Anthropic launches Claude Duet, proactive AI agent
Anthropic Debuts Proactive Agent for Task Completion
Anthropicintroduced Claude Duet on October 21, 2026, a standalone AI-agent application designed to work proactively on behalf of users Anthropic. The product is being released as a waitlisted research preview, marking the company's entry into the expanding market for autonomous agents that handle routine professional work without constant human direction.
Claude Duet is built on Anthropic's Opus 5.5 model and operates as a connected agent that monitors linked applications to surface items requiring attention and complete requested tasks Reuters via Business Insider. The agent can prepare daily briefs summarizing key information and continues working while the user is away, distinguishing it from passive chatbot interfaces that require explicit prompts.
Safety Checks Built Into User Approval
Anthropicsaid the agent "checks with you before it acts on your behalf" CNBC. The company specified that approval is required for consequential actions—sending messages, making payments, signing documents, or deleting files—while permitting routine information gathering and preparation without interruption MacRumors.
The approval safeguard represents a deliberate design choice to balance autonomy with user control. By requiring consent for high-stakes actions while allowing the agent to complete lower-risk tasks independently, Anthropic aims to reduce friction while preserving oversight over decisions that commit resources or alter external systems.
Available to U.S. Subscribers, Expandable Feature Set
Claude Duet is accessible to U.S. subscribers on Anthropic's Pro and Max plans through Claude's website and standalone mobile applications. The agent can integrate with connected apps to perform specific tasks: scheduling meetings, building financial forecasts, and creating web applications Heise. The research-preview status allows Anthropic to gather real-world usage data and refine the agent's decision-making before wider rollout.
The launch positions Anthropic in direct competition with similar agentic products from OpenAI and Meta, each pursuing different approaches to autonomous task execution. Anthropic's emphasis on approval gates for consequential actions reflects the company's stated focus on safety and user agency—a distinction the company is marketing as a competitive advantage in an increasingly crowded agent landscape.
Users interested in testing Claude Duet can join the waitlist through Anthropic's official channels, with availability expanding as the research preview progresses.