Claude Duet Preview: Anthropic's Proactive Agent App

Introduction

On October 21, 2026, Anthropic opened a research preview of Claude Duet, a new app in which Claude acts as a proactive agent instead of waiting for a prompt. In the company's words, it "comes to you with what needs your attention, completes tasks you ask of it, keeps working even when you step away, and checks with you before it acts on your behalf." Anthropic aims it at professionals and small business owners, and says it designed Duet around Claude Opus 5.5.

Two caveats frame everything below. Access runs through a waitlist, so inclusion in a plan does not mean every subscriber can open the app today. And Anthropic calls this an early version that "still might misread a thread or miscalibrate what's urgent." The Verge, which covered the launch on October 21, described Duet as a competitor to Meta's Muse and OpenAI's Dots. That comparison is The Verge's framing; Anthropic's post does not make it.

Feature Overview

Morning brief, then selective alerts. Each morning Duet sends a short brief on the most important items across the user's connected apps. During the day it "keeps tabs on those apps and messages you only when something needs your attention." Anthropic says it checks connected accounts every few hours, because checking nonstop would use up plan limits much faster. Users can tell it what they want to hear about and how often.

Work that continues off-device. Duet has a virtual machine with its own browser, so it does not need the user's laptop or phone to be on. The user describes a task in a short message, Duet works out which apps and steps to use, and a notification arrives when the task is done or needs review. Anthropic's examples range from scheduling meetings to building dashboards, financial or sales forecasts, prototypes and web apps.

Carried-over context in one thread. Duet inherits memory, skills and connected apps from the user's Claude account, so setup is light. The user talks to it by typing or voice in a single ongoing thread, so a quick reply and a week-long project start in the same place.

Approval gates. Duet reads, drafts and prepares work on its own. Before it sends a message, makes a payment, signs a document or deletes anything, it shows exactly what it is about to do and waits for approval. A few actions are off-limits even with approval, such as trading securities or moving money between accounts. Users can disconnect any app Duet has access to.

Connectors. Duet can use the "over 6,000 connectors and plugins" in the Claude Marketplace. Anthropic's examples include finding missed Gmail and Slack messages, checking QuickBooks or Stripe for unpaid invoices, preparing payroll in Gusto, drafting a LinkedIn job post that is published only after the user edits and approves it, and sending agreements through Docusign. Purchases go through Link, Stripe's digital wallet: Duet confirms the item, Link asks the user to approve the spend, and payment uses a virtual card so the real card number is not shared. Anthropic lists two items as "coming soon": a richer visual editor for Shopify stores and sign-in with logins saved in 1Password.

Safety Design

Anthropic says Duet's credential handling follows the design of Claude Managed Agents, so it never directly sees passwords or sign-in tokens. Actions are approved by the auto mode classifier already used for Claude Code, and its defenses against hidden instructions and prompt injection come from Anthropic's browser-use research. Red teams and an external firm penetration tested Duet, and it is covered by Anthropic's public bug bounty program on HackerOne.

The notable addition is a new dedicated classifier and safety system prompt that checks each message. It is designed to reject requests to share the strictest tier of information, including credit card numbers and government identification. Other sensitive information, such as health or personal finances, is let through only when the user has explicitly instructed Duet to send it. Anthropic says it will share more detail about Duet's safety design "in the coming days." Until then these are vendor statements: the announcement does not include penetration-test results or measured classifier error rates.

Usability Analysis

Availability is the first practical question. Anthropic says Duet is included in Pro and Max plans and shares the same usage limits as the rest of Claude, and that it is starting with Pro and Max users in the US. People join a waitlist and Anthropic emails them when Duet is ready. The post gives no separate price, and we found no statement on other regions or on a general-availability date.

Usage limits deserve attention. Anthropic says Duet's background checks count toward plan limits and that larger tasks, such as building a financial model or a deck, need more model turns and background work. The every-few-hours polling cadence is a deliberate trade of immediacy for cost, which means Duet is not a real-time monitor.

Anthropic also lists current gaps. Core functionality was built around work, so personal errands such as booking tickets are not as strong. Duet cannot yet show the user its browser. For now it lives in its own app, separate from the main Claude app. Testimonials in the announcement come from Anthropic staff and early-access business owners, so they are vendor-supplied rather than independent evidence. We have not used Duet; this assessment rests on Anthropic's announcement and The Verge's report.

Pros and Cons

Strengths: the approval gate keeps irreversible steps under user control; the virtual machine lets work finish while devices are off; existing memory, skills and connectors carry over; and the announcement is unusually specific about sensitive-data handling.

Limitations: access is waitlisted and starts in the US; every-few-hours checking limits timeliness; background work draws on the same plan limits as ordinary chat; and the safety claims are not yet independently documented.

Outlook

Anthropic says what it learns from early users will shape how Duet and the main Claude app work together, and that it will keep tuning when Duet decides to message a user. Several questions remain open: how the promised safety write-up looks, whether the waitlist widens beyond the US, and how well the urgency judgment holds up on real inboxes. The "coming soon" items, Shopify design and 1Password sign-in, would widen what Duet can do without a user present, and they will also widen what the approval gates must cover.

Conclusion

Duet is a well-specified research preview that puts approval prompts and sensitive-data checks at the center of a proactive design. Its value will depend on how reliably it judges what is urgent, which Anthropic itself says is still being tuned. It suits US-based Pro and Max subscribers whose work lives in email, chat, invoicing and payroll tools, and who are willing to join a waitlist and spend plan usage on background work. We rate it 3.5 out of 5 on the information available at launch.

Editor's Verdict

Claude Duet Preview: Anthropic's Proactive Agent App brings real, demonstrable value, though with caveats that deserve weighing.

The strongest case for paying attention: the approval gate shows the exact action before any message, payment, signature or deletion goes out, and a few actions stay off-limits even with approval. That alone raises the bar for what readers should expect in this space. Reinforcing that, a virtual machine with its own browser lets tasks finish while the user's laptop and phone are off — practical value rather than just headline appeal. The broader signal worth registering is straightforward: a proactive design reverses the usual chat pattern, because the agent opens with a morning brief and speaks up only when something needs attention. On the other side of the ledger, one constraint is real rather than a marketing footnote: the research preview is waitlisted and starts with Pro and Max users in the US, so it is not open to every subscriber yet. It should factor into any serious decision. Layered on top of that, a check-in cadence of every few hours means Duet is not a real-time monitor, and its background checks count toward plan limits — which narrows the set of teams for whom this is an obvious yes.

For US-based Pro and Max subscribers whose work runs through email, chat, invoicing and payroll tools, joining the waitlist and running a measured trial makes sense, with clear criteria for when to expand or pull back. For everyone else, the safer posture is to monitor coverage and revisit once the use cases that matter to your team are demonstrated in the wild.