Claude Duet: Anthropic’s New A.I. Agent Takes On Dots, Muse and Grok Bot

Anthropic now has an assistant that can work all day and night, too. The company behind Claude has just introduced Claude Duet, an AI agent that works in the background around the clock. It checks email and calendars and only contacts users when it needs something from them. With Duet, Anthropic is catching up with OpenAI, Meta and SpaceXSI, which have already launched AI agents that can keep working without a person watching them. Unlike its rivals, however, Duet is mainly designed for work rather than fun and personal activities.

Who Can Use Duet

Duet is launching as a research preview, which means it is an early version being tested. For now, people must join a waiting list to get access. It is available only to Pro and Max subscribers in the United States. Team and Enterprise customers are "not included yet," Anthropic says on its website. The company has not said when they might get access. People in Europe cannot use it yet, as is also the case with many other new AI agents.

Duet works as a separate app on iOS and Android phones and can also be used in a web browser at claude.ai. Anthropic does not charge extra for Duet. However, the agent uses up part of the allowance included in a subscriber's plan, even when it is doing background work that the user cannot see. It runs on Claude Opus 5.5. Reports say the project was called Claude Loop inside the company.

What Duet Is Supposed to Do

Duet works much like its rivals. It uses its own computer in the cloud, so it can keep working even after a user's phone and laptop have been turned off. Its main features include:

• Morning brief: Every morning, Duet gives users a summary of what is coming up. After that, it only contacts them when something needs their attention.
• Big tasks: The agent is designed to arrange meetings, create dashboards and prepare presentations, documents and spreadsheets for users to review.
• Business focus: Anthropic gives examples such as finding unpaid invoices and writing payment reminders, preparing payroll and getting contracts ready to send.
• Thousands of integrations: Duet can connect to more than 6,000 tools and add-ons. These include Gmail, Slack, Microsoft 365, HubSpot, Stripe, QuickBooks, Docusign and Shopify.
• Memory: Duet automatically brings over memories, skills and connected apps from the user's Claude account. Users manage everything in one chat, using text or voice.
• Approvals: A person must approve actions before Duet sends messages, makes payments, signs documents or deletes data. Reports say that trading securities and moving money between accounts are completely blocked, even if a user approves.

Anthropic's Answer to the Competition

Anthropic has been under pressure to launch its own agent for some time. OpenAI has introduced Dots, an agent that also works around the clock on its own computer in the cloud and can contact users on its own. For example, it can warn them when an email is about to be sent with the wrong attachment. Meta has had success with Muse: The app received more than 10 million downloads in the United States in just 40 days. Elon Musk's AI company SpaceXSI has also offered Grok Bot since August. In an awkward twist, Grok Bot now uses Claude Opus 5.5 itself.

Meta designed Muse for consumers who want help with things like booking trips or shopping. Duet, on the other hand, fits Anthropic's business strategy. Most of the company's money comes from business customers. According to its prospectus for its planned initial public offering (IPO), just two large customers made up about 20 percent of its revenue in the first half of the year, as we reported. Duet is therefore built around tasks such as handling invoices, preparing payroll and managing contracts. However, its first version is available through the Pro and Max plans, so it is aimed more at freelancers and small businesses than at large companies. Reports say it is less developed at personal tasks, such as booking tickets, than at business tasks.

What Problems Duet Could Cause

An AI agent that can access email, financial records and payment systems can create risks. Anthropic itself warns that Duet may misunderstand an email conversation or get the level of urgency wrong. It may also miss something that arrives between checks because it only checks connected services every few hours. The technology publication AlphaSignal reports that early testers saw problems such as a property viewing being added on the wrong day and a client call being scheduled shortly before a flight.

There are other problems that still need to be addressed:

• Security: Giving an AI agent access to email and payment systems creates a risk of prompt injection. This happens when hidden instructions on websites or in messages try to trick an AI into doing something. Anthropic says it uses separate login details, filters to block actions that users have not asked for, and a program that rewards people for finding security bugs.
• Overeager agents: Just a few days ago, Anthropic revealed in its own report that Claude models had sometimes tried to get around obstacles instead of stopping during internal tests. For example, they submitted forms on real websites.
• Little visibility: Users cannot watch what the agent is doing in its cloud browser as it happens.
• Usage limits: Pro subscribers, in particular, may reach their usage limits more quickly. Regular checks and large tasks use computing power throughout the day.