Anthropic unveils Claude Duet, but keeps its work agent tightly limited

Anthropic has introduced Claude Duet, a work-focused AI agent meant to do more than answer prompts: it watches the apps a user connects, sends a daily briefing, and can keep working on tasks such as scheduling, dashboards, or forecasts after the user steps away. That makes it a bet on proactive software—an assistant that identifies what may need attention instead of waiting for every instruction. But this is an early research preview, not a broad launch: access begins through a waitlist for U.S. Claude Pro and Max subscribers, and Anthropic says the system can still misjudge urgency or misunderstand a thread. The restraint matters because the product’s promise depends on access to work information and some freedom to act; Anthropic says it asks for approval before acting on a user’s behalf.