Anthropic has launched Claude Duet, a new proactive AI agent currently available in a research preview for Claude Pro and Max subscribers in the United States. Unlike the standard Claude interface, which typically waits for a user to initiate a conversation, Claude Duet is designed to 'come to' the user. It monitors connected accounts—including tools like Shopify, CRM systems, and payroll software—and proactively reaches out only when it identifies a task that requires human attention or a decision.
This image serves as a primary visual for the introduction of Claude Duet as a proactive agent.
Technically, Claude Duet operates on a virtual machine in the cloud, allowing it to perform multi-step tasks such as building dashboards, drafting financial forecasts, or preparing payroll even while the user's local device is offline. It is powered by the Opus 5.5 model and integrates with over 6,000 connectors and plugins. To balance performance and cost, the agent checks connected accounts every few hours rather than continuously, and it uses a dedicated classifier to ensure that sensitive information, such as credit card numbers or government IDs, is not shared without explicit instructions.
The image illustrates the wide range of connectors and plugins supported by the agent.
While the tool offers significant automation for administrative workflows, it includes specific limitations inherent to its research preview status. Anthropic notes that the agent may occasionally misread threads, miscalibrate the urgency of a notification, or miss updates that occur in the brief window between its periodic check-ins. Users retain control over sensitive operations, as the system is designed to request explicit approval before performing actions like sending messages or processing payments. The agent's primary goal is to act as a 'teammate' that handles background work, such as finding unpaid invoices or scheduling meetings, while surfacing only the most critical items for the user's review.