Anthropic has opened a research preview of Claude Duet, an AI agent that does not wait to be asked. It sends a morning brief, keeps an eye on your connected apps and can carry on working on its own cloud computer after you close the app.
A disclosure first: Ctrl + Alt News uses Claude, made by Anthropic, as an editorial assistant, and Claude Duet is an Anthropic product. We have kept this article to what the reports say and what we could not confirm.
What it is
Duet is described as a “proactive agent”. It draws on your history with Claude, plus the memories, skills and connected apps on your Claude account, and you talk to it in one continuing conversation, typing or speaking. It is aimed at professionals and small-business owners who have more tasks than they can keep track of.
It is built on Claude Opus 5.5, according to 9to5Mac and MacRumors. Anthropic positions it against Meta’s Muse and OpenAI’s Dots, the latter launched a few weeks earlier, according to 9to5Google.
What it does
According to the reports, Duet:
- sends a short morning brief that reviews email and calendar and picks out priorities;
- checks connected apps during the day, MacRumors says every few hours, and alerts you only when something needs attention or a decision;
- runs on its own cloud computer with its own browser, so it can work when your device is off;
- takes on jobs such as scheduling meetings, building dashboards, financial or sales forecasts, prototypes and web apps.
Bigger pieces of work, such as research briefs, financial models, slide decks and dashboards, arrive as editable files in a tab called Make, while a Tasks tab tracks what is in progress. 9to5Mac also lists Stripe, where it can check unpaid invoices and draft payment reminders, and Link and Shop Pay, where it can make purchases with your approval.
The safeguards described
Duet is designed to ask before it sends a message, makes a payment, signs a document or deletes anything, according to 9to5Mac and MacRumors, and only you can mark a task as finished. MacRumors adds that it does not see passwords or logins when it is given account access, and that a dedicated classifier and a safety prompt check each message to guard against leaks of sensitive information.
Those are design claims. As we write, we have found no independent testing of how well they hold up in real use. For a sense of why that matters, see our report on AI agents that reached live systems without authorisation in India.
Who can try it, and what it costs
This is a research preview, with access by waitlist. It is rolling out to Pro and Max subscribers in the United States, with invitations sent in waves. Team and Enterprise plans are not included yet, and Anthropic plans to extend it later.
Duet is included at no extra charge, but its use counts toward your plan’s limits, background work included. So you may reach those limits sooner than you would by chatting alone.
The reports mention only the United States. We found nothing about availability in the UK.
What we could not confirm
We read three reports, and all of them rest on Anthropic’s announcement. We could not open the announcement itself. The reports also disagree on platforms: 9to5Mac lists iOS and Android apps plus the web, while MacRumors mentions the web and an iOS app.
Nobody outside the company has yet reported how reliably Duet works. Treat the safeguards and the claimed tasks as Anthropic’s description until that changes. For how a rival is pitching its own models, see our piece on Meta’s Muse Wonder. We will update this article when the announcement or independent tests are available.
Leave a Reply