Anthropic's Claude Duet is an AI agent that messages you first and asks before it acts
Anthropic has opened a waitlist research preview of Claude Duet, a new app in which Claude works as a proactive agent. It sends you a morning brief, checks your connected apps every few hours, keeps working on its own cloud computer, and asks for your approval before it sends a message, makes a payment, signs a document or deletes anything.
For most of the short history of AI assistants, the deal has been the same: you open the app, you type a prompt, and the model answers. Nothing happens until you ask. Anthropic now wants to flip that direction.
On Wednesday, 21 October 2026, the company opened a research preview of Claude Duet, a new app in which Claude works as what Anthropic calls a proactive agent. In Anthropic's own words, it "comes to you with what needs your attention, completes tasks you ask of it, keeps working even when you step away, and checks with you before it acts on your behalf." The preview is available through a waitlist, starting with Claude Pro and Max subscribers in the United States. CNBC and VentureBeat both covered the launch the same day and framed it as Anthropic's answer to Meta's Muse and OpenAI's Dots.
The three things you need to know
- It comes to you: Every morning Claude Duet sends a short brief on the most important items across your connected apps. During the day it checks those apps every few hours and messages you only when something needs your attention. It runs on its own virtual machine with its own browser, so it keeps working when your phone or laptop is off.
- You stay in charge: Duet reads, drafts and prepares work on its own. But before it sends a message, makes a payment, signs a document or deletes anything, it shows you exactly what it is about to do and waits for your approval. Some things, like trading securities or moving money between accounts, are off limits even if you approve them.
- The catch: It is a waitlist research preview for US Pro and Max subscribers only, with no Team or Enterprise access yet. It runs on Claude Opus 5.5, and its background checks count toward the same usage limits you already use for the rest of Claude.
What Claude Duet actually is
Claude Duet is a standalone app on iOS and Android, and it is also available on claude.ai. It is not a new model. It is a new way of working with an existing one. Anthropic says it designed Duet around Claude Opus 5.5 and built it to decide "what needs you, what can wait, and when to check with you first." According to the Claude Help Center, you cannot choose a different model in the preview. Duet runs on Opus 5.5 and decides by itself how much effort a job needs.
The target group is clear. Anthropic says it built Duet "for professionals and small business owners with more on their plates than they can track." That is a deliberate choice. Meta has mostly marketed Muse toward consumers, and Anthropic itself admits that personal errands such as booking tickets are not "quite as strong" yet, because the core functionality was built around work.
Setting it up is meant to be quick. Duet carries over your memory, skills and connected apps from your existing Claude account. You talk to it in one ongoing thread, by typing or by voice, so a quick question and a project that runs for a week start in the same place. The Help Center adds that you can open separate channels for different topics, and that they all share the same memory.
A day with Duet
The easiest way to understand the product is to walk through a day the way Anthropic describes it.
In the morning, Duet sends a short brief with the most important items from your connected apps. The Help Center says it looks across your email and calendar, picks what most needs your attention, and can start the first steps, for example by drafting replies for you to review.
During the day, it keeps an eye on those apps. This is where the details matter. Duet does not watch your accounts continuously. It checks them every few hours. Anthropic is open about why: "checking nonstop would use up your plan's limits much faster." You can tell it what you want to hear about and how often. The company also says it is still testing Duet's judgment about when to interrupt you and will keep tuning it based on feedback.
Whenever you want, you can hand it a task with a short message. Anthropic lists simple admin jobs such as scheduling meetings, but also larger ones: building dashboards with dozens of tools, creating financial or sales forecasts, prototyping and building web apps. Duet works out which apps to use and which steps to take, then notifies you when the task is done or needs your review. Bigger results come back as files, such as a research brief, a financial model, a slide deck or a dashboard, which appear in a Make tab.
In the background, all of this runs on Duet's own virtual machine with its own browser. That is the key technical difference from a normal chat app: your laptop and phone do not need to be on. For websites that have no connector, Duet uses that cloud browser.
Who decides what
The part of the announcement that will matter most to anyone thinking about giving an AI agent access to email, invoices or payroll is the approval model. Anthropic describes three levels.
On its own, Duet reads your connected apps, drafts replies and prepares work such as documents, decks, sheets and dashboards.
Only with your approval can it send a message, make a payment, sign a document or delete anything. In those cases it shows you exactly what it is about to do and waits.
Never, even with your approval, does it trade securities, move money between accounts or access someone else's account. You can also disconnect any app Duet has access to at any time.
Anthropic lists several safety layers behind this. Duet's handling of credentials follows the design used for Claude Managed Agents, so it never directly sees your passwords or sign in tokens. Its actions are approved by Anthropic's auto mode classifier, the same safeguard used for Claude and Claude Code. Its defenses against prompt injection, meaning hidden instructions planted in emails or web pages, come from Anthropic's browser use research. Red teams and an external firm penetration tested the product, and it is covered by Anthropic's public bug bounty program.
There is also a new piece built for Duet: a dedicated classifier and safety system prompt that checks each message and guards against leaks of sensitive information. It is designed to reject requests to share the strictest tier of data, including credit card numbers and government identification. Other sensitive information, such as health or personal finances, only goes out when you have explicitly told Duet to send it. Anthropic says it will share more details about Duet's safety design in the coming days.
The connectors: where it gets interesting for businesses
Duet can use the more than 6,000 connectors and plugins in the Claude Marketplace. Anthropic's examples read like a small business to do list:
- Find the Gmail and Slack messages you missed and help you reply.
