Anthropic launched Claude Duet as a research preview on Wednesday, October 21, 2026. The new app turns Claude into a proactive agent that flags what needs attention, handles the tasks it’s given and keeps working while its user is away. It still asks for approval before acting on their behalf. Anthropic says it built the product for “professionals and small business owners with more on their plates than they can track.”
Access is by waitlist, with invites going out in waves, and for now it’s limited to Pro and Max subscribers with a US Claude account. Team and Enterprise plans aren’t included yet, and Anthropic hasn’t given a date for other countries. Claude Duet is available as an iOS and Android app and on claude.ai.
A morning brief, then messages only when needed
Each morning, according to Anthropic’s help center, Claude Duet sends a short brief on what matters most across a user’s connected apps, including email and calendar, and can take first steps such as drafting replies to review. During the day, it checks those apps every few hours and reaches out only when something calls for it. Anthropic says checking nonstop would burn through a plan’s usage limits much faster.
Users talk to it in a single thread, by text or voice, with optional topic channels that share the same memory. The tasks Anthropic lists range from scheduling meetings to building dashboards, financial or sales forecasts, prototypes and web apps.
The agent runs on a cloud virtual machine with its own browser, so users can switch off their laptop and phone while it works. It inherits the memory, skills and connectors already set up in a Claude account, and it can search past chats unless that search has been turned off. There’s no model picker: Claude Duet runs on Opus 5.5, covered in our comparison of Claude models, and decides on its own how much effort a job needs. The next day, Anthropic handed the format behind those skills, Agent Skills, to a Linux Foundation group.
More than 6,000 connectors and one-time virtual cards
Anthropic says Duet can tap more than 6,000 connectors and plugins in its marketplace. The examples it gives are everyday small-business chores: finding missed Gmail and Slack messages, drafting reminders for unpaid invoices found in QuickBooks or Stripe, preparing payroll in Gusto, posting a job listing to LinkedIn once approved, sending contracts for signature through Docusign and managing a Shopify store.
To buy things, Claude Duet uses Stripe’s Link digital wallet or Shopify’s Shop Pay. According to Anthropic, each payment uses a one-time card, and the real card number isn’t shared. The company also plans to let the agent sign in to sites with logins saved in 1Password, if users allow it.
The help center notes that some connectors, including Slack, work with limits, and that Microsoft 365 isn’t available yet. The agent won’t call or text other people, but it can send email from connected accounts with approval.
Approval before it sends, pays, signs or deletes
Anthropic says Claude Duet reads, drafts and prepares work on its own. Before it sends a message, makes a payment, signs a document or deletes anything, it shows exactly what it’s about to do and waits for a go-ahead. If approval doesn’t come right away, the agent finishes the rest of the task and brings the request back later. Some actions stay off-limits even with approval, such as trading securities, moving money between accounts or accessing someone else’s account.
Anthropic says the agent never directly sees passwords or sign-in tokens. When a site asks for a password, the user enters it in a separate form on their own device. A new classifier, paired with a dedicated safety system prompt, is designed to keep the most sensitive information, such as credit card numbers and government IDs, from being shared. Health or personal finance details go out only on explicit instruction.
The help center also covers prompt injection, where an email or a web page hides instructions meant to hijack the agent. It concedes that “no defense is perfect,” which is one reason consequential actions require approval.
Anthropic says red teams and an outside firm ran penetration tests, and it promises more safety details “in the coming days.” On October 19, Microsoft described AI agents deployed at work as an insider risk that needs watching like an employee with too many permissions.
Included in Pro and Max, with shared usage limits
During the preview, Claude Duet comes with the Pro and Max plans at no extra cost. Claude’s pricing page lists Pro at $20 a month, billed monthly and before tax (about €18 at the European Central Bank’s October 21, 2026, rate), and Max from $100 a month (about €89). Anthropic hasn’t said what Duet will cost after the preview.
