Claude Duet: Anthropic’s Proactive AI Agent That Works While You’re Away
Your inbox fills up while you’re in meetings. Your Slack pings while you’re on a call. Three QuickBooks invoices have been sitting unpaid for two weeks because nobody found the right moment to chase them. By the end of a full working day, you’ve spent most of your energy managing the flow of information rather than doing the actual work that grows the business.
Claude Duet, launched this week by Anthropic as a research preview, is built for exactly that situation. It’s a proactive AI agent running on Opus 5.5, Anthropic’s most capable model, that monitors your connected apps throughout the day, figures out what actually needs your attention, and handles or queues everything else.
The key word is proactive. This is not a chatbot you go to with questions. It comes to you.
Every morning, Claude Duet sends a short brief covering the highest-priority items across your Gmail, Slack, HubSpot, QuickBooks, and hundreds of other connected tools. Throughout the day, it checks those accounts every few hours and reaches out only when something genuinely requires a decision from you.
Tasks run in the background without needing you at your desk. Claude Duet operates on a virtual machine with its own browser, so it keeps working after you’ve put your phone down and walked away. Describe what you need — draft follow-up emails from HubSpot deal history, prepare payroll in Gusto, build a financial forecast, prototype a web app — and it works through the steps independently. When it finishes, or when it reaches something that needs your input, you get a notification.
The approval model is worth understanding before you form an opinion about this. Before Claude Duet sends any message, makes a payment, signs a document, or deletes anything, it shows you exactly what it plans to do and waits for your confirmation. Some actions are off-limits regardless — trading securities, moving money between accounts. The reading, drafting, and preparation all happen automatically. Every consequential click stays with you.
Anthropic built this on the same trustworthy agents framework that governs Claude Code. Credentials never pass directly through the model. Defenses against prompt injection — attempts by malicious content in an email or document to trick the AI into doing something unintended — are built in. A dedicated classifier actively guards against leaking sensitive information like credit card numbers or government IDs.
The connector library covers over 6,000 tools in the Claude Marketplace. Gmail, Slack, Stripe, QuickBooks, HubSpot, Gusto, DocuSign, LinkedIn, Canva, and Shopify are all there. If your business runs on SaaS tools, Claude Duet can coordinate across them without you switching tabs.
The research preview is live now on iOS, Android, and claude.ai for Pro and Max plan subscribers, available through a waitlist. It’s included within existing plan pricing.
AI assistants have promised to save time for years. Most required you to initiate every single interaction. Claude Duet inverts that model entirely, and for small business owners juggling more than they can track, that difference is significant.
Want to explore how proactive AI agents could benefit your business? Let’s talk.