Anthropic opens Claude Duet, a proactive agent app that asks before it acts

What Anthropic launched

Anthropic has opened a research preview of Claude Duet, a separate app in which Claude acts as a proactive agent rather than a chat partner that waits to be asked. In its announcement on 21 October 2026, the company says Duet "comes to you with what needs your attention, completes tasks you ask of it, keeps working even when you step away, and checks with you before it acts on your behalf." It is aimed at professionals and small business owners "with more on their plates than they can track."

Duet is a standalone app for iOS and Android and also runs on the web at claude.ai. For the preview it is included in Claude Pro and Max plans, starts with subscribers in the US, and is gated by a waitlist. Anthropic says it built the agent around Claude Opus 5.5 and that what makes it useful is judgment: deciding what needs the user, what can wait, and when to check in first.

The launch is Anthropic's answer to a category its rivals opened earlier this month. The Verge framed Duet as a competitor to Meta's Muse and OpenAI's Dots, both of which pitch an assistant that works in the background and reaches out on its own.

How Claude Duet works day to day

Anthropic describes a simple daily rhythm. Each morning Duet sends a short brief on the most important items across the user's connected apps. During the day it keeps tabs on those apps and sends a message only when something needs attention. Users can also hand it tasks, from scheduling meetings to building dashboards that pull from dozens of tools, creating financial or sales forecasts, prototyping, and building web apps.

Everything happens in one ongoing thread, by typing or by voice, so a quick reply and a week-long project start in the same place. Duet carries over the memory, skills and connected apps from the user's existing Claude account, which Anthropic says keeps setup to a minimum. The product page adds that it returns finished work, such as a document, spreadsheet or deck, for review.

The key technical difference from the regular Claude app is where the work runs. Duet has its own virtual machine with a browser, so it does not need the user's laptop or phone to be switched on. A user describes a task in a short message, Duet works out which apps to use and which steps to take in the background, and it sends a notification when the task is finished or needs review.

Duet does not watch connected accounts constantly. Anthropic says it checks them every few hours, because checking nonstop would use up a plan's limits much faster, and users can tell it what they want to hear about and how often. The product page notes the trade-off plainly: Duet can miss something that arrives between check-ins.

Approvals, credentials and the new sensitive-data classifier

Anthropic leans hard on control. Duet reads, drafts and prepares work on its own, but before it sends a message, makes a payment, signs a document or deletes anything, it shows exactly what it is about to do and waits for approval. Some actions are off-limits even with approval, including trading securities and moving money between accounts. Users can disconnect any app Duet has access to.

The safety design reuses pieces Anthropic has shipped elsewhere. Credentials are handled the way they are in Claude Managed Agents, so Duet never directly sees passwords or sign-in tokens. Actions are screened by the auto mode classifier that also guards Claude Code, and its defences against prompt injection come from Anthropic's browser-use research. Anthropic says red teams and an external firm penetration-tested Duet, and that it is covered by the public bug bounty programme.

The new element is a dedicated classifier and safety system prompt that checks each message for leaks of sensitive information. Anthropic's reasoning is that an agent which reads email and sends messages is constantly deciding where information should go. The classifier is designed to refuse to share the strictest tier of data, such as credit card numbers and government identification, while other sensitive information, like health or personal finances, goes out only when the user has explicitly told Duet to send it. Anthropic says it will publish more on Duet's safety design in the coming days.

Connectors, payments and small-business tasks

Duet can use the more than 6,000 connectors and plugins in the Claude Marketplace. Anthropic's examples are practical small-business chores: finding missed Gmail and Slack messages and helping reply, checking Intuit QuickBooks or Stripe for unpaid invoices and drafting reminders, preparing payroll in Gusto, drafting a job post for LinkedIn, sending agreements through Docusign, drafting follow-ups from HubSpot deal history, turning a launch into a Canva campaign, and setting up or managing a Shopify store.

It can also buy things. Purchases go through Link, Stripe's digital wallet: Duet confirms the item, Link asks the user to approve the spend, and the payment uses a virtual card so the real card number is not shared. Duet can also check out with Shop Pay, which Shopify recently opened to agents through its Shop MCP catalogue and checkout tools. Anthropic says sign-in to sites with logins saved in 1Password is coming, with the user's permission.

Anthropic cites early testers. Amin, who owns Shahs of Kabob in South Florida with about 110 employees, used it for accounting, customer support and marketing and said "I don't need to open up 100 tabs." A brand strategist, Mya Dalton, said it noticed a client call fell right before her flight and offered to move it. These are vendor-selected anecdotes, not measured results.

Where Duet fits against Muse and Dots

Proactive agents are suddenly the main battleground among the big labs. OpenAI introduced Dots, its always-on agents, and Meta has pushed Muse as an assistant that acts across apps since its Muse Wonder release. Anthropic's pitch is narrower: work first, aimed at professionals and owners of small businesses, with approval gates and data rules at the centre.

The choice of model matters for cost. Duet runs on Claude Opus 5.5, Anthropic's top general model, and the company is upfront that background checks and large tasks consume plan limits. That is a different bet from an agent priced as a cheap always-on layer.

Duet is also kept apart from the main Claude app for now. Anthropic says what it learns from early users will decide how the two products work together, which suggests the separation is an experiment rather than a final design.

Who can use it and how usage counts

Access starts with Claude Pro and Max subscribers in the US through a waitlist; Anthropic says it will email people when Duet is ready for them. Team and Enterprise plans are not included yet, according to the product page's FAQ.

There is no separate price during the preview. Duet shares the plan's usage limits with the rest of Claude but uses them differently from a normal conversation: its background checks count toward the limits, and larger jobs such as building a financial model or a deck take more model turns and background work. Users who lean on Duet heavily should expect to hit limits sooner in their regular Claude chats.

Limitations Anthropic admits

Anthropic calls this an early version. It says Duet "still might misread a thread or miscalibrate what's urgent," that personal errands such as booking tickets are weaker because the core was built around work, and that it cannot yet show the user its browser. The few-hours check-in cycle means it is not a real-time monitor.

Some questions remain open. Anthropic has not published data on how often the approval prompts or the sensitive-data classifier misfire, how the auto mode classifier performs on Duet's broader set of actions, or what happens when a connected app returns content designed to manipulate the agent. AI Lookout has not tested Duet, which is waitlisted, so the capabilities described here are Anthropic's own account.