Anthropic’s Claude Duet Is an AI Agent That Works in the Background While You’re Away

Image: Anthropic

Meta description: Anthropic’s Claude Duet is a proactive AI agent that can monitor connected apps, prepare briefings and complete delegated tasks in the background. Here’s how it works, what it can do and who can access it.

Artificial intelligence assistants are moving beyond the familiar pattern of answering questions in a chat window. Instead of waiting for users to enter a prompt, a new generation of AI agents is being designed to monitor information, identify tasks that need attention and carry out work with less constant supervision.

Anthropic is taking that approach with Claude Duet, a new AI agent app announced on October 21, 2026. The company describes it as a proactive assistant for professionals and small-business owners who need help managing work across connected applications.

Claude Duet can prepare briefings, keep track of relevant information and continue working on assigned tasks even after a user steps away from their computer or phone. However, it remains an early research preview, with access initially limited to eligible subscribers in the United States.

What is Claude Duet?

Claude Duet is a standalone application that brings Claude into a more proactive, task-oriented workflow. Rather than treating every interaction as an isolated conversation, it is designed to help users manage ongoing responsibilities.

Anthropic says Duet can identify items that need attention, complete tasks users assign and check with them before taking certain actions on their behalf.

The app is available through iOS, Android and the web. During its research preview, access is offered through a waitlist for Claude Pro and Max subscribers in the United States. It shares the usage limits of the user’s existing Claude plan.

That limited availability is important. Claude Duet is not yet a universally available replacement for a traditional digital assistant, and its capabilities may evolve as Anthropic gathers feedback from early users.

How Claude Duet keeps working when you step away

One of the defining features of Claude Duet is its ability to perform work without requiring a user’s personal device to remain active.

Anthropic says Duet operates through a virtual machine with its own browser. A user can describe a task, and the agent can determine which connected applications and steps may be required to complete it. The work continues in the background, with a notification when the task is finished or requires the user’s attention.

This creates a different experience from a conventional chatbot. A user might delegate a task before leaving the office and review the result later, rather than staying in a conversation to guide each step.

Duet can also check connected accounts periodically. Anthropic says it reviews them every few hours rather than continuously, partly to avoid consuming a user’s plan limits too quickly.

From morning briefings to multi-step projects

Anthropic presents Claude Duet as a tool for managing the flow of professional work.

The app can prepare a short morning briefing based on information from connected applications. During the day, it can monitor those accounts and alert the user when something appears to require attention.

Users can also assign specific tasks. Examples described by Anthropic include scheduling meetings, building dashboards, preparing financial or sales forecasts, prototyping ideas and creating web applications.

The important distinction is that these tasks may involve several steps and tools. An agent could, in principle, gather relevant information, organize it and prepare a result rather than simply provide instructions for a person to follow.

That does not mean every task will be completed perfectly. The product is still being tested, and its performance will depend on the task, the connected services and the quality of the information available.

What happens before Claude Duet takes action?

An AI agent that can work across applications needs more than the ability to generate text. It also needs rules governing when it can act and when it must ask for permission.

Anthropic says Claude Duet can read information, draft content and prepare work independently. Before sending a message, making a payment, signing a document or deleting something, it is designed to show the proposed action and wait for approval.

The company also says Duet’s credential-handling design prevents the agent from directly seeing users’ passwords or sign-in tokens. Anthropic has described additional protections intended to prevent sensitive information from being shared improperly.

These safeguards are meaningful, but they should not be confused with a guarantee that an agent will never make a mistake. Anthropic acknowledges that the preview may misinterpret a conversation or incorrectly judge what deserves attention.

The limitations users should understand

Claude Duet is an early product, and several limitations matter for anyone considering it.

First, access is restricted to the US research-preview waitlist for Pro and Max subscribers. Availability in other countries and for other plans should not be assumed.

Second, background activity consumes Claude usage limits. Large tasks may require more processing and model interactions than a short question.

Third, the agent is not equally suited to every kind of task. Anthropic says its initial design is focused on professional work; personal errands such as booking tickets are less developed. The company also says users cannot yet see Duet’s browser while it works.

Finally, connected applications introduce privacy and security considerations. Users should understand what accounts they authorize, what information the agent can access and which actions require approval.

Why Claude Duet matters

Claude Duet reflects a broader change in how AI products are being designed. The next stage of assistant software may not be defined only by smarter answers, but by the ability to manage tasks over time and coordinate work across applications.

That could make AI more useful for people dealing with busy inboxes, meetings, reports and repetitive administrative responsibilities. It could also introduce new questions about reliability, oversight and how much autonomy people are comfortable giving software.

For now, Claude Duet is an early experiment rather than a finished vision of autonomous work. Its progress will depend on whether it can reliably save users time while keeping them informed and in control.

The takeaway: Claude Duet is Anthropic’s move toward a more proactive AI assistant—one that can continue working after a user steps away. Its research preview offers a glimpse of that future, but availability, reliability and user oversight remain central to the experience.