Anthropic Launches Claude Duet, a Proactive AI Agent

Anthropic has unveiled Claude Duet, an AI agent aimed at proactive assistance similar to Meta's Muse and OpenAI's Dots agents. Unlike those agents, Duet initiates contact without being prompted. It interfaces with users' apps, highlighting tasks requiring attention, and can operate in the cloud without a local computer. Claude Duet functions via a virtual machine with browser access to perform tasks.

Designed for professionals and small business owners burdened with too many tasks, Duet is powered by Anthropic's Opus 5.5. Each day, it delivers a concise summary of key items from linked apps, followed by hourly checks and alerts for issues needing attention. While capable of handling simple or complex tasks such as web app development, Duet operates under user approval for actions like sending messages, making payments, signing documents, or deleting files.

Claude Duet is adept at retaining user context, skills, and connected apps within a Claude account, minimizing setup requirements. Conversations occur in a continuous thread, contrasting with separate interactions seen in other agents. In terms of security, Duet does not access passwords or logins when granted access to accounts. It also employs a dedicated classifier and safety system prompt to scrutinize each message, preventing leaks of sensitive data.

Currently, Claude Duet is in a research preview, accessible through a waitlist, and is limited to Pro and Max subscribers in the United States. It is compatible with web interfaces or a standalone Duet iOS app. Duet utilizes existing Claude plan limits, with background checks contributing to these limits.