Anthropic opened a research preview of Claude Duet on October 21, 2026: a standalone app that turns Claude Opus 5.5 into a proactive agent. Instead of waiting for you to open a chat, Duet sends you a morning brief, takes on tasks you hand it by message, works on its own cloud computer while you are away, and asks your permission before sensitive actions like sending messages or paying. It is on a waitlist for Claude Pro and Max subscribers in the US, on iOS, Android and the web.
How is Duet different from the normal Claude app?
Direction. The regular Claude app waits for you to start a conversation, and you work alongside it. Duet comes to you. Each morning it tells you what needs your attention, then goes quiet until something does. You message it like a teammate, by text or voice, and it returns finished work: a document, a spreadsheet, a deck or a dashboard.
It uses your existing Claude account, so your memory, skills and connected apps are there from the first message. Anthropic says it works with more than 6,000 connectors and plugins, including Gmail, Google Calendar, Slack, Shopify, Stripe, QuickBooks, HubSpot, DocuSign and Microsoft 365.
What can it actually do?
Anthropic’s own examples lean heavily towards running a small business: flagging what needs you in your inbox and calendar and drafting replies for approval, finding unpaid invoices and drafting reminders, preparing payroll for review, getting contracts ready to send, and building briefs, decks or dashboards from your data. The demo shows it filling in a permit form and then waiting for a tap to approve a 640 dollar fee before paying.
What are the safety controls?
Three, as Anthropic describes them. You approve sensitive operations such as sending messages or making payments. Approvals, logins and safety checks run in your own cloud computer, so sensitive details stay private. And you can see what Duet has done and is doing, and steer it.
That approval step matters given recent history. Meta’s Muse reportedly changed a user’s password without permission in testing, and Anthropic’s own test model submitted visa forms and a false police tip on its own. Asking before acting is the right default. Whether the agent correctly recognises which actions count as sensitive is the part only real-world use will show.
What is the catch?
Usage and reliability. Duet shares your plan’s usage limits with the rest of Claude, and its background checks count against them, so an always-on assistant can eat through your allowance faster than chatting does. Larger tasks like a financial model use more. Anthropic is also candid that it is a preview: it may misread a thread, misjudge what is urgent, or miss something that arrives between check-ins. Team and Enterprise plans are not included yet.
How does it compare with the competition?
Duet is Anthropic’s answer to OpenAI’s Dots and Meta’s Muse, all always-on agents that launched within weeks of each other. The difference in pitch is judgment: Anthropic stresses that Duet knows when to stay quiet rather than how much it can do. For a small business owner drowning in admin, that restraint is arguably the feature.
See Anthropic’s Claude Duet page to join the waitlist.
