Anthropic Launches Claude Duet, a Proactive AI Agent for Work

Anthropic has introduced Claude Duet, a standalone AI agent designed to watch connected work apps, identify what needs attention and continue tasks while its user is away. The research preview is starting with a waitlist for Pro and Max subscribers in the United States, not a general release for every Claude user.

In its October 21 announcement, Anthropic said Duet can prepare a morning brief from connected accounts and send further messages when something important arises. Users can also assign work by text or voice in an ongoing thread. The product is available as a separate iOS and Android app and on claude.ai for those admitted to the preview.

Duet runs on Claude Opus 5.5 and has a virtual machine with its own browser. That setup lets it continue a task without keeping a phone or laptop switched on. Anthropic describes uses ranging from scheduling and briefing to building dashboards, financial forecasts, prototypes and web apps. It can draw on a user’s Claude memory, skills and connected services rather than starting with a blank account.

The company says Duet can work with connectors in the Claude Marketplace. Its examples include finding missed Gmail and Slack messages, checking unpaid invoices in QuickBooks or Stripe, drafting job posts and preparing documents for signature. These are described as capabilities in an early research preview, not proof that every integration is equally reliable or available to every customer.

Duet is also more selective about interrupting users than a constant notification feed. Anthropic says it checks connected accounts every few hours, partly because continuous checking would consume plan capacity faster. Users can tell it which updates they want and how often. The background checks and larger tasks count against the same usage limits as the rest of a Pro or Max subscription.

A proactive agent that reads messages and prepares transactions raises a practical question; when can it act without its owner? Anthropic says Duet asks for approval before sending a message, making a payment, signing a document or deleting anything. Some activities, including securities trading and moving money between accounts, are off-limits even with approval. Business purchases through Link or Shop Pay involve a further confirmation step.

Anthropic says credentials are handled so the agent does not directly see passwords or sign-in tokens. It also describes safeguards against prompt injection and accidental disclosure of sensitive information. Those are company claims about the preview’s design, not a guarantee that an agent operating across email, files and commerce cannot make mistakes. Anthropic itself says Duet may misread a conversation or misjudge what is urgent.

The launch gives a public form to the industry’s move towards assistants that start work instead of waiting for a prompt. TechBooky previously covered indications of a proactive mode in Claude Code, though that earlier finding was not a product announcement. Duet is a distinct, officially unveiled app aimed at professionals and small businesses.

For now, the most important limits are availability and trust. A US-only waitlist means readers elsewhere cannot assume immediate access, and the shared usage allowance could make heavy background work costly in practice. The stronger test will come when external users can judge whether Duet catches genuinely important work, respects approval boundaries and saves more time than it consumes in supervision.