Anthropic rolls out Claude Duet AI agent with 6,000+ connectors, background tasks and daily briefs

Anthropic has introduced Claude Duet, an AI agent that can take action on its own to help professionals and small business owners. It helps them manage ongoing projects and everyday business tasks by doing assigned work and finding things that need their attention.

Claude Duet AI agent

Claude Duet is built around Claude Opus 5.5. It brings over the user's saved information, skills and connected apps from their Claude account. Users can talk to it through text or voice in one ongoing conversation. This lets it handle quick requests and longer projects without starting a new conversation for each task. The regular Claude app waits for users to start a chat, but Claude Duet can contact them with a morning report that highlights important information from their connected apps.

Claude Duet runs on a virtual computer with its own browser. This means it can keep working even when the user's phone or computer is turned off. It checks connected accounts every few hours and alerts users when something needs their attention. Users can choose what it checks and when it sends updates. They can also review what it has done and what it is doing now, then give it more instructions.

Tasks it can handle include:

• Scheduling meetings.
• Building dashboards with dozens of tools.
• Creating financial and sales forecasts.
• Designing early versions of web apps and building them.
• Preparing documents, spreadsheets, presentations, reports and dashboards.

Features: Claude Duet

Background tasks and daily briefs

Users can give Claude Duet a task with a short text or voice message. It works out which apps and steps it needs, then does the work in the background. Its regular checks help it find things that need attention, so users do not have to open every connected app themselves. However, checking more often uses up their usage limits faster.

Business app integrations

Claude Duet works with more than 6,000 connectors and plugins through the Claude Marketplace. These cover messaging, finance, hiring, document management, marketing and online shopping. Examples include:

• Gmail and Slack: Find messages that were missed and prepare replies.
• QuickBooks and Stripe: Find unpaid bills and draft payment reminders.
• Gusto: Prepare and run payroll, including work hours, bonuses and expense repayments.
• LinkedIn: Write job ads and publish them after users edit and approve them.
• DocuSign: Prepare agreements and send them for review and signing.
• HubSpot: Draft follow-up emails using information about contacts and deals.
• Canva: Create a social media campaign for a product launch.
• Shopify: Set up, design and manage an online store.

Anthropic plans to add a more advanced visual editor for Shopify. It will let users design and preview their online stores before publishing them.

Business purchases and payments

Claude Duet supports online purchases through Link, Stripe's digital wallet, and Shop Pay. When using Link, the agent confirms the item with the user before Link asks them to approve the payment. The payment is then made with a virtual card, so the user's real card number is not shared. Anthropic also plans to let Claude Duet sign in to websites using login details saved in 1Password, but only with the user's permission.

User approvals and controls

Claude Duet must get approval before taking important actions. Users can review what it plans to do and disconnect any linked app whenever they want. Some financial actions are not allowed, even if a user approves them:

• Trading securities.
• Moving money between accounts.

Safety

Anthropic said Claude Duet uses its system for making AI agents trustworthy. The same system also guides Claude Code and Claude in Chrome. Its handling of login details follows the approach used for Claude Managed Agents. This means the agent does not directly access users' passwords or sign-in tokens.

Other security measures include:

• Action checks: An automatic system checks the agent's actions to help stop it from doing things users did not ask for.
• Prompt-injection defenses: The system protects against hidden instructions that could try to control the agent. These protections use research by Anthropic on browser use, along with other safety measures built into the AI model.
• Security testing: Anthropic's red teams and an outside company that tests security have checked the system's defenses. The service is also covered by Anthropic's public program that rewards people for reporting security bugs.
• Sensitive-information checks: A special detection system and safety instructions check messages for private information. The agent refuses requests to share information in its highest-protection category, including credit card numbers and government ID details. Other private information, such as health details and personal finances, can be shared only when users clearly tell the agent to send it.

Anthropic said it will keep improving these protections and share more information about how its safety systems work.

Anthropic employees are also using Claude Duet for their work. Boris Cherny, who leads the Claude Code team, talks through his mental to-do list with the agent while walking. Chelsea Larsson, who leads the team that designs model and product behavior, gives it a task list before spending time with her children and gets a report in the evening. Karan Ahuja, who leads the Claude Duet team, uses it to schedule interviews with users, develop prototypes and release code.

Limitations

Claude Duet is an early release for research and testing. It still has several limitations:

• Task accuracy: It may misunderstand conversations, get the level of urgency wrong or miss something that arrives between its regular checks.
• Personal tasks: It is less capable at personal errands, such as booking tickets, than at work-related tasks.
• Browser visibility: Users cannot yet see the agent's browser.
• Separate app: Claude Duet runs separately from the main Claude app.
• Usage limits: It shares usage limits with the rest of Claude. Background checks count toward these limits. Bigger tasks, such as creating a financial model or presentation, need more steps with the AI model and more background work.

Anthropic said feedback from early users will help it improve how Claude Duet and the main Claude app work together.

Availability

Claude Duet is being rolled out as a research preview for Claude Pro and Max subscribers in the US. Users can join the waiting list and will get an email when they can access it.

The research preview includes:

• Platforms: Separate apps for iOS and Android, plus access through claude.ai on a computer.
• Eligible plans: Claude Pro and Max subscribers in the US.
• Excluded plans: Team and Enterprise.