Summary
Anthropic has opened a waitlist-based research preview of Claude Duet, an AI agent for work that can monitor connected apps and carry out tasks in the background. The preview starts with US Pro and Max subscribers, whose existing Claude usage limits also apply to Duet.
Anthropic opened a waitlist-based research preview of Claude Duet on October 21, 2026. The work-focused AI agent can monitor connected apps, prepare information and carry out tasks in the background, with user approval required before it takes consequential actions. The standalone app is available on iOS and Android, and Duet can also be used on claude.ai.
The preview is starting with Pro and Max subscribers in the United States. Duet is included in those plans and draws on the same usage limits as the rest of Claude. Anthropic says the product is designed for professionals and small-business owners managing multiple ongoing tasks.
A work agent that follows tasks through
Duet is built around Claude Opus 5.5, according to Anthropic. It carries over the user’s Claude memory, skills and connected apps, and provides one ongoing thread for typed or voice requests. Anthropic describes it as proactive: it can send a morning brief about important items across connected apps, then check those accounts every few hours and message the user when something needs attention.
For tasks, Duet uses a virtual machine with its own browser, so it can continue working when the user’s phone or laptop is off. A user might ask it to find missed Gmail or Slack messages, prepare invoice reminders using QuickBooks or Stripe, or draft a follow-up using information in HubSpot. Anthropic says Duet can draw on more than 6,000 connectors and plugins in the Claude Marketplace.
The product is designed to prepare work independently but pause for approval before sending a message, making a payment, signing a document or deleting something. Some actions remain prohibited even with approval, including trading securities and moving money between accounts. For purchases, Anthropic says Duet can use Link or Shopify’s Shop Pay; Link requires the user to approve the spend and uses a virtual card.
Safety controls and early-preview limits
Anthropic says Duet does not directly see passwords or sign-in tokens. The company also describes a new classifier and safety prompt that checks messages for sensitive information. It is designed to block requests to share the strictest category, including credit-card numbers and government identification; health or personal-finance information can be sent only when the user has explicitly instructed Duet to do so. Anthropic says its security work includes red-team testing, an external penetration test and coverage under its public bug-bounty programme.
The preview is an early version. Anthropic says Duet may misread a conversation or misjudge what is urgent, and that it is still tuning when the agent sends notifications. Its core features are built for work rather than personal errands, and users cannot yet view its browser. Background checks and larger tasks count towards the shared Claude plan limits, with tasks such as building a financial model requiring more model turns and background work.