The short answer
Anthropic has publicly documented substantial agent capabilities, but that does not establish a launch meeting this contract’s specific definition before 2027. Its September announcement merged Claude chat and Cowork and described tasks continuing after a laptop closes. The verified rules additionally require a cloud computer dedicated to an individual user and retained between tasks, alongside proactive work and action-taking. This review did not establish that architecture in a qualifying consumer offering. The October 12 breaking snapshot showed 10% for Anthropic. Our assessment remains conditional on evidence of all required capabilities and publicly accessible launch, rather than a new agent brand alone.[1][2][3]
Key takeaways
- 01Claude’s existing background-task functionality is capability evidence, but does not independently establish a persistent user-dedicated cloud computer.[2]
- 02Meta’s benchmark product combines persistent goals, an agent’s own environment, and access controlled by the user.[3]
- 03Personal-data access increases the practical importance of safety and reliability.[4]
What Anthropic has actually announced
On September 16, Anthropic said Cowork and chat were becoming one Claude, initially rolling out to Pro and Max plans. Its examples include a recurring weekly report and progress checking from a phone. The company also described a default that asks before actions, with an option to keep working and escalate matters needing closer review. These are concrete product statements; they should be evaluated separately from speculation about an additional personal-agent brand.[2]
Editorial analysis: the critical distinction is persistence of the assigned computer, not merely how long one job runs. A product can execute an extended task and still create a fresh environment for each session. Conversely, a qualifying product could be added to Claude rather than branded as a separate app. Evidence should establish the architecture and functions directly.
Why the Muse comparison is specific
Meta’s September 8 announcement describes Muse as working toward personal goals through its app or WhatsApp, using a dedicated virtual computer with a browser. It presents user-selected access as central to the design. Those documented features provide a useful comparison framework: an agent’s execution environment, continuing context, ability to take actions, and the way a person communicates with it. Meta’s description is a company account, not an independent certification of performance.[3]
Editorial analysis: a chatbot that drafts a shopping list and an agent that manages a recurring household task impose different demands. The latter needs account access, a way to decide when to act, and a way to recover when a website or instruction changes. A launch headline may conceal a limited preview, waitlist, or narrowly supported workflow, so the actual scope should be recorded.
The reliability issue goes beyond model quality
Reuters’ September 8 reporting on Muse described internal concerns about sensitive-data handling and explained that connecting personal apps raises the stakes of errors. That reporting makes a practical point about this category without establishing anything about Anthropic’s launch schedule. The consequences of an incorrect answer are different from those of an incorrect action executed through a person’s email, calendar, or payment access.[4]
Editorial analysis: the strongest evidence for a mature personal agent would include permissions that users understand, visibility into actions, and reliable recovery or cancellation. These are editorial evaluation criteria, not claims that Anthropic lacks them. Competitive pressure can create incentives to ship, while unresolved product decisions can keep a technically capable system from a broad consumer rollout.
How to interpret the market movement
The breaking list supplies movement magnitude, but not direction. This page therefore does not call the change a rise or fall. The official API supplied the detailed definition summarized below. It is stricter than treating any product described as an agent as eligible.[1]
Editorial analysis: evidence should be weighted by proximity to a usable launch. A research result establishes what a model can do under selected conditions. A demo establishes one shown workflow. A release announcement establishes what the company says it offers. Public access and actual documentation establish who can use it and under what restrictions. Those stages matter more than a rumored name or an isolated social post.
Market resolution criteria
By December 31, 2026, at 11:59 p.m. Eastern Time, Anthropic must offer a publicly accessible agent using a company-provided cloud computer dedicated to a user and retained between tasks. It must initiate tasks proactively and act on websites or connected accounts. Per-task or per-session environments, single-purpose agents, API-only products, and business-only offerings are excluded. Open betas and open rolling waitlists count; closed or private access does not. Paid subscriptions and country restrictions are allowed. Anthropic information is primary, corroborated by credible reporting.
Possible outcomes
These are conditional editorial scenarios, not assigned probabilities or trading recommendations.
A qualifying personal-agent expansion
Anthropic adds sufficient persistent, proactive consumer capabilities to Claude or introduces a separate product, with access that meets the contract’s requirements before the deadline.
Useful agents, different product scope
Claude continues to expand work automation, but the available product remains outside the market’s Muse-like definition. Technical capability alone would not resolve the distinction.
Announcement ahead of access
Anthropic announces plans or a limited demonstration before year-end, while usable qualifying access arrives later. The outcome depends on the full release criteria.
What to watch next
- Anthropic documentation of a persistent cloud computer assigned to the individual user between tasks.
- Documented consumer availability, rollout dates, and restrictions rather than a demo alone.
- Evidence of proactive initiation and action-taking across websites or connected accounts in a qualifying public product.
Common questions
Is Claude Cowork already a Muse-like personal agent?
It has documented extended-task features. This review did not establish that it meets every verified contract requirement, especially the persistent user-dedicated cloud computer.[2]
Has Anthropic announced a launch date for a separate Muse competitor?
This review did not verify such a date. The sources here describe existing Claude functionality and the competitor benchmark, not a committed launch schedule.
Does a 10% probability mean Anthropic lacks agent technology?
No. It measures the observed market’s expectation for a defined event and deadline. It is not a technical assessment of Claude’s abilities.[1]
Sources & evidence
Sources reviewed on October 12, 2026. Dated background is identified below. Links support specific claims; they do not endorse this site’s analysis.
- [1] PolymarketMarketWho will launch a Muse-like personal agent before 2027? ↗
Exact outcome rules verified through the official Gamma API, alongside dated displayed prices. AI context excluded.
- [2] AnthropicPrimaryClaude Cowork and chat are now one Claude ↗
September 16, 2026 · Documented existing workflow features, rollout and approval controls; not a promise of a separate personal agent.
- [3] MetaPrimaryIntroducing Muse: The World’s First Personal AI Agent Built for Everyone ↗
September 8, 2026 · Company description of the competitor’s goals, execution environment and user-controlled access.
- [4] Reuters via Investing.comReportingMeta launches AI agent that can access other apps to send emails, make payments ↗
September 8, 2026 · Independent reporting about Muse and personal-data reliability stakes; not evidence of Anthropic’s schedule.
This is an AI-assisted synthesis of linked evidence, with original conditional analysis. It is not original on-the-ground reporting. Read our editorial policy.