Anthropic launched the all-weather AI agent Claude Duet, powered by Opus 5.5

Anthropic is extending AI competition from model capabilities to task execution.

On Wednesday local time, the company launched a new application called Claude Duet, positioning it as an AI agent capable of running continuously in the background, actively identifying user needs, and executing work tasks, further expanding the product form of Claude.

Unlike traditional conversational assistants that wait for user commands, Claude Duet can aggregate important information from related applications, proactively send notifications when user attention is needed, and handle various tasks ranging from meeting scheduling to financial forecasting and web application development.

However, this product is currently still in the research preview stage and is only available to Claude Pro and Max subscribers in the U.S. through a waitlist. The limited availability means that Claude Duet has not yet entered a full competitive phase aimed at a broader user base.

Anthropic stated that Claude Duet is developed based on its flagship model Claude Opus 5.5 and can continue executing tasks after the user leaves the device, but there are still limitations in assessing the urgency of information and handling certain personal affairs. As Meta and OpenAI are also advancing proactive AI agents, Anthropic is attempting to translate model capabilities into user services.

From Responding to Commands to Proactive Execution

According to the official announcement released by Anthropic, Claude Duet is designed for professionals and small business owners with heavy workloads.

The application aggregates important matters from related applications every morning and continuously checks relevant information throughout the day, proactively sending reminders when it identifies matters that require user attention.

Users can also directly delegate tasks. In addition to routine administrative work such as scheduling meetings, Claude Duet can also build dashboards, create financial or sales forecasts, and develop prototypes and web applications using multiple tools.

The core change is that users do not have to remain in the conversational interface to issue commands step by step; the agent can advance tasks in the background and notify users upon completion or when review is needed.

Anthropic claims that Claude Duet inherits the memory, skills, and connected applications from the user's Claude account, reducing the need for repetitive configuration. The product offers standalone iOS and Android applications and can also be used via the web.

Based on Opus 5.5, Connecting Over 6,000 Tools

Claude Duet is built on Claude Opus 5.5 and is equipped with an independent virtual machine and browser environment, allowing it to continue working even after the user's computer or phone is turned off.

According to Anthropic's introduction, the product can call upon over 6,000 connectors and plugins in the Claude Marketplace, covering work tools such as Gmail and Slack, as well as business services like QuickBooks and Stripe.

These connectors enable the agent to combine information retrieval with subsequent actions. For example, it can check for unprocessed emails and unpaid invoices, draft reminder messages, prepare payroll processing content, and assist in creating marketing materials, managing online stores, or preparing contract documents.

Key steps such as payments and information sending are still subject to an approval mechanism. Anthropic stated that before sending messages, making payments, signing documents, or deleting information, Claude Duet will present the proposed actions to the user and wait for approval.

The product also has dedicated security mechanisms to prevent sensitive information from being improperly disclosed; certain high-risk operations cannot be executed even with user approval.

Research Preview Stage, Capabilities and Usage Scope Still Limited

Claude Duet is currently not a fully open commercial product. Anthropic has initially opened the waitlist application to Pro and Max subscribers in the U.S. During the research preview period, the agent shares subscription usage limits with other Claude functionalities, and background checks and complex tasks will also consume usage limits.

Anthropic also acknowledges that the product may still misinterpret conversational content or incorrectly assess the urgency of matters at this stage. Since the product is primarily developed around work scenarios, its ability to handle personal affairs such as booking tickets is relatively weak, and users currently cannot directly view its browser operation process.

Therefore, the actual value of Claude Duet still depends on whether it can establish a stable workflow between continuously acquiring information, proactively assessing task priorities, and executing actions. For businesses and professional users, background automation is expected to reduce information filtering and repetitive tasks, but misjudgments and sensitive information handling remain issues to be addressed during use.

AI Agent Competition Further Extends

As Anthropic launches Claude Duet, OpenAI and Meta are also advancing AI agents capable of executing tasks on behalf of users. OpenAI previously launched Dots, while Meta introduced Muse, with both products attempting to move AI from answering questions to completing tasks across applications.

The Verge views Claude Duet as a competitor to Meta Muse and OpenAI Dots, noting that Duet integrates memory, skills, and application connection capabilities within the Claude ecosystem, executing work tasks through virtual machines and browsers.

Business Insider points out that Claude Duet is still in the waitlist open stage and cannot immediately compete directly with a broader market and other agent products.

For Anthropic, Claude Duet is not only an extension of the Claude application form but also signifies the company's exploration of how to keep the model continuously involved in users' daily work, rather than just providing one-time Q&A and content generation services. Its subsequent development will depend on the reliability of task execution, user acceptance of proactive services, and whether the product can gradually expand its usage scope from a limited research preview.

Risk Warning and Disclaimer

The market has risks, and investment should be cautious. This article does not constitute personal investment advice and has not considered individual users' specific investment goals, financial situations, or needs. Users should consider whether any opinions, views, or conclusions in this article align with their specific circumstances. Investing based on this is at one's own risk.