Anthropic launches Claude Duet proactive AI agent app

Anthropic's Claude Duet represents a shift from a chat assistant that responds on request to an agent that initiates contact. The company says it built the app for professionals and small-business owners who cannot keep track of all their work, and the design reflects that: a short morning brief, monitoring of connected apps, and messages only when attention is needed. The preview runs on Anthropic's Claude Opus 5.5 model and on a virtual machine with its own browser, which lets work continue while the user's own devices are off.

The app draws on more than 6,000 connectors and plugins from the Claude Marketplace, including Gmail, Slack, QuickBooks, Stripe and Docusign. Anthropic has set explicit boundaries around sensitive actions: Duet asks for approval before sending messages, making payments, signing documents or deleting, and it will not trade securities or move funds between accounts even with approval. It also refuses to share highly sensitive information such as credit card numbers and government ID numbers, and sends health or personal financial information only when the user explicitly instructs it to.

Anthropic is entering a market where Meta and OpenAI are already present which notes common design elements with Meta's Muse and OpenAI's Dots and points to pricing and usage limits as areas to watch. The research preview is limited to US Pro and Max subscribers on a waitlist, so broad availability remains to be announced.