Anthropic opens preview of Claude Duet, a proactive agent that monitors business tools
Claude Duet runs as a standalone app that sends a morning brief, checks connected apps like Slack, QuickBooks and Gusto in the background, and asks before taking action.
On October 21, 2026, Anthropic opened a research preview of Claude Duet, a separate application where Claude operates as a proactive agent rather than a passive chat assistant. The preview is available by waitlist for Pro and Max subscribers in the United States on iOS, Android and the web.
The announcement comes as major AI providers compete to move beyond passive chat boxes into assistants that monitor external systems on their own. Claude Duet arrives as a direct counterpart to recent proactive tools such as xAI’s Grok Bot, Muse and OpenAI Dot.
Unlike standard chat interfaces that wait for a question, Claude Duet connects to workplace software to monitor activity and propose actions. Anthropic demonstrated connections to tools including Gmail, Slack, HubSpot, Intuit QuickBooks, Stripe, Gusto, DocuSign, Canva and Shopify. In practice, Duet sends a summary each morning of items across connected tools, tracks threads during the day and messages the user when something needs attention. It can draft email and Slack replies, draft reminders for overdue invoices, prepare payroll runs in Gusto, draft job postings for LinkedIn, and prepare agreements in DocuSign. With Stripe Link or Shopify Shop Pay, it can make purchases using virtual cards after user confirmation.
Anthropic emphasizes that Claude Duet remains an early preview. The company notes that the agent can misread conversation threads or misjudge what is urgent, cannot yet show its browser, and lives in its own app separate from the main Claude workspace. Background monitoring and multi-step tasks also draw from the same usage limits as standard Claude conversations on Pro and Max plans.
What it means for your work
The shift from a reactive chat box to background monitoring changes how a team works with AI. Instead of remembering to open an app and write a detailed request, the assistant keeps watch over routine administrative handoffs: uncollected invoices, missed messages or scheduling conflicts. With Anthropic, xAI and OpenAI now offering proactive agents, workplace software is increasingly shifting from tools that wait for input to systems that present drafts for review.
At the same time, giving an agent visibility into financial records, customer histories and communication channels introduces practical risks. Anthropic designed Duet to request confirmation before completing irreversible actions like sending a message, running payroll or approving a payment. For business owners and practice managers, the primary discipline is resisting the temptation to approve suggestions blindly. When an assistant prepares twelve invoices or drafts five client replies, each entry still requires verification against primary records.
What to check
If your team joins the preview or plans to evaluate proactive agents:
- Verify tool permissions: Start by connecting low-risk services like calendars or drafting tools before giving an agent access to payroll, invoicing or payment accounts.
- Review before approving: Keep approval authority with the person accountable for the result. Check numbers, dates and recipient addresses before confirming an action.
- Track plan usage: Background checks and multi-turn workflows consume plan limits faster than single chat queries. Keep tabs on usage if multiple team members run intensive background tasks.
Next step: to decide which routine responsibilities in your firm are ready to delegate and which should stay entirely in human hands, Map your own job, task by task, and see where AI helps and where you stay in charge provides a structured one-page framework.