🔍 Read the full analysis: A New Chapter In Always-On AI Agents Begins With Claude Duet on ThorstenMeyerAI.com
Get hardware and tech essentials delivered free with Prime
- Fast, free delivery on millions of items
- Prime Video, Amazon Music and more included
- Member-only deals all year
TL;DR
Anthropic launched Claude Duet on Oct. 21 as a waitlisted research preview for U.S. Claude Pro and Max subscribers. The agent can work in the cloud across connected apps, but Anthropic says sensitive actions require approval and some financial actions are blocked; independent testing is not yet available.
Anthropic launched Claude Duet on October 21 as a waitlisted research preview for U.S. Claude Pro and Max subscribers, adding a cloud-based AI agent that can carry out tasks across connected apps while the user is away. The company says Duet can read information, prepare work and send users updates, but requires approval for sensitive actions such as sending messages, paying and deleting; those safeguards and capabilities have not been independently tested.
Duet runs in a separate app available on iOS, Android and claude.ai. Anthropic says it operates on a virtual computer with its own browser, allowing it to continue working when the user’s phone or laptop is off. Users can give it tasks by text or voice in an ongoing conversation. The company’s examples include scheduling meetings, preparing dashboards, creating sales forecasts and building small web apps.
Anthropic says the agent can send a morning brief based on connected apps and contact users during the day when it needs their attention. It carries over memory and connected apps associated with the user’s Claude account and can access what the company describes as more than 6,000 connectors and plugins in the Claude Marketplace. Examples named by Anthropic include QuickBooks, Stripe, Gusto, DocuSign and Shopify.
In a safety post dated October 23, Anthropic said Duet displays the action, destination and content before it sends a potentially sensitive message, makes a payment or deletes something. The user must approve it. The company says a separate classifier checks whether the proposed action matches the user’s intent. Trading securities, transferring money between accounts and accessing another person’s account are prohibited, even with user approval. Anthropic also says Duet does not see passwords, browser cookies or sign-in tokens.
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 to the Agent Interface
Duet is part of a developing product category in which assistants can carry out tasks in the background rather than respond only to requests in a chat window. The product is designed to handle work across email, finance, scheduling and commerce tools through an agent interface. Its performance will depend on task reliability, the frequency of human review and the visibility of completed actions.
The launch also raises a question about control of the agent “switchboard”: the interface that receives a request, accesses a person’s data and selects which model or service performs the work. The source article reports that SpaceXAI’s Grok Bot may route tasks to models from other companies, including Claude Opus 5.5, and that its model selection cannot be restricted by customers. If agents route work across providers, the service managing access, memory and task flow may shape the user experience as much as the underlying model. The available information does not establish how Duet itself may route tasks to external services.
For organizations, relevant operational considerations include access to company accounts, approval practices, auditability and the ability to stop or reverse actions. Anthropic warns that repeated approval prompts can create approval fatigue, potentially leading users to approve actions without careful review.
Products Worth Considering
Trezor Safe 7 Crypto Hardware Wallet
Ledger Flex + Folio Crypto Hardware Wallet
Logitech MX Master 4 for Mac, Ergonomic Bluetooth Mouse with Haptic Feedback, Ultra-Fast Scrolling, USB-C Charging, for MacOS, iPadOS, Windo
As an affiliate, we earn on qualifying purchases.
A Wave of Background Agents
Claude Duet arrives amid a rapid sequence of agent announcements described in the source material. SpaceXAI opened a Grok Bot beta on August 11, Meta launched Muse on September 8, OpenAI introduced Dots at its September 29 DevDay, and Google released Gemini Agent on October 8, according to the cited account. Anthropic’s October 21 launch followed those releases. The products share a broad pitch: software can act on a user’s behalf in a cloud environment and return with results or a request for attention.
Duet is based on Claude Opus 5.5, according to the source, and is not generally available. Access is limited to U.S. Pro and Max subscribers selected through a waitlist. The source gives no independent product test or comparative performance results, so its descriptions of the agent’s functions and protections should be understood as Anthropic’s statements, rather than verified findings.
Anthropic says connected accounts are checked every few hours, rather than continuously, because more frequent checks would use the user’s plan limits more quickly. Background activity also counts against the same usage limits as other Claude use. The company says users can disconnect an app, but information Duet has already learned remains in memory until that memory is reset.
““Approval fatigue is a serious issue, because too many requests teach people to tap through them.””
— Anthropic, in its Claude Duet safety post
Preview Limits and Open Questions
Duet remains a research preview, and no independent test is included in the source material. Its accuracy, speed, reliability and performance across different connected apps are not established. Anthropic says the early version may misread conversations or misjudge urgency, is weaker at personal errands such as booking tickets, and cannot yet show users its browser while it works.
It is also unclear how often the waitlist will admit users, when access might expand beyond eligible U.S. Pro and Max subscribers, or whether the preview’s features and safeguards will change. The source does not provide detailed terms for business administration, audit logs, data retention across connected services or the extent of any external model routing. Anthropic’s statements that certain actions are blocked and that credentials are not visible have not been independently verified in the material provided.
Disconnecting an app stops future access, according to Anthropic, but information already learned remains until the user resets Duet’s memory. The source does not explain in detail how that stored information is presented, reviewed or removed item by item. Those details may affect how comfortable users and organizations are connecting sensitive accounts.
Testing Access and Safeguards
The next immediate milestone is the ongoing waitlist rollout; Anthropic has not supplied a date for broader availability in the source material. As users enter the preview, testing can assess Duet’s performance on ordinary tasks, how frequently it asks for approval and whether its updates arrive in time to be useful. Independent evaluation is needed to compare its stated controls with its behavior in real use.
Businesses considering a trial can define a narrow set of tasks, use read-mostly access where possible and document which actions require human approval. They can also decide how to revoke app access, reset agent memory and end a pilot before connecting sensitive systems. Individual users can start with low-stakes work, connect only the services they need and keep approval requirements in place for actions that send, pay or delete.
Anthropic’s published limits state that Duet is not a continuously monitoring assistant: checks happen every few hours and consume Claude usage limits. The company has not announced whether it will adjust those limits, add browser visibility or expand the preview, according to the supplied material. The preview’s results may help establish how the time spent reviewing and overseeing delegated tasks compares with the work completed by the agent.
Key Questions
What is Claude Duet?
Claude Duet is Anthropic’s cloud-based AI agent, offered as a waitlisted research preview. The company says it can work across connected apps, prepare tasks and contact users when something needs attention.
Who can use the preview?
According to the source material, access is limited to U.S. Claude Pro and Max subscribers selected through a waitlist. Anthropic has not provided a broader release date in the supplied information.
Can Duet send messages or make payments without approval?
Anthropic says Duet must show the proposed action and wait for approval before sending potentially sensitive messages, paying or deleting items. The company says trading securities and moving money between accounts are blocked altogether.
Does Duet monitor connected accounts continuously?
No. Anthropic says it checks connected accounts every few hours to limit usage. It may miss information that appears between checks, and background activity counts toward the user’s Claude plan limits.
Have Duet’s capabilities and safeguards been independently verified?
The source material reports no independent testing. Descriptions of Duet’s features, approval steps and restrictions are attributed to Anthropic and remain subject to verification as the preview is tested.
Source: ThorstenMeyerAI.com
