09:31Machine-translated from Azerbaijani by AIRead in Azerbaijani
Anthropic Unveils Claude Duet AI Agent
Anthropic has unveiled a new artificial intelligence agent called Claude Duet that can work even when the user is not interacting with it. The service, currently in testing, is being made available in the US market to Pro and Max subscribers through a waitlist. Invitations will be sent out gradually. Anthropic developed the agent for busy professionals and small business owners. The service competes with OpenAI’s Dots and Meta’s Muse agents.
Claude Duet accepts written and voice instructions and monitors connected email, calendars, and other apps. The agent can prepare a brief summary of important matters every morning, draft replies, and provide updates throughout the day about tasks that require a decision. It uses the same memory, skills, and connections as the standard Claude and can also search previous conversations. Channels created for separate topics share the same memory. The agent can schedule meetings and prepare research documents, presentations, financial models, and sales forecasts. It can also use dozens of tools to build dashboards, create prototypes, and work on web apps. The results of major tasks are added as files to the Make section for review and editing. The service runs on a separate cloud computer and browser. As a result, a task continues even if the app is closed or the user’s computer is turned off. Websites without a direct connection open in Claude Duet’s cloud browser. The service works with plugins, including an option designed for small businesses, and can recommend a plugin suited to the task at hand. The Stripe connection checks unpaid invoices and drafts payment reminders. Purchases can be made with Link and Shop Pay after the user gives approval. Ongoing tasks and those requiring attention can be tracked in the Tasks section. Claude Duet sends an update to the main conversation when it completes a task or needs a decision. However, only the user can mark a task as completed. Sending messages, making payments, signing documents, and deleting data require the user’s permission. This restriction is intended to prevent the agent from performing sensitive actions without the user’s knowledge.
Anthropic says the agent can still make mistakes. The service may misunderstand correspondence, fail to accurately assess how important a task is, or miss an update that arrives between checks. All work performed in the background counts against the user’s current limit. Major tasks such as financial models and presentations may use up the limit faster because they require more steps. There is no separate charge for Claude Duet, but the service counts toward the limits of the Pro and Max plans. Team and Enterprise plans are not yet supported, but support is planned for the future. The agent is powered by the Claude Opus 5.5 model. Users who receive an invitation can access the agent through the iOS and Android apps, as well as the claude.ai website.
Claude Duet accepts written and voice instructions and monitors connected email, calendars, and other apps. The agent can prepare a brief summary of important matters every morning, draft replies, and provide updates throughout the day about tasks that require a decision. It uses the same memory, skills, and connections as the standard Claude and can also search previous conversations. Channels created for separate topics share the same memory. The agent can schedule meetings and prepare research documents, presentations, financial models, and sales forecasts. It can also use dozens of tools to build dashboards, create prototypes, and work on web apps. The results of major tasks are added as files to the Make section for review and editing. The service runs on a separate cloud computer and browser. As a result, a task continues even if the app is closed or the user’s computer is turned off. Websites without a direct connection open in Claude Duet’s cloud browser. The service works with plugins, including an option designed for small businesses, and can recommend a plugin suited to the task at hand. The Stripe connection checks unpaid invoices and drafts payment reminders. Purchases can be made with Link and Shop Pay after the user gives approval. Ongoing tasks and those requiring attention can be tracked in the Tasks section. Claude Duet sends an update to the main conversation when it completes a task or needs a decision. However, only the user can mark a task as completed. Sending messages, making payments, signing documents, and deleting data require the user’s permission. This restriction is intended to prevent the agent from performing sensitive actions without the user’s knowledge. Anthropic says the agent can still make mistakes. The service may misunderstand correspondence, fail to accurately assess how important a task is, or miss an update that arrives between checks. All work performed in the background counts against the user’s current limit. Major tasks such as financial models and presentations may use up the limit faster because they require more steps. There is no separate charge for Claude Duet, but the service counts toward the limits of the Pro and Max plans. Team and Enterprise plans are not yet supported, but support is planned for the future. The agent is powered by the Claude Opus 5.5 model. Users who receive an invitation can access the agent through the iOS and Android apps, as well as the claude.ai website.