Claude Duet Review 2026: Agent Capabilities, Waitlist Limits and Comparison to Anthropic Assistants
Claude Duet runs background tasks across connected apps with user approvals required for actions. The research preview is limited to US Pro and Max subscribers.
Claude Duet Research Preview launched on October 21, 2026, as Anthropic’s proactive agent that monitors apps and runs tasks on a cloud virtual machine. It sends daily briefs and handles work such as invoice processing, dashboard creation, and scheduling without constant user input. The tool requires explicit approvals for any external actions and shares session limits with existing Claude subscriptions.
Claude Duet Capabilities and Daily Workflow
The agent connects to more than 6,000 Marketplace plugins including Gmail, Slack, QuickBooks, Stripe, Gusto, HubSpot, DocuSign, LinkedIn, Canva, and Shopify. It checks accounts every few hours and only sends notifications when attention is required. Tasks run in the background on a dedicated cloud VM even when the user’s device is powered off, with results appearing in dedicated Make and Tasks tabs.
Users communicate through a single ongoing thread or topic channels via text or voice. Memory, skills, and connectors carry over from the main Claude account. The system supports purchases through Stripe Link or Shop Pay, and 1Password sign-in is planned. Hard blocks prevent trading securities, moving funds between accounts, or accessing third-party accounts regardless of user approval.
Explicit approval is required before sending emails, making payments, or signing documents. Classifiers scan for sensitive data leaks, and the agent cannot view passwords or tokens directly. Early tester examples include to-do list management, forecasting models, and simple web app prototypes for small business operations.
Access Requirements, Pricing and Usage Limits
Entry is restricted to a waitlist for US Claude Pro and Max subscribers. Team and Enterprise plans are not yet supported. The preview runs exclusively on Claude Opus 5.5 with its 1 million token context window and adaptive reasoning. No separate pricing tier exists; the agent is included in Pro plans at approximately $20 per month or $17 per month billed annually, and Max plans starting at $100 per month.
Background monitoring and complex tasks consume the standard five-hour session and weekly usage allowances. New participants receive one full limit reset valid through November 30, 2026. The agent performs checks at fixed intervals, which can result in missed urgent items. It cannot access local files, view browser sessions, or initiate phone calls during the preview phase.
| Feature | Claude Duet Preview | Standard Claude Chat | Claude Code Tools |
|---|---|---|---|
| Operation mode | Proactive, always-on VM | Reactive, session-based | Reactive, coding-focused |
| Background execution | Yes, persistent tasks | No | Limited to active sessions |
| App connectors | 6,000+ Marketplace | Basic integrations | Code repository focus |
| Approval requirements | Explicit for all actions | Not applicable | Project-specific |
| Access method | Waitlist, US only | Immediate for subscribers | Immediate for subscribers |
Future waves of access are planned, with adjustments based on tester feedback around urgency detection and additional integrations. No timeline exists for non-US expansion or general availability.
Frequently asked questions
How do I join the Claude Duet waitlist?
US users with active Claude Pro or Max subscriptions can join the waitlist through their account settings on claude.com. Team and Enterprise accounts are not eligible during the current preview phase.
Does Claude Duet use extra compute beyond standard limits?
Background tasks and complex operations count against the same five-hour session and weekly limits as regular Claude use. New preview participants receive one reset valid until November 30, 2026.
Can Claude Duet make payments without approval?
No. Every payment, email, or document action requires explicit user confirmation. Trading securities and inter-account transfers remain blocked even with approval.
What model powers Claude Duet in the preview?
The agent runs on Claude Opus 5.5 with 1 million token context and adaptive reasoning. Users cannot select alternative models during this stage.
Comments 7