Claude Duet Is Anthropic’s Answer to OpenAI’s Dots and Meta’s Muse

Anthropic has launched Claude Duet, a proactive AI agent that messages you before you message it, roughly two months after Meta and OpenAI kicked off the race for always-on AI agents. The app opened as a research preview on October 21 for Claude Pro and Max subscribers in the US. It sits behind a waitlist and runs on iOS, Android, and the web. Anthropic Labs co-lead Mike Krieger described it on X as the company’s “take on giving everyone a trusted partner with your work.”
That pitch is narrower than it sounds. Duet runs on Claude Opus 5.5, and Anthropic’s launch post says it was built for “professionals and small business owners with more on their plates than they can track.” It isn’t chasing Meta’s consumer crowd. After going through Anthropic’s announcement, its support documentation, the early press coverage, and the first hands-on reports, we found the reasoning behind that choice, and its trade-offs, are easier to see than the launch hype suggests.
Claude Duet starts your day with a brief, then goes quiet
The core idea is simple. Every morning, Duet reads across the apps you’ve connected, mainly your inbox and calendar, and sends a short brief on what actually needs you. After that, it checks those apps every few hours and stays silent unless something deserves your attention. It’s closer to a chief of staff than a chatbot.
You don’t open new chats with it. Everything happens in one running thread, typed or spoken, and you can spin off channels for separate topics that still share the same memory. Your Claude memory, skills, and connectors come along from day one, so a new user isn’t starting from zero. Work in progress lives in a Tasks tab, while finished files such as decks, briefs, and financial models land in a Make tab.
Anthropic’s best example from early testing is a small one, which is exactly why it works. A real estate broker’s morning brief flagged a property showing booked on her calendar for the wrong day. She never asked Duet to look for scheduling mistakes. It noticed anyway.
Duet inherits every app already connected to your Claude account. Before your invite arrives, go through your Claude connector settings and remove anything you wouldn’t want an always-on agent reading.
Invoices, payroll, and contracts are the real pitch
Read Anthropic’s list of supported jobs and a pattern jumps out. Duet can check QuickBooks or Stripe for unpaid invoices and draft polite reminders. It can prepare payroll in Gusto with hours and bonuses filled in, get agreements ready for signature in DocuSign, and draft follow-ups using deal history in HubSpot. It can also write a job post for LinkedIn, turn a launch into a Canva campaign, or manage a Shopify store, drawing on more than 6,000 connectors and plugins.
One early tester runs a South Florida restaurant group with about 110 employees and used Duet to work through invoices store by store, alongside accounting and customer support work. That’s the customer Anthropic is chasing: someone with a dozen open tabs and no assistant.
Duet can also buy things for the business. Purchases go through Stripe’s Link or Shopify’s Shop Pay, and each one uses a one-time virtual card that only works for that order, so your real card number never changes hands.
Claude Duet rides on a $20 plan, while Dots starts at $100
On a technical level, the three big agents are close cousins. Each one runs on its own virtual machine with a browser, so it can keep working after your phone is locked. The overlap was obvious enough that one widely shared post on X greeted the launch with a simple “Dot, meet Duet.”
The clearest difference is the price of entry. As Engadget points out, Duet comes with Claude’s $20 per month Pro plan. OpenAI Dots is currently limited to ChatGPT’s Pro tier, which starts at $100 per month. Meta Muse went the consumer route instead, and TechCrunch reports Meta says it has passed 10 million users, with shopping among the most popular activities.
Not everyone is cheering for a third option. In the comments under MacRumors’ launch story, the most upvoted reactions were tired rather than excited. One reader called the push to hand everything to agents exhausting, while another noted that four companies had shipped nearly identical products within weeks of each other.
Anthropic admits it isn’t great at booking tickets
The most revealing line in Anthropic’s announcement is an admission. The company says Duet is weaker at personal errands like booking tickets, and that it can’t yet show you its browser while it works. Launch posts don’t usually volunteer weaknesses.
