026-03-23] Dispatch, the earlier "persistent" Claude feature, ran on the user's own computer, and the host machine had to stay running (Tessl, March 2026). Anthropic's pattern is local-first, with cloud added gradually. The 2026-07-07 and 2026-10-06 cloud moves, per the colleagues' reports, are the first steps away from that. That leaves about 11 weeks, and no leak, waitlist or teaser exists. https://tessl.io/blog/how-anthropic-is-turning-claude-into-an-always-on-agent-and-what-it-learned-from-openclaw - [2026-10-14] Deadline arithmetic: it is 78 days to 12/31, which includes the Thanksgiving and December holiday slowdown. Anthropic's calendar is already busy, with the 5.5 model family rolling out through October, a company IPO reportedly in progress (Wikipedia, 2026-10), and the unified-Claude rollout still unfinished for Team and Free. It is a plausible choice to finish that rollout rather than launch a separate agent product. https://en.wikipedia.org/wiki/Dario_Amodei - [2026-10-14] Technicality: even if Anthropic ships something larger, it must be "launched and publicly accessible" and be personal rather than business-only. Cowork-on-Team and Enterprise rollouts are business-oriented. The resolver, per the criteria, uses "official information from the listed company". Anthropic's official pages call cloud sessions isolated and temporary, so the official text points to exclusion unless Anthropic changes it. https://support.claude.com/en/articles/14479288-claude-cowork-architecture-overview BOTTOM LINE: On Anthropic's own pages, Cowork is built from per-session, destroyed-at-end sandboxes, which the criteria exclude. No sign of a persistent per-user agent is visible in the release notes or trackers, so NO rests on status quo unless Anthropic announces something new within 78 days. The main risk to NO is a lenient resolver counting the unified Claude (snapshotted per-conversation VMs, cloud schedules, memory) as Muse-like. NOT FOUND: any Anthropic statement, leak, waitlist or job posting for a dedicated always-on per-user cloud computer. Any Anthropic comment on whether Cowork counts as "Muse-like". Any October release-note entry showing a persistent agent. Other venues' or sibling markets for this event.