Claude Duet Keeps Working After You Close the Laptop

Most AI assistants wait to be asked. Anthropic's newest product tries the opposite approach. On Tuesday the company opened a research preview of Claude Duet, a standalone app where Claude works as a proactive agent. In Anthropic's own words, it "comes to you with what needs your attention, completes tasks you ask of it, keeps working when you step away, and checks with you before it acts on your behalf."

That last clause is the one worth reading twice. Duet is pitched at professionals and small-business owners who want a hand with the daily grind of email, calendars and follow-ups. It opens with a morning brief drawn from the apps you connect, then checks back every few hours and flags anything that looks like it needs you. Reuters described it as a proactive assistant built around that rhythm of brief, check, nudge.

What is actually running

Under the hood, Duet runs on Claude Opus 5.5 inside a cloud virtual machine with its own browser. Because the work happens on Anthropic's machine rather than yours, tasks carry on after you close the laptop or pocket the phone. The app connects to more than 6,000 services, Gmail and Slack among them, and it is available on the web and on iOS and Android.

Anthropic has put guardrails around the autonomy. Duet asks for approval before anything sensitive, such as sending a message, making a payment or signing a document, and the company says it does not see your passwords or logins when you grant it access to an account. It is honest about the limits too. Anthropic notes that Duet can misjudge how urgent something is, and that it is weaker at personal errands like booking tickets than at the office work it was built for. Access is gated behind a waitlist, and for now it is open only to Pro and Max subscribers in the United States.

The proactive-agent race

Duet does not arrive in a vacuum. 9to5Mac framed it squarely as Anthropic's answer to OpenAI's Dots and Meta's Muse, the rival proactive assistants that also try to work in the background and surface what matters. The pitch across all three is similar: stop making people prompt, and start having the software notice.

There is something genuinely useful here, and something a little uneasy about it. An agent that works through the night on its own cloud computer is a real productivity gain. It is also a thing that acts while nobody is watching, which is exactly why Anthropic built the approval step in. Readers who have followed the wider story will recognise the tension. We have written about whether a shopping agent will spend your money without asking, and about banks locking agents out of moving customer funds altogether. Duet's whole design is an attempt to answer that worry before it lands: keep working, but ask first.

For now it is a preview, with a waitlist and a short leash. Whether people want an assistant that comes to them, rather than one they summon, is the question the research preview is really there to test.