A Deep Dive into Claude Duet: Features, Availability, and How Anthropic’s Proactive AI Agents Work
Artificial intelligence (AI) is increasingly entering the mainstream, and AI assistants are playing a pivotal role in this shift. In recent months, the capabilities of AI assistants have evolved beyond simple query responses and content generation; they can now handle tasks that previously required users to switch between multiple applications. Anthropic’s new application, Claude Duet, takes this a step further by enabling the AI assistant to proactively execute actions. It also consolidates a user’s existing Claude context into a dedicated workspace, while maintaining a safeguard that requires user approval for critical actions.
What is Claude Duet?
Claude Duet is a new application from Anthropic that transforms Claude into a proactive AI agent capable of performing tasks across connected applications and services. According to Anthropic, Claude Duet can initiate interactions when it identifies content that may require the user’s attention—a departure from traditional chatbots, which typically wait for a prompt before responding within a conversation.
The application migrates the memory, skills, and connected apps from a user’s existing Claude account, which the company notes reduces the initial setup effort. Furthermore, Claude Duet can continue operating even when the user is away from their computer or mobile device. It runs on a virtual machine equipped with its own browser, allowing it to carry out tasks in the background without requiring the user’s own device to remain powered on.
As Anthropic is still in an early research preview phase, it continues to test its ability to determine which information is significant enough to warrant a user’s attention.
Claude Duet Availability
Anthropic has opened a waitlist for the research preview of Claude Duet. Initially, this feature is available to Claude Pro and Max subscribers in the United States. Interested users can join the waitlist and will receive an email notification once access becomes available.
At present, this application operates independently of the main Claude app. Anthropic has stated that feedback from early users will help inform future improvements regarding the integration of the two applications.
Usage limits for Claude Duet are shared with a user’s existing Claude subscription plan. This means that background analysis also counts toward these limits; furthermore, complex tasks—such as creating financial models or presentations—may require more model interactions and backend processing than standard conversations.
Features and Use Cases of Claude Duet
Claude Duet is designed to facilitate interoperability across applications, track ongoing tasks, and seek user authorization before executing critical actions. It offers proactive assistance and handles background tasks. In short, this AI assistant can periodically check your connected accounts—every few hours, for instance—and notify you if it identifies any issues that may require your attention. You can also specify which types of information you wish to receive and how frequently you would like to get updates.
Once assigned a task, Claude Duet determines which connected applications to utilize and carries out the necessary steps in the background. It then notifies you when a task is complete, or when a decision or review is required on your part.
Claude Duet supports over 6,000 connectors and plugins available via the Claude Marketplace. According to Anthropic, these integrations enable the AI to work in tandem with a range of services used for communication, accounting, recruitment, sales, document management, and online retail.
Furthermore, this AI assistant can assist users in making business-related purchases. It supports Link—Stripe’s digital wallet—and can be used with Shopify’s Shop Pay for online shopping. Anthropic also plans to introduce a feature that, with user permission, will allow for logging into websites using credentials stored in 1Password.
Although Claude Duet is capable of independently reading information, drafting messages, and preparing tasks, Anthropic states that it does not have full authorization to execute all actions. It must first propose a course of action and await human approval before carrying out tasks such as sending messages, making payments, signing documents, or deleting content.
Even with user approval, certain activities remain prohibited—for example, securities trading and the transfer of funds between accounts.
Anthropic notes that Claude Duet is currently in an early version and may occasionally misinterpret the content of a conversation or misjudge the urgency of a situation. The company is actively collecting user feedback to improve how the system handles notifications and prioritizes tasks.