Anthropic Introduces Personal AI Agent Claude Duet

Anthropic has introduced Claude Duet, a standalone app where Claude acts as a proactive agent. Instead of waiting for individual prompts for every action, the new tool independently identifies tasks that require a user’s attention, executes assigned tasks in connected apps, and keeps working in the background.

Anthropic says it is primarily aimed at entrepreneurs, professionals, and small business owners who handle a high volume of routine and analytical tasks on a daily basis.

How Claude Duet Works in Practice

Claude Duet runs within a virtual machine equipped with its own browser. This means users don’t need to keep their laptop or smartphone turned on to complete tasks. They simply describe a task in a short text or voice message, and the AI ​automatically selects the necessary tools and steps.

Powered by the Claude Opus 5.5 language model, the agent can perform a wide range of tasks. For instance, it can send a brief daily report each morning covering key events from connected applications, build analytical dashboards that consolidate data from dozens of tools, generate financial and sales forecasts, handle administrative tasks such as scheduling meetings and rescheduling them in the event of conflicts, and even create prototypes and develop web apps.

Users receive a notification once a task is completed or requires their review. Throughout the day, the agent checks a user’s accounts and reaches out only when something genuinely requires a response. Checks occur every few hours, not continuously, because continuous monitoring would quickly exhaust usage limits.

Duet carries over the memory, skills, and connected apps from a user’s Claude account, so setup takes very little work. In practice, this means Duet already knows what Claude has learned about you, the saved instructions you’ve set up for specific tasks, and the apps you’ve linked.

According to the developers, the primary focus is on judgment - the agent determines what requires a user’s immediate attention, what can wait, and when it is appropriate to ask for their input.

Services That Can Be Connected

According to Anthropic, Claude Duet supports over 6,000 connectors and plugins from the Claude Marketplace. They enable integration with work, financial, and communication platforms.

The agent can find missed messages in Gmail or Slack, create LinkedIn posts, prepare agreements via DocuSign, set reminders for unpaid invoices in QuickBooks or Stripe, generate payroll reports in Gusto, draft follow-ups based on deal history in HubSpot, set up and manage Shopify stores, and more. Any connected application can be disconnected at any time.

Human Oversight and Security Measures

Duet can independently read content, draft messages, and gather materials. However, before sending a message, signing a document, making a payment, or deleting any data, it displays its intended action and waits for user confirmation. And there are actions, such as trading securities or transferring funds between accounts, that are prohibited even with user consent.

Anthropic says Duet will soon be able to sign in to sites using logins saved in 1Password, with the user’s permission. The company says the agent never directly sees passwords or sign-in tokens, because Duet’s handling of credentials follows the same design as Claude Managed Agents. Additionally, security mechanisms restrict the transmission of sensitive information. Specifically, the system is prohibited from transmitting bank card numbers or government-issued identification details. It may only access information regarding health or personal finances upon the user's explicit instruction.

Purchases are made via Link by Stripe, using virtual cards. The agent specifies the item, Link asks the user to confirm the expense, and the payment is then processed with a virtual card, ensuring the actual card number is never shared. Shop Pay by Shopify is also supported.

The company says it continues to develop safeguards against prompt injection, an attack where someone hides instructions in an email, web page, or document that the agent reads, hoping to trick it into acting against the user’s wishes.

At the same time, Anthropic acknowledges that this research preview is prone to errors. For instance, the agent may occasionally misjudge the urgency of a task or misinterpret the context of previous conversations. Certain personal tasks, such as booking tickets, are currently handled less efficiently. Additionally, for now, users cannot watch what the agent is doing in the browser.

Who Can Try It and When

A research preview launched on October 21, 2026, is limited to Pro and Max users in the US.

Access is limited for now. To test the agent, you first need to join the waiting list.

Eligible users can try it via the claude.ai website and dedicated iOS and Android apps.

There is no additional fee for using the agent - Duet is included in Pro and Max subscription plans and shares usage limits with the rest of the Claude ecosystem.

Why this launch matters

Claude Duet puts Anthropic into direct competition with other companies developing proactive personal AI agents, including OpenAI’s Dots and Meta’s Muse, both launched recently.

The broader shift is from AI as a tool you operate to AI as an assistant that can take responsibility for parts of a workflow. That could change how companies approach automation: instead of building a separate automation for every task, they may increasingly delegate work to agents that can use existing business applications.

The important question that remains for businesses is how reliably these agents can execute tasks, respect permissions, protect sensitive data, and handle exceptions without constant human supervision.