Claude Duet Is Here: Anthropic's Proactive AI Agent That Works While You Sleep

Most AI assistants wait for you to ask. Anthropic's newest product does the opposite: it comes to you first. On 21 October 2026, Anthropic opened a research preview of Claude Duet, a standalone app in which Claude works as a proactive agent, checking connected work apps, preparing a morning brief, and continuing delegated tasks in the background, even when your phone or laptop is off.

Duet is not a new model. It is built around Claude Opus 5.5, Anthropic's existing top model, and it is included in Claude Pro and Max plans during the preview, sharing those plans' usage limits. What is new is the direction of the conversation. Instead of you opening Claude and typing a question, Duet starts the conversation when it finds something that needs your attention.

That shift sounds small, but it changes what an assistant is for. A chatbot helps you do a task you already know about. A proactive agent is supposed to notice the task you missed, the invoice that went unpaid, the meeting that needs scheduling, or the message that needs a reply, and bring it to you before it becomes urgent.

What Duet actually does each day

Each morning, Duet sends a short brief on the most important items across the user's connected apps. During the day, it checks those accounts every few hours, not continuously, and messages only when something needs a decision. Anthropic says checking constantly would use up a plan's limits much faster, so the periodic rhythm is a deliberate design choice, not a limitation it forgot to mention.

Duet runs on its own virtual machine with its own browser, so a task keeps progressing after someone closes their laptop. When the work is done or needs review, a notification arrives. That matters for longer jobs, such as building a financial model, preparing a slide deck, or drafting a dashboard, which do not fit neatly into a single chat session.

The connection layer is the Claude Marketplace, which Anthropic says offers more than 6,000 connectors and plugins, including Gmail, Slack, Google Calendar, Google Drive, Intuit QuickBooks, Stripe, Shopify, HubSpot, Gusto, Docusign, Notion, Canva, Microsoft 365 and Ramp. Those connections are what let Duet move from chatting to doing: finding missed messages, checking for unpaid invoices, running payroll for review, or preparing documents.

Credit: Claude logo, via Logodix


What it will and will not do on your behalf

Anthropic lists concrete jobs Duet can handle. It can find missed Slack and Gmail messages and draft replies for review. It can check QuickBooks or Stripe for unpaid invoices and draft reminders. It can run payroll in Gusto with hours and bonuses filled in, for human review. It can draft a job post for LinkedIn after a human edits and approves it, and prepare Docusign agreements for signature.

Duet can also spend money on a user's behalf through Stripe's Link digital wallet or Shopify's Shop Pay, paying with a virtual card rather than exposing a real card number. That is a powerful capability, and Anthropic has put guardrails around it: Duet pauses for approval before sending messages, making payments, signing documents or deleting items. Trading securities and moving money between accounts are prohibited, even if a user approves.

Those guardrails are important because proactive agents have more opportunities to misunderstand than chatbots do. A chatbot misunderstands a question you asked. A proactive agent can misunderstand a message you never saw, a deadline you did not set, or a payment you did not intend. Anthropic itself cautions that the preview may misread message context or judge urgency incorrectly, and that it is less effective at personal errands such as booking tickets.

Who can use it right now

Access is narrow, and that is deliberate. The research preview is available through a waitlist, starting with Pro and Max subscribers in the United States, on iOS, Android and claude.ai. Team and Enterprise plans are not included in the preview, though Anthropic says it plans to bring Duet to them later. During the preview, Duet is included in Pro and Max at no separate fee, but background checks and larger tasks draw on the subscriber's existing usage limits.

That means Duet is currently an early test, not a tool every Claude customer can switch on today. Anthropic says it will email waitlisted users as soon as their access is ready, and that feedback from early users will help shape how the agent works with the main Claude app. Access is tied to the email address given on the waitlist, so users are advised to use the same address as their Claude account.

For users outside the United States, or on Team and Enterprise plans, the honest answer is that Duet is not available yet, and Anthropic has not said when that changes. That geography limit is the first thing to check before getting excited about the product.

Credit: Anthropic office illustration, via Quickwork


Why proactive agents are hard

The hardest part of a proactive agent is not capability. It is judgement: deciding what needs you, what can wait, and when to check first. Anthropic says Duet is designed around exactly that question. But judgement is also where errors are most costly. A missed message is annoying. A wrongly sent payment, a wrongly signed document or a wrongly deleted file is serious, which is why Duet's approval pauses exist.

There is also a cost question that users should understand. Background checks count toward shared usage limits, and larger tasks such as building a financial model or a deck require more model turns. A proactive agent that checks every few hours is therefore not free, even if there is no separate fee. Frequent checks and big jobs will use plan capacity faster, and users who run Duet heavily may hit limits sooner than they expect.

Privacy is the other hard question. Duet works because it can see across email, calendar, documents, finance and messaging. That breadth is its value, and also its risk. Users should check which apps they connect, what Duet can read, and what it stores, before handing it the keys to their work life.

What Duet means for the assistant race

Duet arrives as every major lab moves from chatbots to agents that can act. OpenAI, Google and Meta are all building agents that can use tools, browse and complete tasks. Anthropic's bet is different in emphasis: not the most tools, but the most careful judgement about when to act and when to ask.

Whether that bet works will depend on the preview. If Duet's morning briefs are genuinely useful, and its approval pauses feel protective rather than annoying, it could become the way many professionals start their day. If it misreads urgency, burns through usage limits, or sends the wrong draft, users will turn it off quickly, because a proactive agent that creates work is worse than no agent at all.

Credit: AI assistant workspace photo, via CodingFlow


Key Takeaways

  • Anthropic opened a research preview of Claude Duet on 21 October 2026, a standalone app in which Claude works as a proactive agent, built around Claude Opus 5.5.
  • Duet sends a morning brief, checks connected apps every few hours, and continues tasks on its own virtual machine with its own browser, even when the user's device is off.
  • It connects via the Claude Marketplace's more than 6,000 connectors and plugins, including Gmail, Slack, Google Calendar, QuickBooks, Stripe, Shopify and Notion.
  • Access is waitlisted, for Pro and Max subscribers in the United States only, on iOS, Android and claude.ai, with no separate fee during the preview but shared plan usage limits.
  • Duet pauses for approval before sending messages, making payments, signing documents or deleting items, and prohibits trading securities or moving money between accounts.
  • Anthropic cautions the preview may misread context or urgency, is less effective at personal errands like ticket booking, and background checks consume plan limits.

Claude Duet is Anthropic's answer to a simple question: what if your assistant noticed the work before you did? The preview is narrow, the guardrails are real, and the judgement is unproven. But if it works, the morning brief may become the most useful thing Claude does all day.

Sources

Comments

Post a Comment