Claude Duet: The Agent That Comes to You (Not the Other Way Around)
Direct answer
Anthropic opened on October 21, 2026 the research preview of Claude Duet: a standalone app where Claude works as a proactive agent. Instead of waiting for your question, Duet sends a morning brief with what most needs your attention across your connected apps, checks those apps every few hours, messages you only when something needs a decision, and keeps working in the background. It runs on Claude Opus 5.5, on its own cloud virtual machine with its own browser, so a task continues even after you close your laptop or phone. Access is by waitlist, only for United States subscribers on Pro and Max plans; Team and Enterprise plans are not included yet.
What Duet does in a workday
- Morning brief: it reviews email, calendar and connected apps and presents the most important items first, with drafts ready for your review.
- Background work: scheduling meetings, building dashboards with dozens of tools, creating financial or sales forecasts, prototypes and web apps, according to Anthropic's list.
- Inherited memory and connectors: it uses the memory, skills and apps already connected to your Claude account. Anthropic cites more than 6,000 connectors and plugins in the Claude Marketplace, including Gmail, Slack, Google Calendar, QuickBooks, Stripe, Shopify, HubSpot and Notion, according to coverage of the announcement.
- Money with approval: it can find unpaid invoices in Stripe or QuickBooks and draft reminders, and it can buy with your approval through Stripe's Link wallet or Shopify's Shop Pay, using a virtual card instead of exposing your real card number.
The guardrails (and why they matter more than the features)
- It asks approval before acting: Duet shows the exact action and waits for your approval before sending a message, making a payment, signing a document or deleting anything.
- Some actions are off-limits even with approval: trading securities or moving money between accounts is out of bounds, according to coverage of the announcement.
- Shared usage limits: Duet is included in Pro and Max plans during the preview, but it shares those plans' usage limits. Background checks count, so an agent that checks constantly can burn through your allowance faster than chat alone.
- Anthropic's own warning: Duet might misread a thread or miscalibrate what is urgent, and it is weaker on personal errands like booking tickets. This is a research preview, not a finished product.
What it means for your business
- The agent moves from answering to watching: the real change is not a smarter chatbot but a system that works between conversations. That helps with follow-up (the money lost by not responding in time), but it also multiplies the points where it can misunderstand something.
- Human approval is the design, not a patch: if you delegate work to a proactive agent, demand this pattern: the agent proposes, you approve the sensitive stuff, and the forbidden stuff stays forbidden by design.
- Do not buy the promise; test the flow: the preview is limited and waitlisted. Meanwhile, the useful exercise is to list which follow-up tasks in your business you would delegate with prior approval, and which ones never.
Related links
- What to do when your AI agent makes a mistake
- How to tell if your AI is working: the 5 KPIs that matter
- The AI agent that qualifies prospects: how it separates the curious from buyers
Frequently asked questions
What is Claude Duet?
A standalone Anthropic app, in research preview since October 21, 2026, where Claude works as a proactive agent: it sends a morning brief, checks your connected apps every few hours and continues tasks in the background, asking approval before acting on your behalf.
Who can use it today?
Only United States subscribers on Pro and Max plans, via a waitlist. Team and Enterprise plans are not included in this preview. It is included in those plans and shares their usage limits.
What model does it run on, and where does it work?
On Claude Opus 5.5, on its own cloud virtual machine with its own browser, so tasks continue even after you close your laptop or phone.
Can it send money or sign without my approval?
No: it shows the exact action and waits for your approval before sending a message, making a payment, signing a document or deleting anything. Some actions, such as trading securities or moving money between accounts, are forbidden even with approval.
