Claude’s round-the-clock AI agent “Claude Duet” is here too! Powered by Opus 5.5, it proactively tracks tasks, completes work, and even notifies you.

After xAI, Meta, and OpenAI launched their own always-on AI agent tools one after another, it’s finally Anthropic’s turn! It has just officially announced the launch of the new “Claude Duet.” In addition to continuously carrying out tasks you assign in the background without interruption even if you shut down your computer, it will also regularly check the apps you’ve connected, proactively track important matters, and proactively send notifications when a task is completed or when it needs your attention.

More importantly, there’s no extra cost—it’s included directly in the subscription plan (though it shares usage limits with Claude). For now, it’s only available to Claude Pro and Claude Max subscribers in the US.

Anthropic Launches Claude Duet: an AI Agent That Proactively Reaches Out to You and Keeps Working Even When Your Computer Is Off

Claude Duet is a standalone app that supports iOS and Android, and on computers it is used through the claude.ai website. It is powered by Anthropic’s latest Opus 5.5 model:

How is it different from using regular Claude? The simplest way is to see how it accompanies you through a workday.

Every morning when you wake up, Claude Duet has already organized the key points for you and will send a brief summary bringing together the most important items from your connected apps, such as email, calendar, and Slack. You don’t have to open each app one by one to check—just look at this summary first, and you’ll know what to tackle first today.

Then Claude Duet checks your connected accounts every few hours, but it only messages you about things that genuinely need your attention. As for why it doesn’t monitor constantly, Anthropic says continuous checking would use up plan quota too quickly, which is why it’s designed this way. You can also tell it directly what kinds of notifications you want to receive and how often.

All interactions stay in the same continuous conversation thread—no need to start a new conversation. You can type or use voice.

Claude Duet has its own cloud virtual machine and browser, so you don’t need to keep your laptop or phone on. Just describe the task in one sentence, and it will decide which apps to use and what steps to take, run it in the background, and only send a push notification when it’s done or needs your review.

Of course, before sending a message, making a payment, signing a document, or deleting anything, Claude Duet will show you the full list of actions it is about to take and will not act until you approve. In addition, there are some things it won’t do even if you agree, such as buying or selling securities or transferring funds between accounts.

Claude Duet can use more than 6,000 connectors and plugins available on Claude Marketplace. The following are official usage examples:

• Find overlooked Gmail and Slack messages and help with replies.
• Check QuickBooks or Stripe for unpaid invoices and draft payment reminders.
• Prepare and run payroll in Gusto, with hours, bonuses, and reimbursements filled in automatically.
• Draft a recruitment post; once you’ve revised and approved it, post it directly to LinkedIn.
• Prepare the contract and send it for review and signing via Docusign.
• Draft a follow-up email using the contacts and deal records in HubSpot.
• Turn a product launch into a complete set of social media marketing assets in Canva.
• Set up, design, and manage a Shopify store, with a more complete visual editor coming later so you can preview your storefront before going live.

Even more noteworthy is that Claude Duet will directly carry over the memories, Skills, and connected Apps in your Claude account, so the work context you’ve built up from using Claude normally doesn’t need to be taught again.

In terms of security, Anthropic has also made significant enhancements, including:

1. Claude Duet handles login credentials using the same architecture as Claude Managed Agents and cannot itself see your password or login token.
2. Every action must be vetted by the auto mode classifier to prevent it from doing things you didn’t ask for.
3. Defenses against hidden instructions and prompt injection come from Anthropic’s past research on browser agents.
4. Before launch, it also underwent penetration testing by an internal red team and an external cybersecurity company, and was included in a public bug bounty program.

More notably, Anthropic has also added a dedicated sensitive information classifier and safety system prompt to Duet, checking every message. The highest-tier information, such as credit card numbers and ID document numbers, will be directly refused from being sent out. Sensitive data such as health and personal finance will only be sent when you explicitly instruct it. Anthropic says it will publish more safety design details in the coming days.

Claude Duet is currently still in research preview and is only available to Claude Pro and Max subscribers in the United States; those interested need to first go toOfficial product pageJoin the waitlist, and you’ll receive an Email when it’s your turn. The Team and Enterprise plans are not available yet.

It’s included in the subscription at no extra charge, but it shares the quota with other Claude usage, and it consumes quota differently from regular conversations; background checks also count. For large tasks such as financial models or presentations that require more model turns and background computation, usage will be higher.

Below are the key positioning differences among the various AI agents currently available:

Source: KOCPC Chinese