Claude Duet Explained: Features, Access, and Usage Limits
What is Claude Duet? Explore Anthropic's proactive AI assistant, its features, US waitlist access, shared usage limits, and important preview restrictions.
Claude Duet is Anthropic's new proactive AI assistant: it can accept tasks through messages, keep working on a cloud computer, and contact you when work needs attention. Anthropic announced the research preview on October 21, 2026. It follows a rapid series of personal-agent launches from other major AI companies.
The useful question for professionals is how much coordination an assistant can remove from the working day. A service that prepares a useful brief before a meeting could save several manual steps. One that repeatedly interrupts with low-priority suggestions could create another inbox to manage.
Checked October 22, 2026. This explainer is based on official announcements and documentation, not a hands-on review. Preview features and access conditions can change.
What is Claude Duet?
Anthropic describes Duet as an agent that can initiate contact and take on work through typed or spoken messages. It brings over memory, skills and connected apps from your Claude account. Its separate cloud computer and browser let work continue after you close your own device.
For the reader, the distinction is about workflow: you can delegate an outcome, return later and review the result. That makes the quality of the handoff important. An instruction should explain the desired result, what information to use and when the assistant should stop for a decision.
See Anthropic's Claude Duet announcement for the launch details.
Four personal-agent launches in less than three months
| Product | Announcement date | Official source |
|---|---|---|
| Grok Bot | August 11, 2026 | Grok Bot launch reference |
| Meta Muse | September 8, 2026 | Meta announcement |
| OpenAI dots | September 29, 2026 | OpenAI announcement |
| Claude Duet | October 21, 2026 | Anthropic announcement |
The sequence suggests growing competition around delegated work. It does not establish that these products have identical permissions, coverage or reliability. A useful comparison needs the same real task, connected services and review criteria across each assistant.
What can Claude Duet do?
According to the official product page, Duet can send morning briefings, surface inbox and calendar priorities, draft replies, and prepare documents, decks or dashboards. Business examples include invoice reminders, payroll preparation and contracts awaiting review. It is available through a standalone iOS or Android app and on the web.
Anthropic also describes larger assignments such as forecasts, prototypes and web apps. Its announcement gives an early-tester example in which Duet prepared a client-call brief and spotted that the call fell just before a flight. This is a vendor-reported example, rather than independent evidence of everyday reliability.
A practical first task could be preparing a one-page meeting brief with links to the relevant messages, open questions and decisions needed. The brief has an obvious audience and a result you can inspect quickly. By comparison, an open instruction to manage an entire business leaves much more room for assumptions.
Who can access the research preview?
Anthropic's getting-started guide says access requires a US Claude account, a Pro or Max subscription and a waitlist invitation. Invitations arrive in waves. Team and Enterprise plans are not included yet.
During the preview, Duet is included without an additional subscription charge. Paying for Pro or Max does not by itself guarantee immediate access. Users in Vietnam are outside the announced US-only preview.
Background work uses your Claude allowance
Duet shares your plan's usage limits with the rest of Claude. Morning briefings and background checks count toward that allowance, and larger jobs can require more work than a short chat. Anthropic says Duet checks connected accounts every few hours rather than continuously.
For a frequent Claude user, the useful measure is work saved per unit of allowance. A polished briefing may be worthwhile if it replaces repeated searching and sorting. An update that restates information you already know has less value. Start with a narrow brief and assess whether its contents change what you do next.
Approvals, restrictions and privacy details
Anthropic says Duet is designed to show the proposed action and request approval before sending messages, paying, signing documents or deleting content. Trading securities and moving money between accounts remain prohibited even with permission.
The help guide lists further preview limits: users cannot watch or take over its browser; it cannot make phone calls or send SMS to others; Microsoft 365 is not available yet. Disconnecting an app stops future access but retains memory of earlier information. Duet can also misread a thread or misjudge urgency.
This makes access, memory and action approval three separate things to consider. An assistant may have permission to read a source without having permission to send its contents elsewhere. For related questions about stored workplace information, read our guide to AI meeting assistant privacy and meeting data.
Netzender's take: test the handoff before expanding the job
Our assessment is that Duet's most useful early test is a recurring task with a clear finish line. Choose one workflow, such as preparing tomorrow's meeting notes, and evaluate it over several working days.
- Check accuracy: can you trace names, deadlines and claims to the original material?
- Check usefulness: does the result reduce preparation or merely reformat information?
- Check interruptions: are the notifications relevant enough to deserve attention?
- Check total effort: include time spent correcting outputs and reviewing approvals.
These are suggested evaluation criteria, not results from a Duet test. For broader platform selection, our ChatGPT vs Claude vs Gemini for work guide provides additional context.
Claude Duet gives Anthropic a new way to put AI into ongoing work. Its value for an individual user will depend on whether it produces useful, verifiable results with a manageable review burden.
Netzender Editorial Team