Anthropic opens Claude Duet preview for AI work that continues between chats
The assistant checks connected accounts and prepares work without a fresh prompt. Its background activity also consumes subscribers’ existing Claude usage limits.
Loading page…
The assistant checks connected accounts and prepares work without a fresh prompt. Its background activity also consumes subscribers’ existing Claude usage limits.
Listen to this story
The preview turns Claude into a persistent work agent: Duet can use its own cloud computer and browser to monitor connected accounts, prepare files, and resume tasks, while requiring approval for consequential actions. It is included for US Pro and Max subscribers, but invitations roll out in waves, Team and Enterprise are excluded, and background work uses existing five-hour and weekly allowances. Its usefulness will depend on trust and quota management as well as task execution; checks every few hours and no browser handoff limit responsiveness and user oversight.
Duet runs on Claude Opus 5.5 with no model selector during the preview, and inherits account memory, skills, and connected apps.
Link and Shop Pay purchases require approval and use one-time cards limited to the approved amount; Link also requires a US payment method.
Disconnecting an app stops future access but does not erase information remembered earlier; users can separately turn off searches of past chats.
Claude can now start work without waiting for another question. Anthropic opened a waitlist on October 21 for Claude Duet, a research-preview assistant for US Pro and Max subscribers. The company says it checks connected apps, prepares work and keeps going while users are away—but asks before sending messages, paying, signing documents or deleting information.
The preview is included in Pro and Max subscriptions, with access through standalone iOS and Android apps and claude.ai. Invitations will arrive in waves; joining the waitlist does not secure an access date. Team and Enterprise plans are not eligible yet. Anthropic is targeting professionals and small-business owners, with jobs ranging from meeting preparation to invoice reminders and financial forecasts.
Duet inherits memory, skills and connected apps from a user’s Claude account. People can send typed or voice messages through an ongoing thread instead of restating the context each time. It runs on Claude Opus 5.5, with no model selector during the preview. Its own cloud computer and browser let it continue tasks even when the user’s phone or laptop is off.
Anthropic says Duet sends a morning brief drawn from connected apps, then checks those accounts every few hours. It can draft replies and follow up on quiet threads. Larger assignments can return as editable files, including research briefs, financial models, slide decks and dashboards. A Tasks tab separates ongoing jobs from the main conversation, highlights work needing input, and lets only the user mark a task complete.
The company’s early examples emphasize unsolicited preparation, not just faster answers. In Anthropic’s announcement, brand strategist Mya Dalton said Duet prepared a brief for a potential-client call, noticed it fell shortly before a flight and offered to reschedule it. That is a company-selected tester account, rather than an independent measure of how reliably Duet spots conflicts.
Reading, drafting and preparing work can happen without another prompt. Anthropic says Duet shows the proposed action and waits for approval before sending a message, making a payment, signing a document or deleting anything. If approval has to wait, it can finish the rest of the task and return to the request later. Securities trading and transfers between accounts remain prohibited even with permission.
Purchases use Stripe’s Link wallet or Shopify’s Shop Pay, connected inside Duet rather than through its connector directory. These payment tools are separate from the Stripe and Shopify connectors used to work with business accounts. The two checkout routes have distinct controls:
Approval is only one safeguard. Anthropic says Duet keeps passwords and sign-in tokens outside the model’s direct view. A new classifier—a screening system—and safety instructions are designed to block sharing credit-card numbers and government identification. Sending health or personal-finance information requires an explicit instruction. The company also warns that emails and webpages can contain malicious instructions, known as prompt injection, and that no defense is perfect.
There is no extra preview fee, but there is no separate pool of usage either. Duet draws from the same five-hour session and weekly limits as ordinary Claude chats. Morning briefs, account checks and other background work count toward those limits, so subscribers may exhaust their allowance sooner. Anthropic says checking accounts every few hours, rather than continuously, helps contain that consumption.
That schedule also means an urgent message can arrive between checks. And users cannot watch Duet’s browser or take it over during the preview. Anthropic acknowledges it may misread a thread or misjudge urgency, and says personal errands such as ticket booking are less developed than its work-focused tasks.
Account controls have another important distinction: disconnecting an app stops future access, but does not erase information remembered from earlier conversations. Users can separately turn off the setting that lets Duet search past chats. Microsoft 365 is not supported yet, and Duet cannot make phone calls or send text messages to other people.
Loading discussion...
Join the conversation
Explain what background help would earn its share of your allowance.
Be the first to share a perspective or an experience.
Reader comments
Newest comments first. Replies stay oldest first.