Claude Duet Debuts as Anthropic’s Always-On Proactive Agent

Anthropic has officially introduced Claude Duet as an early test version called a research preview. This new service is a big change for Claude. Instead of only answering questions when users ask, Claude can now work on its own and take action without waiting for instructions each time.

Claude Duet does not need to stay open in a browser tab. It runs on its own virtual computer in the cloud, which means it uses a computer on the internet instead of the user's device. This allows it to keep working in the background, even when users close the app or turn off their laptops.

How Claude Duet Works in the Background

Claude Duet can check connected email inboxes, calendars, and project management tools while users sleep. Each morning, it can show users a list of important tasks and what they should do next.

The assistant does more than simply send notifications. It can write draft follow-up messages, find meetings or events that overlap, and point out urgent messages. Users can check these updates on the web or through special Claude Duet apps for iPhones and Android phones.

Claude Duet can also handle more complicated jobs that involve several steps. It does this using its own virtual computer in the cloud. For example, it can find unpaid bills, collect information for research, create dashboards that display useful data, and organize project files. Users do not need to watch over it all the time.

If a website does not offer a direct API connection, which is a way for different services to share information, Claude Duet can use its built-in cloud browser instead. This lets the assistant move through websites and online services and collect the information it needs on its own.

Built-in Safety Rules and User Approval

To help keep connected accounts and services safe, Anthropic built Claude Duet with strict rules that keep people in control. The assistant automatically stops and asks users for clear permission before doing anything sensitive.

These actions include sending emails to other people, making payments through Stripe or Shop Pay, signing documents, and deleting records. A special Tasks tab lets users see which jobs are still running. Users must also give the final approval when required before tasks can be completed.

Claude Duet uses Anthropic's main AI reasoning model, Claude Opus 5.5, to power its work. It also keeps the user's existing saved information, skills it has learned, and OAuth connections, which let it securely connect to other services. This means users can continue using their previous information and settings without entering everything again.

Availability and Rollout Details

Claude Duet is currently being released slowly as an experimental research preview. People who want to try it must join a waitlist. At first, only individual Claude Pro and Claude Max subscribers in the United States can get access.

Anthropic confirmed that tasks running in the background count toward the usage limits already included in a subscriber's plan. Users do not have to pay an extra fee for this feature. Anthropic also plans to make Claude Duet available to Team and Enterprise subscribers in a later stage of the rollout.