Anthropic takes on Muse with Claude Duet, its new AI agent

Your AI assistant can now proactively manage tasks across your connected apps and take action with your approval.

Anthropic has launched Claude Duet, an AI agent designed to compete with Meta's Muse and OpenAI's offerings. Available as a research preview on iOS, Android, and the web, Claude Duet is built on Claude Opus 5.5. It functions by providing daily briefings on connected apps and notifying users of critical updates. The agent can handle tasks ranging from scheduling meetings to complex operations like building dashboards, generating forecasts, and prototyping web applications. It integrates with Claude's memory, skills, and connected apps, utilizing a virtual machine with a browser for task execution. To ensure safety, Claude Duet includes a dedicated classifier and safety system to prevent the sharing of highly sensitive information, and it prompts users for approval before executing significant actions. Anthropic notes that this is an early version, with limitations in handling personal errands and potential for occasional misinterpretations or miscalibrations of urgency. Access is currently restricted to Pro and Max subscribers in the US via a waitlist.

Anthropic's introduction of Claude Duet marks a significant step in the evolution of AI agents capable of proactive task management and execution across multiple applications. By integrating directly with users' workflows and offering a degree of autonomy coupled with human oversight, Duet signals a shift towards more integrated and utility-focused AI tools for professionals. This move challenges the current landscape where AI is often reactive, pushing the competitive envelope for agents that can truly augment daily productivity by handling routine and complex tasks.

The robust safety features and user approval mechanisms implemented in Claude Duet address critical concerns surrounding AI agents acting on sensitive data. This approach, which prioritizes user control and data security, could set a new standard for the development of such tools, especially as they become more deeply embedded in business and personal operations. It also suggests that future AI agents will need to balance powerful capabilities with stringent safeguards to gain widespread trust and adoption in the market.