What is Claude Duet? Anthropic's new proactive agent, explained

Claude Duet is a research preview, and Anthropic is rolling it out by waitlist rather than opening it to everyone at once. The starting group is Claude Pro and Max subscribers in the United States. Team and Enterprise plans are not included yet, and Anthropic has not said when, or whether, that will change.

To try it, you join the waitlist from the Claude app or from Anthropic's Claude Duet pages. There is no separate signup flow outside your existing Claude account: Claude Duet uses your Claude account as is, including your memory, skills, and connectors, so there is nothing extra to configure once you're let in.

Once you have access, Claude Duet is available as a standalone app for iOS and Android, and also on claude.ai on the web. You message it the way you'd message a teammate, by text or voice, with no prompts or setup, and your history stays in one running thread rather than resetting with every conversation.

During the research preview, Claude Duet shares your plan's usage limits with the rest of Claude, and it uses them differently than an ordinary back-and-forth conversation. Its background checks, the quiet monitoring it does between messages, count toward those limits too. A larger task, like putting together a financial model or a slide deck, takes more model turns and more background work than a quick question does, so it draws on the limit faster.

An agent that can send messages, make payments, sign documents, and connect to over 6,000 apps and plugins is also an agent that can do real damage quickly if it acts on a wrong guess. Anthropic's answer is a hard checkpoint: before Claude Duet sends a message, makes a payment, signs a document, or deletes anything, it shows you exactly what it's about to do and waits for your approval. Nothing in that list goes out without you seeing it first.

This is the detail worth paying attention to if you're deciding whether to trust an agent with real accounts and real inboxes. It is not a vague promise to 'be careful.' It is a specific, named list of the actions that always stop for a human: sending, paying, signing, deleting. Anthropic also says Claude Duet cannot trade securities or move money between accounts, even with your approval, which puts a hard line in front of the riskiest category of action rather than leaving it to the approval step alone.

The practical effect is that you stay the one who decides, even on a day when Claude Duet has done a lot of the actual work. You read a drafted reply instead of writing one from scratch, and you approve a prepared payroll run instead of building it. The checking-first habit this encourages, reviewing what an agent is about to do rather than assuming it went well, is the same habit worth building whenever you delegate a real task to AI, agentic or not.

Anthropic draws the comparison specifically against the regular Claude app, not against every other Claude product. In the standard Claude app, the app waits for you to open a chat, and you work on something alongside Claude in that session. Claude Duet comes to you instead: it tells you what needs your attention each morning, takes tasks you hand it by message, and reaches out when a task is done or needs your approval.

Claude Duet is not a separate account or a reset. It uses your existing Claude account, including your memory, skills, and connectors, from the first message. What changes is the direction of the interaction: in the regular app, you initiate; with Claude Duet, it initiates, within the bounds of the approval step described above.

Anthropic's own materials describe this one comparison, Claude Duet against the standard Claude app, in detail. They do not lay out a feature-by-feature comparison against Claude Code or Claude Cowork, Anthropic's other products, so this page does not invent one. What's confirmed is narrower and more useful than a guess: Claude Duet is the proactive, message-first way to work with your existing Claude account, and the standard app is the version you still open yourself.

It is a research preview, and Anthropic says that plainly rather than hiding it in fine print. It may misread a thread or miscalibrate what's urgent, and it can miss something that arrives between one check-in and the next. Treat its judgment about what deserves your attention as useful, not infallible, especially in the first weeks of using it.

Two categories are off the table even with your approval. Claude Duet cannot trade securities or move money between accounts, full stop, regardless of what you approve. And Anthropic says personal errands, such as booking tickets, aren't quite as strong a fit for it yet as business and administrative tasks are.

There's also a visibility gap worth knowing about: Claude Duet cannot yet display its own browser to you, so while it shows you what it's about to do before an approved action goes out, you cannot watch it work inside that virtual machine step by step the way you might watch a screen share. For now, the approval step is the window you get into its process, and it's the one Anthropic built the product around.