Anthropic unveils Claude Duet: a proactive agent that works for you, even when you're away

Anthropic has opened a research preview of Claude Duet, a new app in which Claude acts as a proactive agent. According to the company, the system comes to users with what needs their attention, completes the tasks it is given, keeps working even when the user logs off, and checks in before acting on their behalf. The article, dated October 21, 2026, says access is via a waitlist, starting with Pro and Max subscribers in the United States.

The product is aimed at professionals and small-business owners who have more on their plate than they can keep track of. It is available as a standalone app on iOS and Android, as well as on claude.ai. During the preview it is included in the Pro and Max plans and shares the same usage limits as the rest of Claude. For now it lives in its own app, separate from the main Claude app, and Anthropic says what it learns from early users will shape how the two are integrated.

The pitch rests on the system's judgment. Anthropic explains that it designed it around Claude Opus 5.5 so that it can decide what the user needs, what can wait and when it is best to ask first. Each morning it sends a short summary of the most important items across connected apps. Throughout the day it monitors those apps and flags things only when something needs attention. It also takes on tasks of varying size, from administrative chores such as scheduling meetings to more complex ones, such as building dashboards drawing on dozens of tools, creating financial or sales forecasts, building prototypes and developing web apps.

How it works. Claude Duet inherits the memory, skills and connected apps from the user's Claude account, so setup takes little effort. Users type or talk to it in a single continuous thread, so a quick answer and a week-long project start in the same place. It has a virtual machine with its own browser, so it does not need the user's laptop or phone to be on: the user describes the task in a short message, the agent decides which apps to use and which steps to take, and it notifies the user when it is done or needs review.

On notifications, Anthropic specifies that Claude Duet checks connected accounts every few hours, because doing so continuously would exhaust plan limits much faster. The company is still testing its judgment about when to send messages and will adjust it based on testers' feedback. Users can also specify what they want to hear about and how often.

User control. Claude Duet reads, drafts and prepares work on its own, but before sending a message, making a payment, signing a document or deleting something, it shows exactly what it is about to do and waits for approval. Some actions are off-limits even with approval, such as trading securities or moving money between accounts. Users can disconnect any app the agent has access to.

Connectors. According to the article, it can use more than 6,000 connectors and plugins from the Claude Marketplace. Examples cited include: finding pending Gmail and Slack messages and helping reply to them; reviewing unpaid invoices in Intuit QuickBooks or Stripe and drafting reminders; preparing and running payroll in Gusto; drafting a job posting and publishing it on LinkedIn after the user's approval; preparing agreements and sending them for signature with Docusign; drafting follow-ups using HubSpot history; turning a product launch into a social campaign in Canva; and setting up, designing and managing a Shopify store. A more complete visual editor for designing and previewing the store before publishing it is coming soon.

Purchases. Claude Duet can buy things for the business with Link, Stripe's digital wallet: it confirms the item with the user, Link asks for approval of the spend and payment is made with a virtual card, so the real card number is not shared. It can also use Shopify's Shop Pay to pay at online stores. Soon, with permission, it will be able to sign in to sites using credentials stored in 1Password.

Security and privacy. Anthropic says Claude Duet builds on years of work on safety, privacy and agents. It follows the company's framework for trustworthy agents, which also guides Claude Code and Claude in Chrome. Credential handling follows the design of Claude Managed Agents, so the agent never directly sees passwords or access tokens. Actions are approved by the company's auto mode classifier, the same safeguard that protects Claude and Claude Code from taking unrequested steps. Defenses against hidden instructions and prompt injection come from its research on browser use. In addition, red teams and an outside firm carried out penetration testing, and the product is part of the public bug bounty program.

For sensitive information, a dedicated classifier and a safety system prompt are debuting that review every message and guard against leaks. They are designed to refuse requests to share the most tightly restricted tier of data, such as credit card numbers and government IDs. Other sensitive information, such as health or personal finance data, is sent only when the user has explicitly instructed it. Anthropic says it will keep strengthening these protections and will share more details in the coming days; indeed, a related page on Claude Duet's safety, privacy and security is dated October 23, 2026.

Testimonials. The article says Anthropic employees use Claude Duet to get more done and also to leave more room for thinking. Boris Cherny, who leads the Claude Code team, talks through his mental to-do list with it during walks. Chelsea Larsson, who leads model and product behavior design, hands it her list before heading off with her children and gets the results in a nightly report. Karan Ahuja, who leads the Claude Duet team, uses it to schedule user interviews, prepare prototypes and even ship code.

Small-business owners also took part. Jess Provencher (Pro Homes) gets a summary every morning and recounts that the agent once flagged, unprompted, a property showing that had been put on the wrong day in the calendar. Amin Ebrahimi, owner of Shahs of Kabob in South Florida with about 110 employees, has used it for accounting, customer service and marketing, and has reviewed invoices store by store; he says this way he does not need to open 100 tabs. Mya Dalton, a brand strategist, says Claude Duet prepared a briefing for a call with a prospective client and noticed it fell right before her flight, so it offered to move it.

Limitations. Anthropic acknowledges that this is an early version that may misread a thread or misjudge what is urgent. It was built around work, so personal errands, such as booking tickets, are less robust, and it cannot yet show its browser to the user. On usage limits, it clarifies that background checks count against them and that large tasks, such as a financial model or a presentation, require more model turns and more background work than a normal conversation.

In short, Claude Duet is Anthropic's bet on an agent that takes the initiative, with its own virtual machine, integration with thousands of business tools and a design that leaves the important decisions to the user. Those interested can join the waitlist and will receive an email when they get access.