Anthropic has introduced Claude Duet, an AI agent currently available in a research preview. The tool is designed to perform tasks autonomously, such as organizing daily priorities, managing emails, tracking invoices, and developing web applications. It operates on the Claude Opus 5.5 model and is accessible via iOS, Android, and web platforms.
The agent functions through an independent virtual machine and browser, allowing it to continue tasks even when the user is offline. Users can issue commands via text or voice. Upon completion, the system notifies the user or requests approval for final actions. It integrates with over 6,000 connectors and plugins from the Claude Marketplace, including services like Gmail, Slack, and Stripe.
Security features include a system that avoids direct access to user passwords or login tokens. The agent requires explicit user approval before sending messages, making payments, or signing documents. Additionally, a classifier is in place to block the leakage of sensitive data like credit card numbers or government IDs.
The preview is currently limited to Pro and Max subscribers in the United States, with usage limits shared with other Claude features. Anthropic acknowledges that the software is in an early stage, noting potential issues with task prioritization and the inability to display the browser operation process to the user.
