Claude Duet Review: Agent App Features, Pricing, and Integration Options for Professionals

Claude Duet review readers want concrete details on how Anthropic’s new agent app handles daily professional tasks. Launched October 21, 2026 as a research preview, the app runs background monitoring, sends periodic briefs, and creates editable outputs from connected services while requiring approval for any sensitive steps.

The service works through a single persistent thread that carries memory and skills from the main Claude account. It operates in its own cloud environment so work continues after the device powers off, and it produces files such as documents, decks, and dashboards without constant user input.

Claude Duet Features, Integrations, and Pricing

Key functions include morning briefs, checks every few hours, and automatic creation of separate task threads. Outputs appear as editable files ready for further editing. The app connects to more than 6,000 Marketplace connectors covering Gmail, Slack, Google Calendar and Drive, QuickBooks, Stripe, HubSpot, DocuSign, LinkedIn, Canva, and Shopify. Microsoft 365 integration is not yet available. Payments can be completed through Stripe Link or Shop Pay using virtual cards.

Access requires a Claude Pro or Max subscription and entry through a waitlist. The preview remains limited to the United States, with Team and Enterprise options still in planning. Background operations draw from the same five-hour session and weekly limits that apply to the base plan. One full limit reset is granted upon admission and remains usable until November 30, 2026.

Pro costs $20 per month or $17 per month when paid annually. Max starts at $100 per month. No additional fees apply during the preview period, though heavy background use can exhaust the shared quotas faster than standard chat sessions.

Security Controls, User Reports, and Limitations

Every message, payment, signature, or deletion requires explicit user approval. The system never executes securities trades or transfers between accounts. Credentials are collected through separate forms so the model never receives passwords or tokens directly. A dedicated classifier blocks attempts to request credit card numbers or government IDs outright, while health or financial data handling occurs only on direct instruction. The preview passed red-team testing and includes an active bug bounty program.

Early user feedback from sites such as Every.to and posts on X highlights strong performance on multitasking, context carryover, and complex task outputs. Some reviewers note consistent voice interaction and useful proactivity, while others mention occasional misjudged urgency, slower navigation compared with competing agents, and variable latency. Reports of coding quality remain positive, yet several users cite gaps in full autonomy and reliability during longer background runs.

Limitations include fixed use of Opus 5.5 with no model selection, absence of desktop native apps, and no browser visibility or takeover controls. The preview can misread thread context or over- or under-estimate task priority. Team features and multiplayer editing are not yet implemented, and performance on personal errands such as ticket booking stays weaker than on professional workflows.

Competitor Comparison and Final Verdict

Compared with similar agent tools, Claude Duet offers deeper Marketplace connector access and explicit approval gates that some alternatives lack. Its background VM continues work after the device is off, a feature not matched in every competing product. Limits tied to the base subscription can constrain heavy users more than flat-rate competitors, and the US-only rollout plus lack of Microsoft 365 support narrow its reach versus broader platforms.

The preview continues with planned iterative updates and wider geographic and plan availability expected later. Current constraints on autonomy, model choice, and platform coverage remain the main trade-offs for users evaluating the service against established alternatives.