Anthropic Unveils Claude Duet, Its AI Agent App for Professionals

Anthropic entered the race for proactive AI assistants on Thursday, unveiling Claude Duet, a standalone AI agent app designed to act on a professional's behalf rather than wait for instructions. According to Reuters, the app is available on iOS, Android and claude.ai through a waitlisted research preview, and is initially open only to US subscribers on Anthropic's Pro and Max plans.

The launch marks a shift in how Anthropic positions Claude. Instead of a chatbot that answers questions, Claude Duet is built to notice what needs attention and take action, with the user approving anything sensitive before it happens, reported by Reuters in its coverage of the announcement.

What Claude Duet actually does

Each morning, Claude Duet sends a short briefing on the most important items across a user's connected apps. It then re-checks those accounts every few hours and reaches out only when something genuinely needs attention, according to WebProNews. Tasks run on a virtual machine with its own browser, so the user's own phone or laptop does not need to stay on while the agent works.

The range of delegated work is deliberately professional. Claude Duet can schedule meetings, build financial forecasts and assemble web apps, according to Reuters. For small-business owners, the examples skew toward the daily grind: finding unpaid invoices and drafting reminder emails, preparing payroll for review, getting contracts ready to send, and putting together briefs, decks or dashboards, reported by AI Weekly.

The connector list is the part that makes this credible as a work tool. Claude Duet plugs into more than 6,000 connectors and plugins in the Claude Marketplace, including Gmail, Slack, Microsoft 365, Stripe, Shopify, Intuit QuickBooks, DocuSign, Google Drive, HubSpot, Gusto and Notion, according to AI Weekly's review of the product page. It can also make purchases through Stripe's Link digital wallet and Shopify's Shop Pay, reported by Reuters.

Approval gates and hard limits

Anthropic has drawn unusually explicit lines around what the agent is allowed to do unsupervised. Before Claude Duet sends a message, makes a payment, signs a document or deletes anything, it shows the user exactly what it is about to do and waits for approval, according to WebProNews. Trading securities and moving money between accounts are off-limits even with approval.

A new classifier is designed to reject requests to share credit card numbers and government identification, and sensitive information such as health or personal financial details goes out only when the user has explicitly told the agent to send it, reported by WebProNews. Anthropic added that sensitive operations such as approvals, logins and safety checks run inside the user's own cloud computer, so private details stay isolated, according to AI Weekly.

The company is also candid about what the preview cannot yet do. Claude Duet can misread a thread, misjudge what is urgent, and miss something that arrives between check-ins, Anthropic acknowledges. It remains less effective at personal tasks such as booking tickets, reported by Reuters. Background monitoring is not free either: the app's background checks draw from the same Pro or Max usage limits as a chat window, according to AI Weekly, and Anthropic says it deliberately does not check connected apps nonstop because that would burn through those limits faster.

The race for the always-on assistant

The timing is no accident. OpenAI launched always-on "dots" agents last month that pursue user goals across apps such as Slack and Microsoft Teams, while Meta's Muse AI agent drew millions of downloads and fueled investor enthusiasm for autonomous AI services, according to Reuters. Anthropic's answer is to differentiate on initiative: the product page's pitch is that Claude Duet comes to the user with what needs attention instead of waiting to be prompted, reported by AI Weekly.

For now the audience is deliberately narrow. Claude Duet is a research preview on a waitlist, limited to US Pro and Max subscribers, with Team and Enterprise plans not yet included, according to AI Weekly. There is no announced date for broader availability. But the launch signals where the agent wars are heading: not smarter chat, but assistants that watch your work while you are away and knock on the door only when a human decision is genuinely required.