Claude Duet is Anthropic’s answer to a question the other labs asked first: what if your AI assistant spoke up before you opened the chat window? Anthropic announced it on October 21, 2026 as a research preview, and it puts Claude into the same “proactive agent” race as OpenAI’s Dots, Google’s Gemini Agent and Meta’s Muse. Here is what it does, who can use it, and what the early coverage leaves unanswered.
What Claude Duet Actually Is
Duet is a separate app inside the Claude account, aimed at professionals and small business owners who have more to track than they can watch. According to MacRumors, it connects to your apps, flags what needs attention, and reaches out first instead of waiting for a prompt. It also keeps everything in one ongoing thread rather than a pile of separate conversations.
It runs in the cloud on its own virtual computer with a browser, so it keeps working after you close the app. It also carries over the memory, skills and connected apps you already set up in Claude, which means very little extra setup.
How Claude Duet Works Day to Day
The reporting describes a few clear behaviors:
- Morning brief: a short summary of the important items across your connected apps.
- Background watching: it checks those apps every few hours and alerts you only when something needs attention.
- Tasks on request: you message it by text or voice for admin work such as scheduling, or bigger jobs like dashboards, forecasts and web apps.
- Finished work: results come back as a doc, sheet or deck for you to review.
One example from 9to5Mac’s coverage is a Stripe connection that checks unpaid invoices and drafts payment reminders. That is the kind of small, repeated chore a proactive agent is meant to take off your plate. The memory side of this matters too, and our explainer on why agents keep forgetting things covers why a single running thread is a bigger deal than it sounds.
The Approval Gate: What It Won’t Do Alone
Letting an agent act for you is the risky part, so Anthropic built in approval steps. Per the launch coverage, Claude Duet asks before it sends a message, makes a payment, signs a document or deletes a file. Reports also say it does not see your passwords or logins when you give it access to an account, and that a dedicated classifier checks each message for sensitive information leaks. SiliconANGLE adds that the design follows Anthropic’s “trustworthy agents” framework, the same one behind Claude Code and Claude in Chrome.
The weak point of any approval system is volume. If an agent asks too often, people start clicking yes without reading. Whether Duet strikes the right balance is something only real use will show.
Who Can Use It and What It Costs
Access is narrow for now:
- It is a research preview behind a waitlist.
- It is limited to Pro and Max subscribers in the United States.
- Team and Enterprise plans are not included yet.
- There is no separate price. Its work, including background checks, counts against your existing plan limits.
Several outlets say it is built on Opus 5.5, though the SiliconANGLE piece does not state the model. Outlets also differ on platforms: MacRumors mentions the web and an iOS app, while SiliconANGLE lists web, iOS and Android. Treat the platform list as unconfirmed until Anthropic’s own pages settle it.
The cost angle is worth watching. An agent that checks your apps every few hours spends usage even when you are not typing, so heavy users may burn through limits faster than they expect. Anthropic has not published how much background work a typical day uses. For a sense of how pricing across the Claude line is shifting, see our look at Claude Haiku 5.5’s price cut.
What We Don’t Know Yet
Everything above comes from Anthropic’s announcement as relayed by the press. No independent testing exists, since the product launched this week behind a waitlist. Anthropic itself calls it early: Duet can misread a thread or misjudge what is urgent, and it is built around work, so personal errands like booking tickets will not be treated as urgent. It also can’t show you its browser while it works, which limits how closely you can watch what it does.
The open questions are the practical ones. How many false alerts will it send? How often will it ask permission? How fast does background usage eat into plan limits? And when does it reach people outside the US or on team plans? Until early users answer those, Claude Duet is a promising design more than a proven tool.
The wider pattern is clear, though. Every major lab now sells an assistant that works unprompted, and the differences will come down to trust, cost and how little attention they demand. If you already use Claude, joining the waitlist is cheap. If you do not, wait for the first independent reviews before changing your workflow.
Sharing Is Caring