Anthropic launches Claude Duet for proactive AI task management

A proactive approach to AI

Anthropic has introduced Claude Duet, a new application where the AI functions as a proactive agent rather than a standard chatbot. Instead of waiting for a user to type a prompt, the agent is designed to monitor connected applications and alert the user when attention is needed. It aims to help professionals and small business owners manage complex workflows more efficiently.

The service is currently available as a research preview for subscribers of the Pro and Max plans in the United States. Users can join a waitlist to gain access to the tool, which is available as a standalone app for iOS and Android as well as through the claude.ai website. Anthropic has not yet announced a timeline for expanding the release to other countries.

The agent is powered by the Claude Opus 5.5 model and integrates with a user's existing Claude account. This means it inherits the memory, skills, and app connections already established by the user. While it is included in the Pro and Max tiers during the preview, it shares the same usage limits as standard interactions.

Background tasks and automation

One of the primary features of Claude Duet is its ability to perform work in the background. Every morning, the agent sends a summary of important events from the user's connected apps. Throughout the day, it checks these accounts every few hours and reaches out only when a specific action or piece of information requires human input.

Users can assign tasks via text or voice commands, ranging from scheduling appointments to creating dashboards and financial forecasts. To achieve this, the agent operates within a virtual machine that includes its own web browser. This architecture allows the AI to continue working on tasks even when the user's smartphone or laptop is turned off.

The system supports a wide array of integrations through the Claude Marketplace. There are more than 6,000 connectors and plugins available, allowing the agent to interact with popular platforms like Gmail, Slack, Stripe, HubSpot, and DocuSign. This connectivity allows the agent to move beyond simple text generation into functional task execution.

Safety and user permissions

Anthropic has implemented a system where the agent must ask for permission before performing sensitive actions. Before the AI can send a message, trigger a payment, sign a document, or delete any data, it must show the planned action to the user for approval. Certain high-risk actions, such as trading securities or transferring funds between accounts, are strictly excluded from these automated permissions.

For authorized purchases, the agent utilizes specific digital wallets like Stripe's digital wallet with virtual cards or Shopify's Shop Pay. To maintain security, the agent does not have direct access to passwords or login tokens. It also uses a dedicated classifier to prevent the leakage of confidential information, such as credit card numbers or identification data.

The company has also applied research regarding browser control to protect against prompt injections. This is a type of attack where a user tries to trick an AI into performing unauthorized actions. The tool was tested by internal red teams and an external firm, and it is currently part of the public bug bounty program.

Current limitations and context

Despite its capabilities, Claude Duet is still in an early stage of development. Anthropic noted that the agent might still misinterpret or misread certain threads, which is a critical area for improvement. Some specific types of tasks, such as booking tickets, are currently handled less effectively by the agent than other routine workflows.

Users should also be aware that complex tasks and deep background checks consume more of the plan's usage limits than standard conversations. This means that heavy automation might exhaust the available capacity for the day faster than typical chat interactions. The company is positioning this as a tool for specific professional needs rather than a general-purpose replacement for all human tasks.

The launch of Claude Duet follows a trend of several companies introducing persistent AI agents. Other notable examples include Meta's Muse, OpenAI's dots, and SpaceX Super Intelligence's Grok Bot. Anthropic's move signals a growing competition in the space of autonomous agents that can navigate web apps and manage multi-step workflows independently.

What to expect next

As the research preview continues, Anthropic will likely refine the agent's ability to handle complex, multi-step instructions. The company's focus on safety and permission-based actions suggests that they want to build trust with business users who handle sensitive data. We can expect more connectors to be added to the marketplace to broaden the agent's reach across different industries.

The success of Claude Duet will depend on how well it manages the balance between being proactive and being intrusive. If the agent can successfully reduce the 'busy work' of checking multiple apps without overwhelming the user with notifications, it could become a standard tool for remote workers. For now, the focus remains on gathering feedback from the initial group of US-based Pro and Max subscribers.

Future updates will likely address the current weaknesses in specialized tasks like ticket booking. As the underlying models improve, the agent's ability to understand nuance and context will sharpen. This evolution will determine whether Claude Duet becomes a primary assistant for daily operations or remains a niche tool for specific automated workflows.