Claude Duet explained: what Anthropic’s proactive agent does and who can use it

Anthropic opened a research preview of Claude Duet on 21 October 2026, describing it as “a new app where Claude works as a proactive agent”. Unlike the main Claude app, which waits for the user to open a chat, Duet is built to come to the user first: Anthropic says it sends a morning brief, checks connected apps every few hours, comes to the user “with what needs your attention” and returns finished work for review. Access is by waitlist, and for now only Pro and Max subscribers in the United States can join it.

What Anthropic says it does

According to Anthropic’s announcement, Duet is aimed at “professionals and small business owners with more on their plates than they can track”. The product page lists its core jobs as handling the inbox, calendar and follow-ups: flagging what needs attention, drafting replies for approval and preparing the morning brief. Tasks can be handed to it by message, from scheduling a meeting to, in Anthropic’s words, “building dashboards with dozens of tools, creating financial or sales forecasts, prototyping, and building web apps”.

The agent runs on its own virtual machine with its own browser, so the user’s laptop or phone does not need to stay on while it works, Anthropic says. It checks connected accounts every few hours rather than continuously, because, as the company puts it, “checking nonstop would use up your plan’s limits much faster”. It carries over memory, skills and connected apps from the user’s existing Claude account, and users can message it “through typing or voice”.

Duet is powered by Opus 5.5, according to the product page, and works with more than 6,000 connectors and plugins. The Next Web listed Gmail, Slack, QuickBooks, Stripe, HubSpot and Shopify among them.

Approvals and limits on what it may do

Anthropic says the agent reads, drafts and prepares on its own, but “before it sends a message, makes a payment, signs a document, or deletes anything, it shows you exactly what it’s about to do and waits for your approval”. Some actions are off-limits even with approval, including trading securities or moving money between accounts.

Purchases are possible through Link, Stripe’s digital wallet, which asks the user to approve the spend and pays with a virtual card so the real card number is not shared, according to the announcement. Duet can also use Shopify’s Shop Pay at online stores, and Anthropic says signing in with logins saved in 1Password is “coming soon”. Users can disconnect any app Duet has access to.

Privacy and safety design

Anthropic says Duet draws on its existing agent safeguards, including prompt-injection defences, and that red teams and an external firm penetration-tested it. The company is also shipping a new classifier and safety system prompt that checks each message for leaks of sensitive information. By default, it is designed to reject requests to share the strictest tier of data, such as credit card numbers and government identification; health or personal-finance details are passed on only when the user explicitly instructs it to send them. Anthropic says Duet’s handling of credentials follows the design of its Managed Agents, so it “never directly sees your passwords or sign-in tokens”. The product page adds that approvals, logins and safety checks run in the user’s own cloud computer, so sensitive details stay private. Anthropic said it would publish more about the safety design “in the coming days”.

Known limitations

Anthropic itself lists several things that are not ready. It says Duet “still might misread a thread or miscalibrate what’s urgent”, can miss something that arrives between check-ins, is weaker at personal errands such as booking tickets, and cannot yet show the user its browser. TechCrunch also reported the weaknesses with personal errands and browser visibility. The Next Web reported that early reviewer Dan Shipper of Every found its email drafts weaker than a competing agent’s; “There’s no real voice mode yet, just transcription,” Shipper wrote.

Availability, plans and usage

Duet is available as a standalone app on iOS and Android and on the web at claude.ai, Anthropic says. During the research preview it is included in Pro and Max plans; Team and Enterprise plans are not yet covered. The Next Web reported there is no extra charge, but work done by Duet counts against the same usage limits as the rest of Claude. Larger tasks, such as building a financial model or a deck, use more, Anthropic says.

Engadget’s wording differs on one point: it said Pro subscribers, who pay $20 a month, would get access “once the agent rolls out in general availability”, while Anthropic’s own pages say the preview itself is included in Pro and Max plans, with a waitlist to get in. For readers in Romania and elsewhere outside the US, the preview is not currently offered, and neither the announcement nor the product page gives a date for other countries.

How it compares with rivals

TechCrunch framed Duet as Anthropic’s entry into a crowded field, naming Meta Muse and OpenAI Dots among the rivals; Anthropic pitches Duet at work rather than everyday personal tasks. Engadget pointed out that Dots is currently available only on OpenAI’s Pro tier, which starts at $100 a month, whereas Anthropic is putting Duet in its $20 Pro plan. The Next Web reported that OpenAI introduced Dots in September and that Meta has launched Muse. Not all coverage was warm: Engadget wrote that Anthropic “is late to the party” and that for some, “sharing their entire digital life with an AI agent, is just too much to ask.”

What is still unknown

Anthropic’s announcement and product page do not give post-preview pricing, dates for Team or Enterprise access, international rollout dates, a general-availability date, data-retention periods or numerical usage allowances.

Header photo: Man with smartphone and laptop (Unsplash) by Alejandro Escamilla alejandroescamilla, CC0, via Wikimedia Commons, cropped.

We use AI in our daily operations; posts are reviewed by people before publishing; automated/generated posts are marked as such where needed.