
Key points
- Claude Duet operates as a proactive agent built on Opus 5.5, running in a cloud-based virtual machine with its own browser.
- The tool integrates with over 6,000 applications, including Gmail and Slack, to send morning briefs and alerts on pending items.
- Users must approve sensitive actions like payments or deletions, though the agent cannot trade securities or transfer funds between accounts.
Anthropic introduced Claude Duet on October 21 as a proactive artificial intelligence agent that monitors connected applications and completes tasks without requiring constant user input.
How it unfolded
- Anthropic announced the research preview of Claude Duet on October 21, positioning it as a competitor to Meta’s Muse and OpenAI’s Dots agents.
- The company opened a waitlist for the tool, limiting initial access to Pro and Max subscribers located in the United States.
- Early reviewers, including Dan Shipper of Every, tested the agent before the public launch, noting its professional focus and current limitations in voice features.
Who is affected
The agent targets professionals and small business owners who manage numerous digital tasks and need assistance tracking important items across multiple platforms. According to MacRumors, Anthropic designed Duet for users who find it difficult to monitor all necessary details in their daily workflows. The tool aims to reduce the cognitive load on these users by automating routine administrative checks and providing timely summaries of urgent matters.
Duet connects to more than 6,000 applications, including Gmail, Slack, QuickBooks, Stripe, HubSpot, and Shopify, according to The Next Web. This broad integration allows the agent to pull data from various sources and present a unified view of pending tasks. Users can ask the agent to schedule meetings, prepare sales forecasts, or build simple web applications, leveraging the memory and skills stored in their existing Claude accounts.
What happens next
The research preview is available through the web, a standalone iOS app, and Android devices, as reported by The Next Web. Users do not pay an extra fee for Duet, but the background checks and tasks performed by the agent count toward their existing Claude plan usage limits. This means heavy use of the proactive agent could deplete the monthly or daily quotas associated with Pro or Max subscriptions.
Anthropic has outlined specific limitations for the current preview. The agent may misjudge the urgency of certain messages, and personal errands like booking tickets remain incomplete features. Users cannot watch the agent’s browser activity in real time, and voice capabilities are limited to transcription rather than a full voice mode, according to early reviewer Dan Shipper. The company states that Duet will not see passwords or logins when accessing accounts, relying on a dedicated classifier to guard against sensitive information leaks.
The background in plain language
Proactive agents differ from traditional chatbots by initiating actions rather than waiting for prompts. They operate within virtual environments that mimic real browsers, allowing them to navigate websites and applications as a human would. This architecture requires robust approval workflows, as the agent can perform tasks like sending emails or making payments. Teams usually review these permissions carefully to prevent unauthorized actions, ensuring the agent only executes tasks that align with established security policies and user intent.
What to do and how to stay safe: Anthropic
- Review the list of connected applications and revoke access for any services the agent does not need to monitor or control.
- Verify that approval settings are enabled for all sensitive actions, including payments, document signing, and file deletions, before delegating tasks.
- Monitor usage limits closely, as background checks and automated tasks consume the same quota as manual interactions with the AI model.
- Check the agent’s morning briefs and alerts regularly to ensure urgent items are correctly identified and addressed in a timely manner.
Step-by-step guide: Injection attacks examples: how prompt leaks break AI systems
General security guidance from the Circuit Observer newsroom. It is not confirmed advice from the organisations named in this story.
Frequently asked questions
Can Claude Duet trade stocks or transfer money between bank accounts?
No, Anthropic states that the agent blocks trading securities and transferring money between accounts altogether, though it can make purchases via virtual cards.
Is Claude Duet free to use for existing subscribers?
The tool is included at no extra cost for Pro and Max subscribers, but its background checks and tasks count toward existing usage limits.
How does Claude Duet handle sensitive information like passwords?
The agent does not see passwords or logins when given access to accounts and uses a dedicated classifier to guard against sensitive information leaks.



