Anthropic launches Claude Duet, a proactive Opus 5.5 agent for professionals

Anthropic has announced Claude Duet, a proactive AI agent aimed at professionals and small business owners, released as a research preview. According to The Verge, the launch puts Anthropic in competition with Meta's Muse and OpenAI's Dots, two agents chasing the same idea: an assistant that contacts you first rather than waiting to be opened.

The agent runs on Claude Opus 5.5 and inverts the usual chatbot pattern. Instead of sitting idle until a conversation starts, Duet sends a short morning summary covering the most important items across a user's connected apps, then monitors those apps through the day and only contacts the user when something requires attention or a decision.

What Duet can do

Anthropic's product page describes work spanning simple and complex requests. At the light end, a single message can get a meeting scheduled or a dashboard assembled. At the heavier end, The Verge cites jobs such as financial and sales forecasting, prototyping, and building web apps. Small-business chores are a particular focus: tracking down unpaid invoices and drafting the reminders, preparing payroll for review, and getting contracts ready to send. Duet can also flag items in an inbox or calendar and draft replies for approval.

The agent reuses a user's existing Claude account, so memory, skills, and connected apps are available from the first message, and the entire history sits in one thread. The Verge reports that Duet operates on a virtual machine with a browser for carrying out work, and finished output returns as a document, spreadsheet, or deck for review. Interaction works over text or voice, with no prompts or setup required.

Guardrails and privacy

Before sensitive steps such as sending payments, sending messages, or deleting data, Duet asks for approval. Anthropic says those approvals, along with logins and safety checks, run in the user's own cloud computer so that sensitive details stay private. The Verge additionally reports a new dedicated classifier and safety system prompt intended to stop the agent from sharing the most sensitive category of information, such as credit card numbers.

Availability and constraints

Duet ships as a standalone app on iOS and Android and is accessible on the web at claude.ai. Access is waitlisted and restricted to Pro and Max subscribers in the US; Team and Enterprise plans are not included yet. During the preview, Duet shares a plan's usage limits with the rest of Claude, and its background activity counts against them. Anthropic notes that larger jobs, like a financial model or a deck, consume more model turns and background work.

Anthropic is candid about the rough edges. The company says Duet may misinterpret a conversation thread, misjudge how urgent something is, or miss items that arrive between check-ins, and The Verge reports that Anthropic considers it weaker at personal errands such as booking tickets.

Why it matters

Duet is another step in the shift from chatbots that answer questions to agents that work autonomously in the background and initiate contact. The competitive framing is telling: with Meta and OpenAI shipping comparable products, proactive agents are becoming table stakes for major AI vendors rather than experiments.

The design choices matter just as much. Duet will touch payroll, invoices, and contracts, so the approval prompts, isolated cloud execution, and the new classifier for sensitive data represent Anthropic's attempt to make autonomous access to business-critical information acceptable. There is a visible trade-off in the details: background work drawing down existing subscription limits gives early adopters a concrete cost to weigh, and the acknowledged failure modes — missed messages, bad urgency calls — are precisely the errors that hurt most in a business context. The research preview label signals Anthropic knows this; the waitlist will show how much tolerance professionals have for that uncertainty.