Claude Duet: Anthropic opens preview of background agent app

Key takeaways

• Anthropic has opened a research preview of Claude Duet, a standalone app in which Claude works as a proactive background agent for professionals and small business owners.
• Duet sends a morning brief, watches connected apps during the day, and carries out tasks on its own cloud virtual machine, returning to the user when it needs a decision.
• It runs on Claude Opus 5.5 and is available by waitlist to Pro and Max subscribers in the US, on iOS, Android and claude.ai. Anthropic has not announced a separate price.
• Sending messages, paying, signing documents and deleting anything require explicit approval. Duet will not trade securities or move money between accounts, even if the user approves.
• Anthropic says Duet can still misread a thread or misjudge urgency. It enters a crowded field that includes Meta's Muse and OpenAI's Dots.

What is Claude Duet, the Anthropic agent app?

Claude Duet is Anthropic's new agent app that acts first instead of waiting for a prompt. Announced October 21, 2026, as a research preview, it sends a morning brief, watches connected apps and runs tasks on its own cloud computer, checking in before acting. For Anthropic's latest safety findings, see our analysis of the Claude Mythos 5 AISI incident.

The launch drew wide attention on October 22. TechCrunch called the app Anthropic's bet on agents and placed it in a growing market of "always-on" assistants. Digital Trends also covered the release, and the product's official X account announced the research preview. In a separate MIT preprint, 13 Claude Opus 4.7 agents spent 30 hours on a simulated island with no goal.

Duet is separate from the main Claude chat app. It carries over the memory, skills and connected apps from a user's Claude account, so there is little to set up. Typed and voice messages go into a single, ongoing thread.

What Claude Duet actually does

Anthropic built Duet around three jobs: briefings, monitoring and background tasks.

Morning briefs and monitoring

Each morning, Duet summarizes the most important items across the user's connected apps. During the day it checks those apps and messages the user only when something needs attention. Anthropic says connected accounts are checked every few hours rather than continuously, which saves usage. Because of this, the company warns that Duet can miss something that arrives between check-ins.

Background work on its own computer

Tasks run on a virtual machine with its own browser, which Anthropic's product page calls "your own cloud computer." This means the user's laptop or phone does not need to stay on while a job runs. Duet sends a notification when a task is finished or needs review. For now it cannot show the user its browser while it works, according to Anthropic and TechCrunch.

Anthropic's examples range from small admin jobs to larger projects:

• finding missed messages in Gmail and Slack;
• chasing unpaid invoices in QuickBooks or Stripe;
• preparing payroll in Gusto for the user to review;
• drafting a LinkedIn job posting for approval and sending agreements through Docusign;
• setting up a Shopify store;
• building dashboards, decks, financial or sales forecasts, prototypes and web apps.

Anthropic says Duet can draw on more than 6,000 connectors and plugins from the Claude Marketplace. TechCrunch reported the same figure.

Approved purchases with Link and Shop Pay

Duet can buy things online, with limits. It uses Link, Stripe's digital wallet: Duet confirms the item with the user, then Link asks the user to approve the spend. Payment goes through a virtual card, so the real card number is not shared. Duet can also check out at online stores through Shopify's Shop Pay. Anthropic says sign-in to websites with logins saved in 1Password, with the user's permission, is "coming soon."

Claude Duet at a glance

Who can get Claude Duet, and what it costs

Access starts with US users on Claude's Pro and Max plans, who must join a waitlist on the product page. The product page says Team and Enterprise plans "are not included yet." Anthropic has not said when other countries will be added.

There is no extra fee, but Duet is not unlimited. During the preview it shares the subscription's usage limits with the rest of Claude. Background checks count toward those limits, and large jobs, such as building a financial model or a slide deck, use more model turns and so more of the allowance. Digital Trends noted the same trade-off: the more work Duet does in the background, the faster a user reaches the plan's cap.

Safeguards and stated limits

Anthropic lists several layers of protection:

• Approval before sensitive actions. Before Duet sends a message, makes a payment, signs a document or deletes anything, it shows the exact action and waits for the user to approve it.
• Actions it refuses. Even with approval, Duet will not trade securities or move money between accounts.
• Credential handling. Anthropic says Duet uses the same design as Claude Managed Agents, so it "never directly sees your passwords or sign-in tokens."
• Sensitive data filter. A dedicated classifier and system prompt reject requests to share the most sensitive data, such as credit card numbers and government ID numbers. Health and personal finance details are shared only when the user explicitly says so.
• Prompt-injection defenses. These draw on Anthropic's research into browser use. The company says red teams and an outside firm tested the app, and it is covered by Anthropic's public bug bounty program.
• User control. Users can watch and steer what Duet does, and they can disconnect any app it has access to.

Anthropic is also open about weak spots. It says Duet "still might misread a thread or miscalibrate what's urgent," and that personal errands, such as booking tickets, are weaker than work tasks. TechCrunch also reported that Duet is not good at errands or ticket booking. Anthropic says it will keep tuning when Duet messages users, based on feedback from testers.

How Claude Duet compares with rival personal agents

Duet arrives as several large technology companies race to build agents that run in the background. TechCrunch named Meta's Muse, OpenAI's Dots, Hark, Instinct, Grok Bot and Fo as rivals, and Digital Trends said Meta's Muse helped start the trend before OpenAI followed with Dots.

Many of these products now target business users. According to TechCrunch, Meta has released a small-business version of Muse that connects to other apps, and Dots integrates with professional tools. Public reports so far do not give a full, feature-by-feature comparison, so the table below sticks to what has been reported.

The clearest difference in Anthropic's pitch is its narrow focus on business work and its approval gates, rather than broad consumer errands. Bullquill has covered the wider agent race, including concerns about how OpenAI's agents were monitored internally and OpenAI's models and Codex arriving on Google Cloud.

What it means

For Anthropic, Duet is a move from a chat assistant to an agent that works on its own schedule. That model needs a lot of computing power, because background checks and long tasks use many model turns. This is why Anthropic ties the app to existing usage limits instead of offering unlimited background work. The cost of running AI at scale is a live issue for the company, as Bullquill reported in its look at Anthropic's reported first-half loss.

For small businesses, the appeal is clear: chasing invoices, preparing payroll and spotting missed messages are routine but time-consuming jobs. The risk is also clear. An agent that can pay and sign needs strong controls, and Anthropic's answer is to require approval for every sensitive step and to block securities trading and money transfers outright. Payments through tools such as Link and Shop Pay are part of a wider push toward agent-led commerce, a theme also seen in recent funding for agentic payment start-ups.

Many questions remain open, including when Duet will reach users outside the US, when Team and Enterprise customers will get access, and how well it performs once more testers use it. Readers who want to understand how such agents call tools can start with Bullquill's guide on building an AI agent with tool calling in Python.

Sources

• Anthropic: Introducing Claude Duet (October 21, 2026)
• Anthropic: Claude Duet product page
• TechCrunch: Claude Duet app is Anthropic's bet on agents
• Digital Trends: Anthropic unveils Claude Duet
• Claude (@claudeai) on X: research preview announcement