🔍 Read the full analysis: The Switchboard Takes Center Stage In Claude Duet’s AI Agent Race on ThorstenMeyerAI.com
Prime made for students and young adults
- Fast, free delivery for dorm and study essentials
- Prime Video and Amazon Music included
- Member-only deals
TL;DR
Anthropic launched Claude Duet as a waitlisted research preview on October 21, 2026, for U.S. Pro and Max subscribers. The service can work across connected apps on a cloud computer, but Anthropic says sensitive actions require approval and some financial activities are blocked. The wider agent race also puts attention on who controls app access, memory and the choice of model behind an assistant.
Anthropic launched Claude Duet on October 21 as a waitlisted research preview for U.S. Pro and Max subscribers, bringing its Claude assistant into the race to handle work across apps from a cloud computer. The company says Duet can prepare work and monitor connected accounts, while requiring user approval for sensitive actions; independent testing has not been reported in the source material.
Duet is a separate app available through iOS, Android and claude.ai. Anthropic says it uses Claude Opus 5.5 and carries over the memory, skills and connected apps associated with a user’s Claude account. It runs tasks on a virtual machine with its own browser, allowing work to continue while a user’s phone or computer is offline. Anthropic says Duet can draw on more than 6,000 connectors and plugins in the Claude Marketplace.
The company describes a service that sends a morning brief, accepts tasks by text or voice, and messages the user when attention is needed. Its examples include tracking unpaid invoices through QuickBooks or Stripe, working with payroll in Gusto, handling agreements through DocuSign, and assisting with Shopify storefront tasks. Anthropic also says purchases use Stripe’s Link wallet with a one-time card limited to a specified merchant and amount.
In a safety post dated October 23, Anthropic said Duet must show a user the proposed action, destination and content before it sends a potentially sensitive message, pays, or deletes something. A separate check is intended to assess whether the action matches the user’s request. Anthropic says trading securities, moving money between accounts and accessing another person’s account are blocked, even with user approval. These are company descriptions, not independently verified test results.
Infographic 01 · The permission ladder
What Claude Duet does alone, what it asks first, what it never does.
Anthropic’s three tiers for an agent that keeps working while you are away. Health or financial details go out only with your explicit consent.
Does it alone
no approval
- Reads connected apps: mail, calendar, Slack, invoices
- Sends a morning brief, then flags what needs you
- Drafts replies, reminders, decks, dashboards
- Prepares payroll or contracts for your review
- Checks accounts every few hours, not nonstop
Asks you first
approval card shows action, destination, content, recipients
- Sending a potentially sensitive message
- Paying (one-time card via Link, limited to merchant and amount)
- Signing a document
- Deleting anything
- Never on “don’t ask me again”: paying or sending money, deleting an order, creating an account
Never, even if you approve
blocked
- Trading securities
- Moving money between accounts
- Accessing someone else’s account
- Sharing sign-in credentials, full card or bank numbers, government ID numbers
- Seeing your passwords, cookies or sign-in tokens (kept off the model’s machines)
“Approval fatigue is a serious issue, because too many requests teach people to tap through them.”Anthropic, Claude Duet safety post
These are Anthropic’s stated rules for the research preview. No independent test has been published yet.
Sources: Anthropic, Introducing Claude Duet (21 Oct 2026); Anthropic, Duet safety post (23 Oct 2026); Claude Duet FAQ.
Powered by Thorsten Meyer AI
Control Moves Beyond the Model
Duet’s launch adds another major technology company to a contest over assistants that act beyond a chat window. The immediate question for users is whether a cloud agent can reliably handle routine work without making consequential decisions on its own. The answer depends not only on the model, but also on the scope of its app permissions, approval process and persistent memory.
The source article frames the “switchboard” — the layer controlling access to inboxes and other apps, retaining user context, and selecting which model handles a task — as a potential point of leverage. That is an interpretation rather than a confirmed market outcome. But the concern is concrete: if an assistant routes work across services, users and businesses need to know which systems can access their information and who decides where a task is processed. For companies evaluating agents, that makes permission boundaries and an exit plan as relevant as the assistant’s task performance.
Anthropic’s stated safeguards also do not remove the human-factors problem. The company itself warns that approval fatigue can lead people to approve requests without careful review. A system that pauses for permission may still create risk if prompts are frequent, unclear or easy to dismiss.
Popular right now

