What the source published

Anthropic's product page introduces Claude Duet, a proactive agent meant to take on the work that does not need you personally. It is powered by Opus 5.5 and is currently a research preview on a waitlist for Claude Pro and Max subscribers in the US; Team and Enterprise plans are not included yet. It is a standalone app on iOS and Android, and on a computer you open it on the web at claude.ai.

What sets it apart is how it works: Duet figures out what needs your attention, keeps working while you are away, and only comes back when something genuinely needs your call. It uses your Claude account, including your memory, skills and connected apps, and everything lives in one thread. You hand it work by message, in text or voice, with no prompts or setup to write, and it returns finished work such as a doc, sheet or deck for your review.

The examples Anthropic gives are concrete: a morning brief that flags what needs you across your inbox and calendar and drafts replies for your approval; small-business legwork such as finding unpaid invoices and drafting reminders, preparing payroll for you to review, and getting contracts ready to send; and assembling a brief, deck or dashboard on request.

Sources: [1] [2]

Try it: hand work over in this order

  1. Check eligibility first: you need a Claude Pro or Max plan in the US, and during the research preview you join the waitlist and wait for an invite.
  2. Connect your inbox, calendar and other apps, then review the memory and skills it inherits from the Claude app — that is the basis for its judgment about what matters.
  3. Spend the first week on the morning brief only: read what it flags and the replies it drafts, approve nothing automatically, and calibrate how well it judges urgency.
  4. Then hand it one bounded task by message, such as finding overdue invoices and drafting the reminders for you to read, and check the returned doc or sheet line by line.
  5. It asks before sensitive operations like sending messages or paying; keep the habit of reading every draft before you approve, instead of batch-approving to save time.
  6. Watch your limits: during the preview Duet shares your plan's usage limits with the rest of Claude and its background checks count toward them, while larger tasks take more model turns.

Sources: [1]

The full picture: how it divides work with regular Claude

Anthropic's own framing is that the Claude app waits for you to open a chat and work alongside it, while Duet comes to you: it tells you what needs attention each morning, takes tasks you hand it by message, and reaches out when one is done or needs your approval. It works on its own computer in the cloud, and approvals, logins and safety checks run in that cloud computer so sensitive details stay private. You can see what it has done and is doing, and steer it.

The preview caveats are stated plainly too: it may misread a thread or miscalibrate what is urgent, and it can miss something that arrives between check-ins. Treat it as a tireless first pass — screening and drafting are its job; judgment stays with you, especially for anything with money or other people on the other end.

Sources: [1]

How to read this

Duet matters because the contest has moved: the last generation competed on who answers best, this one competes on who moves your work forward while you are away. With OpenAI pushing always-on agents as well, subscribers should judge these products less on raw answers and more on whether the flags are accurate, the drafts usable, and the background activity respectful of plan limits.

The sensible adoption path is narrow first: let it watch just your inbox and calendar, accept only briefs and drafts as outputs, and widen its remit only after a few weeks of calibration. During a research preview, both the feature set and the usage rules can change, so keep anything mission-critical on your own checklist as well.

Sources: [1] [2]

Key takeaways

  • Claude Duet is a proactive agent: a morning brief first, then tasks you hand over by message, reporting back when done or when it needs approval.
  • It runs on Opus 5.5 and is a waitlisted research preview for US Pro and Max subscribers only; Team and Enterprise are not included yet.
  • It reuses your Claude memory, skills and connected apps, works in its own cloud computer, and must ask before sensitive operations.
  • Background checks share and consume your plan limits during the preview, so watch usage for a week before giving it bigger jobs.
  • Anthropic admits it can misjudge urgency or miss new arrivals: start small, accept briefs and drafts, keep the decisions yours.

Sources: [1]