Claude Duet is Anthropic’s latest experiment in moving AI agents beyond waiting for a question. The research preview can proactively identify issues such as a missed customer email, an overdue invoice or a meeting entered at the wrong time, then begin investigating or preparing a response before a user asks. Running on a virtual machine with its own browser and built on Claude Opus 5.5, the agent offers a concrete view of how Anthropic wants assistants to participate in everyday business operations.
The product is available through a separate mobile app for iOS and Android, as well as claude.ai, and is included for individual Pro and Max subscribers during the preview. Anthropic has not announced a Duet API, an enterprise rollout schedule or access for Team and Enterprise subscriptions. That distinction matters for technology leaders because the preview’s business promise arrives before many of the administrative controls associated with organizational deployment.
Proactive Assistance Across Business Systems
Duet produces a daily briefing from connected services and periodically checks more than 6,000 Claude Marketplace connectors and plugins for developments worth surfacing. Users can control which subjects it flags and how frequently it updates them. Text and voice conversations remain persistent, while account memory, skills and existing Claude connections provide the agent with previously supplied context.
Its cloud environment allows work to continue after a user closes a laptop or turns off a phone. Duet determines which applications and steps an assignment requires, performs the work remotely and alerts the user when it finishes or needs a decision. Anthropic’s examples include investigating unpaid invoices in QuickBooks or Stripe, drafting customer outreach from HubSpot information and assembling agreements for review and signature through Docusign.
Other announced workflows cover preparing Gusto payroll using employee hours, bonuses and reimbursements; drafting recruitment listings for review before publication on LinkedIn; developing Canva campaigns; and configuring or maintaining a Shopify store. A storefront design interface with prepublication previews is planned for later. Duet also supports purchases through Stripe Link and Shopify Shop Pay, with Link requesting spending authorization and using a virtual card that conceals the underlying card number. Planned access to credentials stored in 1Password remains forthcoming.
These examples position Duet as an operational layer across finance, customer management, human resources, marketing and commerce. They also make connection permissions consequential. Authority to summarize an invoice is different from authority to initiate a payment, while access to a customer record may carry obligations that an ordinary personal assistant cannot automatically assume.
Approval Boundaries and Security Constraints
Anthropic draws an explicit line between preparing work and taking consequential action. Duet can inspect information and assemble drafts independently, but it must request approval before sending a message, making a purchase, signing an agreement or deleting material. Securities transactions and transfers between financial accounts remain prohibited even when authorized, and users can revoke an application connection.
The company also describes a sensitive-information classifier intended to prevent inappropriate disclosure. Anthropic says the system blocks its most restricted categories, including payment-card numbers and government-issued identifiers, while allowing disclosure of some health and personal financial information after an explicit instruction. Other safeguards include credential isolation, an action-approval classifier and defenses against prompt injection.
These are vendor descriptions rather than proof that an agent will never make an unauthorized disclosure or take an inappropriate action. Anthropic acknowledges that Duet can misunderstand conversations and assign the wrong urgency to an issue. It also cannot yet display its browser to the user, limiting visibility into part of its work. Internal security teams and an outside firm tested the infrastructure, and Anthropic says its public bug bounty covers Duet, but neither fact substitutes for evaluation in a buyer’s own environment.
Why Claude Duet’s Enterprise Test Is Still Ahead
Duet is not the first persistent assistant to promise proactive help, and cloud execution alone does not distinguish it from Claude’s existing products. Claude Code Remote Control connects a phone to a coding session on a user’s computer, while cloud-based Cowork tasks and scheduled jobs can continue after a laptop is shut. Anthropic is also combining chat and Cowork into one Claude experience. Duet’s differentiator is therefore its dedicated app, persistent conversation, routine account checks and judgment about when to interrupt the user.
The competitive landscape is moving quickly. Meta’s Muse promises proactive assistance and action on a dedicated cloud computer, while OpenAI’s Dots is pursuing a similar model. Sensor Tower estimates reported by Yahoo Finance placed Muse at 10.9 million downloads following its September debut and 2.5 million daily active users on October 16. Such early traction does not establish Duet’s performance, but it shows that demand for agents capable of carrying out tasks is already attracting substantial attention.
For businesses, a sensible first evaluation would emphasize bounded assignments with easily checked outputs, such as briefing preparation, invoice-follow-up drafts or meeting coordination. Payroll, purchasing and publishing warrant closer scrutiny of approvals, permissions and auditability. Anthropic has not disclosed task-level costs, quantified productivity improvements or benchmark results, so teams will need to measure completed work, correction time and subscription allowance consumption themselves.
Pro includes audit logs and custom retention controls under Enterprise rather than the individual preview, while Claude Duet’s value will ultimately depend on whether its interventions help more often than they require correction. Claude Duet offers businesses a timely way to test that proposition, but it does not yet provide the enterprise controls, rollout details or comparative evidence needed for an unrestricted operational role.