- Check Intuit QuickBooks or Stripe for unpaid invoices and draft reminders.
- Prepare and run payroll in Gusto, with hours, bonuses and reimbursements filled in.
- Draft a job posting and, after you edit and approve it, post it to LinkedIn.
- Prepare agreements and send them for review and signature through Docusign.
- Draft follow up emails using the contact and deal history in HubSpot.
- Turn a product launch into a social campaign in Canva.
- Set up, design and manage a Shopify store.
Duet can also buy things for your business. With Stripe's Link wallet, it confirms the item with you, Link asks you to approve the spend, and the payment goes through a virtual card so your real card number is not shared. It can also check out with Shop Pay from Shopify. Coming soon, with your permission, it will be able to sign in to sites using logins saved in 1Password.
The Help Center lists the current limits honestly. Some connectors, such as Slack, work with limits in Duet, and Microsoft 365 is not available yet. Duet does not make phone calls or send text messages to other people. With your approval, it can send emails from the accounts you connected. It cannot see the files on your computer, and it can read your work in Claude but not change it.
How testers use it
Anthropic shared a few examples from early users. Real estate broker Jess Provencher gets a rundown each morning, and one day Duet flagged, without being asked, that a property showing was on her calendar for the wrong day. Amin Ebrahimi, who owns the Shahs of Kabob restaurants in South Florida with about 110 employees, used it for accounting, customer support and marketing. "I don't need to open up 100 tabs," he said. Brand strategist Mya Dalton says Duet prepared a brief for a client call and noticed the call was right before her flight, then offered to move it.
Inside Anthropic, Claude Code lead Boris Cherny talks through his mental to do list with Duet on walks, and the team that builds Duet uses it to schedule user interviews, build prototypes and even ship code. As VentureBeat points out, these are company selected stories, not independent tests.
The race for personal agents
Duet does not arrive in an empty market. CNBC reports that Meta's Muse, which Meta calls a personal AI agent, launched last month and rose to the top of Apple's App Store. On Wednesday, Mark Zuckerberg told analysts that more than 10 million people in the US and Canada have set up Muse personal agents over the past month and a half. OpenAI introduced its always on agents, called Dots, at its annual developer conference a few weeks later.
VentureBeat compared the three. Muse is reached through its app or WhatsApp and focuses on everyday tasks, purchases and travel, with a free entry point. Dots are powered by GPT-6 Astra, work through ChatGPT and messaging channels such as Slack and Teams, and start at $100 per month for an individual Pro subscription, with an Enterprise beta that administrators can enable. Duet's entry point is the $20 per month Claude Pro plan, but it has no Team, Enterprise or API access yet. VentureBeat also notes that human approval and cloud isolation are not unique to Duet, because Muse also asks permission for sensitive actions and runs after the app is closed.
So the differences are less about raw capability and more about focus. Duet bets on work: invoices, payroll, contracts, client calls. Muse bets on everyday life. Dots sit somewhere in between, with a clearer path into large companies.
The fine print on usage
The point that is easy to miss is cost. During the preview, Duet shares your plan's usage limits with the rest of Claude, and it uses them differently than a normal chat. Its background checks count toward those limits, and larger jobs like a financial model or a slide deck need more model turns and background work. There is no separate charge for Duet, but every check it runs on your behalf is drawn from the same budget you use for chatting and coding. VentureBeat notes that Anthropic gave no task level cost estimates, productivity numbers or benchmark results.
Anthropic is also clear that this is an early version. Duet "might misread a thread or miscalibrate what's urgent," it can miss something that arrives between check ins, and it cannot yet show you its browser, so you cannot watch it work or take over. For now it lives in its own app, separate from the main Claude app.
Why it matters
Duet is a small product launch with a big idea behind it. Chat assistants are reactive tools. Agents that watch your inbox, notice problems and start the work before you ask are something different: they behave more like a junior colleague than a search box. That makes the boring details the important ones. How often does it check? What can it do alone? What does it need you for, and what will it refuse even if you say yes?
Anthropic's answers are fairly conservative. Duet drafts freely but asks before anything leaves your hands, it keeps passwords out of the model's view, and it refuses some financial actions outright. That is a sensible starting point for a product that can touch payroll and payments. The bigger question is whether people will find an agent that interrupts them useful or annoying, and whether the time saved is worth the usage limits it burns in the background.
Dany's take
I like the direction, and I like that Anthropic is honest about the trade offs. A morning brief that catches a showing on the wrong day or a call right before a flight is exactly the kind of help most busy people actually want. The approval model also feels right to me: let the agent read and draft, but keep the send, pay, sign and delete buttons with the human. What I am less sure about is the usage budget. If background checks eat into the same limits I use for real work, I will have to decide how proactive I can afford my assistant to be. I also want to see the browser. An agent that buys things and sends emails on my behalf should let me watch what it is doing. For now this is a US only waitlist, so most of us will watch from the sidelines, but the race between Duet, Muse and Dots is clearly the next big battle in AI. Would you let an AI agent message you first? Tell me in the comments under the video.
What this means for you
If you are a US Pro or Max subscriber, you can join the waitlist now. Access opens in waves. If you try it, start with bounded tasks whose results you can easily check, such as the morning brief, draft replies or meeting coordination, before you connect payroll or payments. Review every approval request carefully, because the preview can misread a thread. Disconnect apps you do not need it to touch. And keep an eye on your usage, since every background check counts. If you are outside the US or on a Team or Enterprise plan, there is no date yet, so watch for Anthropic's promised safety details and the first independent reviews.
Sources
Source: Anthropic