Anthropic’s new Claude Duet agent can ask you first when something needs your attention.
Anthropic announced Claude Duet on October 21. Duet is a proactive AI agent that checks a user’s connected apps, flags what needs attention, and completes some tasks in the background. It is a research preview with a waitlist, available first to Claude Pro and Max subscribers in the United States, and built on Claude Opus 5.5.
Anthropic is pitching Duet to professionals and small business owners. Each day it prepares a short briefing of action items from the user’s connected apps. During the day, it checks those apps and messages the user only when something needs attention.
Users can also give Duet tasks such as:
- It can handle simple administrative work, like scheduling meetings.
- It can take on big jobs, such as building dashboards, creating financial or sales forecasts, prototyping, and developing web apps. It returns extensive tasks as files the user can review and edit.
- It connects to Stripe to check for unpaid invoices and draft payment reminders.
- It can make purchases through Link and Shop Pay with the user’s approval.
- It uses its own cloud browser for websites without a connector.
Duet runs on its own cloud-based computer, so it keeps working even after the user closes the app. It carries over memory, skills, and connected apps from the user’s Claude account, and Anthropic says it requires little setup. It runs on iOS, Android, and claude.ai.
Anthropic also says Duet reads, drafts, and prepares work on its own, but it’s designed to ask before sending a message, making a payment, signing a document, or deleting anything.
Anthropic also describes three protections for sensitive data:
- Duet never directly sees a user’s passwords or sign-in tokens.
- A new classifier and system prompt are designed to stop Duet from leaking sensitive information.
- Information that may relate to a user’s health or personal finances passes through only if the user explicitly tells Duet to share it.
Anthropic said it would share more about Duet’s safety protections in the coming days.
Duet costs nothing extra, but it uses the same usage limits as the rest of Claude. Background checks count toward those limits, and Anthropic says users could reach them “sooner than with chat alone.” Time-consuming tasks, such as building a financial model or a deck, take more model turns.
Usage limits are also the reason Duet does not check constantly. Anthropic says it checks connected accounts every few hours because nonstop checking would use up plan limits much faster. A new urgent email may therefore wait for the next check before Duet notices it.
Anthropic also points to these limits of the early preview:
- Team and Enterprise plans are not supported yet, and invitations will go out in waves.
- Anthropic says this early version “still might misread a thread or miscalibrate what’s urgent.”
- Anthropic says it will keep tuning when Duet messages users, based on tester feedback. Users will be able to set how often Duet messages them and which topics it covers.
A review by developer Flavio Copes, based on Anthropic’s documentation rather than hands-on testing, lists three more gaps. Microsoft 365 is not available yet. Duet cannot access local files, make phone calls, or send texts. Users also cannot inspect or take over Duet’s browser in the preview.
Duet arrived about six weeks after Meta’s Muse and about three weeks after OpenAI’s Dots. Like the other two, Duet is a proactive agent with a virtual machine and its own browser.
| Agent | Announced | Availability | Safety approach |
| Meta Muse | September 8 | US rollout on iOS, Android, and the web | A separate Sentinel agent reviews actions before they reach the internet |
| OpenAI Dots | September 29 | Listed under ChatGPT Pro, not included in Free, Go, or Plus | Idle work is limited to read-only research, with rules on when to ask for approval |
| Claude Duet | October 21 | Waitlist for US Pro and Max subscribers | Asks before sending, paying, signing, or deleting, plus a classifier for sensitive data |
All three are designed to keep working in the background. Meta and OpenAI describe their agents differently:
- Meta says Muse can send emails, book travel, fill out forms, and complete a checkout.
- OpenAI says Dots can conduct research, draft documents, and develop software. Users can message it in ChatGPT, Slack, or Microsoft Teams.
Duet puts Anthropic in the same proactive agent category that Meta and OpenAI entered first. It is still a research preview, so we can only judge how well it decides when to message users and when to ask for approval once waitlisted Pro and Max subscribers start using it. For now, you can join the waitlist on the Claude Duet product page.