Shark Pet Cordless Stick Vacuum, 40-Minute Runtime, Lightweight, IX141

Dyson V8 Cordless Vacuum

Ninja Foodi DZ401 DualZone XL Air Fryer, 10 Qt, 6-in-1, Grey
As an affiliate, we earn on qualifying purchases.
A Crowded Agent Launch Calendar
Claude Duet arrived amid a succession of agent announcements described in the source article: SpaceXAI opened a Grok Bot beta on August 11, Meta launched Muse on September 8, OpenAI presented Dots at DevDay on September 29, and Google released Gemini Agent on October 8, according to TechTarget as cited by the article. Anthropic announced Duet on October 21. These dates show a fast-moving rollout, but do not establish that the services have equivalent features or performance.
The products share a broad proposition: an assistant can continue working through a cloud environment, retain context and return with a result or a request for input. Their capabilities, availability and safeguards differ, and the supplied material does not provide independent comparisons. The source also reports that Grok Bot may route tasks to models from other companies, citing an October 6 post by Elon Musk and an xAI security FAQ. The FAQ wording supplied says the team model allowlist does not govern Grok Bot and that customers do not have a model picker. That example makes model routing part of the control question, though it does not establish how all agent products choose models.
“for professionals and small business owners with more on their plates than they can track”
— Anthropic, in its description of Claude Duet
Early Limits and Open Questions
Duet remains a waitlisted research preview, and the supplied material includes no independent evaluation of its accuracy, security or real-world performance. Anthropic says the early service may misread message threads or judge urgency incorrectly. It also says personal errands such as booking tickets are weaker, the service cannot yet show users its browser, and it checks connected accounts only every few hours to conserve plan usage. Those checks could miss developments between scans.
Other practical details remain unclear, including how often users will encounter approval prompts, how well the separate action check performs, and how Duet behaves across different connectors. Anthropic says background checks count against usage limits shared with the rest of Claude, while large tasks may require more model turns. Disconnecting an app stops new access, but information Duet has already learned remains until its memory is reset, according to the company.
The source material does not establish a launch date for broader availability or provide comparable, independently tested results for Duet and rival agents. It also does not show how the market will settle around model routing, customer choice or data handling. Those questions should remain open rather than being treated as settled by product announcements.
Testing Before Wider Adoption
Anthropic’s next steps for Duet beyond the current waitlisted preview are not specified in the supplied material. Users who gain access can begin with low-stakes tasks, connect only the apps they need, and keep approval requirements in place for actions involving messages, payments or deletion. They should also review what remains in memory after disconnecting an app.
Businesses considering agent tools can run small, documented pilots with read-mostly permissions and a defined way to end the trial and remove access. They can record which tasks the system handles, what errors occur, how many approvals it requests and which services or models process the work. Those observations would help distinguish a useful workflow from a promising product description. Further assessment will depend on Duet’s performance as more users enter the preview and on details Anthropic releases about availability, controls and limits.
Key Questions
What is Claude Duet?
Claude Duet is a separate Claude app that Anthropic says can perform tasks across connected services on a cloud computer and send users briefs or updates. It launched as a waitlisted research preview on October 21, 2026.
Who can use the preview?
According to the supplied source, access is limited to U.S. Pro and Max subscribers through a waitlist. The material does not give a date for general availability.
Can Duet send messages or make payments without approval?
Anthropic says Duet must request approval before sending potentially sensitive messages, paying or deleting items. The company says some actions, including trading securities and moving money between accounts, are blocked even if a user approves them.
What are Duet’s stated limitations?
Anthropic says Duet may misread threads or misjudge urgency, is weaker at personal errands such as booking tickets, cannot yet show its browser, and checks connected accounts every few hours. The company says these checks use shared Claude plan limits.
Has Duet been independently tested?
No independent test is included in the supplied source material. The product features, safeguards and limitations described here are attributed to Anthropic.
Source: ThorstenMeyerAI.com
