Anthropic has opened a research preview of Claude Duet, a background agent that keeps working on a user's connected accounts after they step away and checks in before taking any consequential step. The product is aimed at US subscribers on the Pro and Max plans, who can join a waitlist and use it on iOS, Android or claude.ai.
What Duet Does
Duet runs on Claude Opus 5.5 inside its own virtual environment, which includes a browser. That design means a user's phone or laptop does not need to stay switched on for work to continue. According to MindStudio's guide, the agent connects mainly to email and calendar. Each morning it reviews incoming messages, flags items that need attention and drafts replies for the user to review. Throughout the day it tracks open threads and follows up on conversations that stall. Users can interact with it by typing or speaking.
Daily.dev's summary adds that Duet checks connected apps every few hours, sends morning summaries and supports more than 6,000 integrations, including Gmail, Slack, Stripe and Shopify. Those integration figures come from secondary coverage, and the independent Claude Cowork guide repeats the 6,000-plus connector count without citing an official Anthropic announcement. Readers should treat the number as reported rather than confirmed.
The tool is a meaningful step beyond a chat window. Instead of waiting for a prompt, it is designed to notice what needs attention and prepare work in advance, which is the core promise of the agentic AI category.
Access and Pricing
Duet is limited to US users on paid plans. The free tier does not include it, and the Claude Cowork guide reports that Team and Enterprise plans are not included yet. Access runs through a waitlist, and there is no separate fee. Duet draws on the same usage limits as the rest of Claude, and its background checks count toward those limits.
According to daily.dev's summary, Duet carries over memory, skills and app connections from an existing Claude account, so setup is light. The product is available on iOS, Android and the web.
The Approval Model
The most important design choice is the approval gate. Duet asks for explicit permission before it sends messages, makes payments, signs documents or deletes files, and it shows the planned action first. Drafting and flagging happen without approval. Daily.dev reports that trading securities and transferring money between accounts are blocked outright.
The Claude Cowork guide also describes a sensitive-information classifier meant to stop leaks of items such as card numbers and government ID numbers. It advises users to grant the narrowest access that gets the job done. That advice reflects a practical reality: a background agent acts when no one is watching. A draft sitting in an inbox costs little, but an email sent to the wrong client or a payment approved in error can be expensive to undo.
Limits and Open Questions
Duet is still a limited test. MindStudio's guide notes that it is in research preview, that it is available only to US users and that access is granted from a waitlist rather than instantly. The guide gives no timeline for leaving preview or for expanding internationally.
The sources reviewed for this article do not include independent performance figures, such as the accuracy of morning summaries or how often drafts need heavy editing. Claims about how well Duet works in daily life remain untested by outside measurement. Anyone evaluating it should run a short trial with limited permissions before granting wider access.
Why It Matters
Duet is part of a shift from assistants that respond when asked to agents that monitor, prioritise and prepare work on a schedule. It also illustrates the trade-off this category now faces. The more an agent can do unattended, the more its usefulness depends on permission design, narrow access and clear records of what it did.
The Claude Cowork guide says Claude Cowork merged into the main Claude experience in September 2026, alongside Docs, Slides and Design. That consolidation suggests Duet is being positioned as part of a wider Claude workspace rather than a standalone experiment. Because the guide is an independent site and not an Anthropic source, that detail should be checked against official announcements.
For businesses and power users, the practical question is not whether background agents will arrive, but how to govern them. Duet's approach, which lets agents draft and flag freely while requiring sign-off for anything irreversible, is likely to become a common template across the industry.
What to Watch
- Whether Anthropic adds Team and Enterprise plans to the preview.
- When Duet leaves research preview and whether it expands beyond the US.
- Whether independent tests confirm the connector count and the reliability of morning briefings.
- How rival agent products adopt similar approval gates and access limits.