The agent draws from the pool it shares with the rest of Claude (the five-hour session limit and the weekly cap), at the same rate as a chat. Because background work counts too (the morning brief, check-ins, larger jobs), Anthropic warns that subscribers may hit their Claude plan’s usage limits sooner than with chat alone. Everyone admitted to the preview gets one full limit reset, usable until November 30, 2026.
Meta and OpenAI got there first
According to CNBC, Meta’s personal agent, Muse, launched in September and climbed to the top of Apple’s App Store. A few weeks later, OpenAI unveiled Dots, its “always-on agents.” During Meta’s quarterly earnings call on Wednesday, Mark Zuckerberg told analysts that more than 10 million people in the US and Canada had set up a Muse agent in a month and a half. Two days later, Meta launched Muse in Japan, its first country outside North America.
Engadget says Anthropic is late to the party, but notes that Dots is limited to OpenAI’s Pro tier, which starts at $100 a month (about €89), while Duet is included starting with Claude’s Pro plan. Meta, CNBC adds, mostly targets consumers. Anthropic says it built Duet around work and that personal errands, like booking tickets, aren’t as strong. Google, meanwhile, added a small-business hub to the Gemini app on October 20.
Hand-picked testimonials and admitted flaws
All the user feedback published so far comes from Anthropic. The announcement quotes Anthropic staff, including Boris Cherny, who leads the Claude Code team, and three outside testers. According to Anthropic, Amin Ebrahimi, who owns Shahs of Kabob in South Florida (about 110 employees), uses Claude Duet for accounting, customer support and marketing. Jess Provencher, a broker and owner of the real estate agency Pro Homes, says the agent, unprompted, flagged a property showing that was on her calendar for the wrong day.
Anthropic also admits that this first version may misread a thread, misjudge what’s urgent or miss a message that arrives between check-ins. The agent can’t show its browser, can’t see files on a user’s computer or all of their projects, and can read work done in Claude but not change it. None of the official pages (announcement, product page, help center) says where the agent’s virtual machines run.
On Reddit, where the launch was among the most-followed posts in Claude subreddits overnight, one user was mainly pleased that Duet comes with the Pro plan and hoped that would hold in a global rollout.
What it means
With Claude Duet, Anthropic moves from an assistant that waits for a prompt to an agent that takes the initiative, and it’s explicitly aiming at professionals and small businesses. Memory, skills and connectors already existed in Claude. The new parts are the virtual machine that keeps working without the user, the periodic check-ins and the approval flow for sensitive actions.
Outside the US, Duet isn’t available yet. Examples built around American software such as QuickBooks and Gusto point to a product designed first for that market, and the missing Microsoft 365 connector shuts out companies whose email runs on that suite, at least for now.
The biggest unknown is cost. Background work eats into the same quotas as chat, and Anthropic hasn’t set a price or a Duet-specific limit for after the preview. Safety ultimately rests on a person approving sensitive actions, and the help center itself links that requirement to the limits of prompt-injection defenses.
Four things are worth watching: the safety details promised for the coming days, a date for Team and Enterprise plans, Microsoft 365 joining the connectors, and any rollout beyond the US.
Sources
- Introducing Claude Duet, Anthropic, October 21, 2026
- Claude Duet, product page, Anthropic, accessed October 22, 2026
- Get started with Claude Duet, Claude Help Center, Anthropic, accessed October 22, 2026
- Claude pricing, Anthropic, accessed October 22, 2026
- Anthropic announces preview of Claude Duet, a personal AI agent to take on Muse, Dots, CNBC, October 21, 2026
- Claude Duet is Anthropic’s answer to Meta Muse and OpenAI Dots, Engadget, October 21, 2026
- Euro foreign exchange reference rates, US dollar, European Central Bank, October 21, 2026
Header image: MACCUS chart, data from Anthropic (Claude Duet announcement, product page and help center), October 21-22, 2026
