Anthropic Opens Research Preview of Claude Duet, a Proactive AI Agent That Works While You Step Away

Anthropic has opened a research preview of Claude Duet, a standalone application in which Claude acts as a proactive agent instead of waiting for the user to ask for help. The company opened a waitlist on October 21, 2026, initially limiting access to Pro and Max subscribers in the United States. The launch places Anthropic in a fast-moving race among frontier AI labs to build agents that keep working between conversations rather than simply answering one prompt at a time.

What Duet does differently

Most AI assistants sit idle until prompted. Duet is designed to come to the user first. Each morning it can assemble a brief covering the items it believes deserve attention, then continue monitoring connected services throughout the day and message the user when something needs a decision. Delegated work continues in the background: the agent runs on its own virtual machine with its own browser, so tasks keep progressing even after the owner closes a laptop or switches off a phone, and the user is notified when work finishes or needs review.

Anthropic pitches the app mainly at professionals and small-business owners who have more work on their plates than they can track across every app. Example jobs named by the company include flagging what needs attention in the inbox and calendar, drafting replies for approval, setting up meetings, building dashboards that pull data from many tools, preparing financial or sales forecasts, and prototyping or building full web apps.

Key details of the preview

Duet is available through a standalone app for iOS and Android and on the web at claude.ai, and it is powered by Claude Opus 5.5. It works with more than 6,000 connectors and plugins from the Claude Marketplace, including services such as Gmail, Slack, Google Calendar, Google Drive, Intuit QuickBooks, Stripe, Shopify, HubSpot, Gusto, Docusign, Notion, Canva, and Microsoft 365. Setup is meant to be lightweight: Duet carries over a subscriber’s stored memory, skills, and app connections from their existing Claude account.

The agent checks connected accounts every few hours rather than continuously. Anthropic says that choice is deliberate, because constant checking would consume a plan’s usage limits much faster. Background checks and larger tasks draw on the same usage allowance as the rest of a subscriber’s Claude plan, and bigger jobs, such as building a financial model or a presentation, can burn through those limits faster because they require more model turns. Team and Enterprise plans are not included in the preview.

Approval-first guardrails

Anthropic is giving Duet considerable autonomy but keeps several clear limits in place. Before Duet emails or messages anyone, makes a payment, signs a document, or deletes data, it lays out the planned step and waits for the user’s go-ahead. Trading securities and transferring funds between accounts are barred outright, approval or not. Anthropic also warns that Duet can misread a thread, misjudge the urgency of an item, or miss an update that arrives between check-ins, and that a drafted action still deserves human review before it affects customers, colleagues, or money.

Why it matters

Duet is part of a broader industry shift in which the unit of competition is moving from raw model capability to agents that get things done across applications. Meta pushed the idea into the spotlight with Muse, OpenAI followed with Dots, and Anthropic is now entering with a product aimed squarely at workplace and small-business tasks rather than personal errands. The real test of Duet will be trust: whether users believe the agent’s judgment enough to delegate recurring work, accept a morning brief as a starting point rather than a complete audit, and keep approval switched on for consequential actions. For now, it is an early, waitlisted preview with useful ideas and stated safeguards, plus acknowledged limits that make human oversight non-negotiable.

Sources

• Digital Trends, “Anthropic unveils Claude Duet, its answer to AI agents like Dots”: https://www.digitaltrends.com/computing/anthropic-unveils-claude-duet-ai-agent/
• ITMunch, “Claude Duet: What Anthropic’s New Proactive AI Agent Can and Can’t Do”: https://itmunch.com/claude-duet-anthropic-proactive-ai-agent/
• BetaNews, “Claude Duet: Anthropic opens proactive AI agent preview”: https://betanews.com/article/proactive-ai-agent-claude-duet/
• AI NEWS roundup, “Week of October 19 to 26, 2026”: https://medium.com/@davidakpovi/ai-news-week-of-october-19-to-26-2026-8adc02ca81c0