Anthropic launches Claude Duet, a background AI agent for professionals

Anthropic launched Claude Duet on October 21, 2026, a proactive AI agent app that works in the background to handle tasks, monitor connected accounts, and surface what needs your attention — even when you are not actively using it. The app is available by waitlist for US Pro and Max subscribers on iOS, Android, and claude.ai.

Unlike conventional AI assistants that wait for a question, Claude Duet checks your connected apps every few hours and sends a message only when something actually needs you. Each morning it delivers a brief of important items from your email, calendar, and other connected services.

What Claude Duet can do

The agent handles tasks ranging from scheduling meetings and building financial models to writing code, generating dashboards, and creating full web apps. It runs inside its own virtual machine, so your laptop or phone does not need to stay on for longer jobs to finish.

Claude Duet connects to more than 6,000 services through existing integrations, including Gmail, Slack, QuickBooks, Stripe, Gusto, HubSpot, and Shopify. It can also make purchases using Stripe’s Link wallet, but only after you approve each transaction.

Human approval required for consequential actions

Anthropic built explicit approval requirements into every action that carries consequence: sending messages, making payments, signing documents, or deleting data all require user confirmation before the agent proceeds. The system also blocks the agent from accessing card numbers, government IDs, and passwords.

Memory, skills, and app connections already set up in your Claude account carry over into Duet automatically. The app runs separately from the main Claude interface on iOS and Android and on claude.ai.

Current limitations

Anthropic notes that Claude Duet may misjudge urgency, handles personal errands such as booking tickets less reliably than professional tasks, and cannot yet show the browser it uses for web-based work. Features including 1Password sign-in and a visual Shopify editor are listed as coming soon with no dates given.

Usage counts against existing Pro or Max plan limits. Background monitoring and complex tasks like financial models consume more of those limits. Anthropic said it will share additional safety details in the coming days as the research preview expands.

The launch comes as AI companies race to move beyond reactive chatbots toward agents that operate continuously in the background. By requiring explicit approval for any action that could have real-world consequences, Anthropic is signaling a cautious approach to autonomy — a notable design choice given the growing debate over how much authority AI agents should have without a human in the loop.