Anthropic has unveiled a research preview of Claude Duet, an always-on personal agent powered by Claude Opus 5.5. Designed to compete with autonomous offerings like SpaceXAI’s Grok Bot and OpenAI’s Dots, the standalone app proactively manages ongoing tasks without requiring a continuous open chat session. Currently available via waitlist to Pro and Max subscribers in the US across iOS, Android, and web, Duet runs within isolated virtual machines in the cloud so user devices do not need to remain powered on.
Duet integrates with services like Gmail, Slack, Stripe, and Shopify using over 6,000 marketplace connectors to manage scheduling, financial forecasts, and payroll preparation. To maintain security, the agent operates under a human-in-the-loop framework where user approval is explicitly required before sending messages, signing documents, or completing payments via virtual cards. Anthropic noted that early limitations include periodic background check intervals to preserve subscription limits and reduced accuracy on personal errands compared to core workplace tasks.
Why it matters
Shift toward proactive, continuous agent execution changes how professionals delegate asynchronous administrative, financial, and operational workflows.
Demonstrates mandatory safety architecture for commercial agents, requiring explicit human approval for sensitive actions and externalized credential handling.
Highlights infrastructure trade-offs, as periodic background checking is necessary to prevent users from rapidly exhausting API and compute subscription limits.
Source: the-decoder.com