The admission makes sense once you look at who already pays Anthropic. Meta can put Muse in front of billions of people through its apps. Anthropic’s strength is a base of developers, knowledge workers, and businesses already using Claude, Claude Code, and Cowork. Duet isn’t trying to be the AI personal assistant that orders your dinner. It’s trying to be the one that catches the overdue invoice before your accountant does.
Nothing gets sent, paid, signed, or deleted without your yes
Duet reads, drafts, and prepares work on its own. The line is drawn at four actions: sending a message, making a payment, signing a document, or deleting anything. Before any of those, it shows you exactly what it’s about to do and waits. If you can’t approve right away, it finishes the rest of the task and brings the approval back later.
Some things stay off the table even with your approval. Duet won’t trade securities, move money between accounts, or access someone else’s account. When a website asks for a password, you type it into a separate form on your own device, and it goes straight to Duet’s browser without passing through the chat. A new classifier screens every outgoing message and refuses to share the most sensitive information, like card numbers and government ID. Health and financial details only go out if you explicitly tell Duet to send them.
Anthropic is also candid about the hardest problem. Duet reads email and web pages written by strangers, and those can hide instructions meant to manipulate it. The company says no defense against that kind of prompt injection is perfect, which is a big part of why the approval step exists. Red teams and an outside security firm tested Duet before launch, and it’s covered by Anthropic’s public bug bounty.
Give Duet read-and-draft work first, like inbox triage or invoice reminders. Move on to payments and signatures only after it has earned your trust on jobs where a mistake costs nothing.
The quiet cost hiding in your Claude limits
Here’s the detail most launch coverage skipped. Duet doesn’t come with its own allowance. It draws on the same five-hour session and weekly limits as everything else you do in Claude, Claude Code included. Engadget put it bluntly, saying that sounds like “a recipe for the new agent to blow through your rate limits.”
Anthropic clearly saw this coming. According to its Get started with Claude Duet guide, the morning brief and background check-ins count toward your limits too, which is exactly why Duet checks in every few hours instead of constantly. The trade-off is that something urgent can land between checks and sit there. Early discussion on Reddit suggests the bigger drain will come from heavy tasks you approve, like building a deck or a financial model, rather than the check-ins themselves. That commenter was still waiting for access, though, so treat it as an educated guess.
To ease people in, Anthropic gives everyone admitted to the preview one full limit reset covering both the five-hour and weekly limits. It’s valid until November 30, 2026.
Don’t spend your one-time reset on day one. Save it for the week you test Duet’s heaviest jobs, like a full slide deck or a financial model, since that’s when it will draw on your plan the hardest.
What ten hours with Claude Duet turned up
Hands-on reports are starting to arrive, and interest is high. One breakdown video from an AI channel aimed at non-technical users has passed 274,000 views in two days. The most detailed test so far comes from YouTuber Riley Brown, who spent about 10 hours using the app for his business, content, and presentations.
His takeaways line up with Anthropic’s design choices. He found the interface snappy and leaned heavily on channels to keep work organized. He used voice to edit slide decks and noted that every task gets its own thread. His most interesting observation is that Duet feels a lot like Claude Projects, only proactive. He also set aside a full section for the negatives, a reminder that this is very much a preview.
Who gets in first, and who keeps waiting
Access requires three things lining up: a Pro or Max plan, a US Claude account, and an invitation from the waitlist. Invitations go out in waves with no promised date. The most common snag is a mismatched email. Access is tied to the address you gave the waitlist, and it has to match both your Claude account and the Apple or Google account on your phone.
Search interest from the UK, Canada, and Australia is already climbing, but Duet is US-only for now, and Team and Enterprise plans are still on the “later” list. The preview has other gaps too. You can’t choose a model, Duet won’t make phone calls or text other people, Microsoft 365 isn’t supported, and Slack works with restrictions.
None of those gaps will decide whether Duet works. Two things will: how good its judgment is about what deserves your attention, and how much of your Claude plan it quietly uses up along the way. Anthropic has bet that a focused, careful agent beats a flashy one. The waitlist is where that bet gets tested.
Frequently asked questions
It's Anthropic's proactive agent app, released as a research preview on October 21, 2026. It sends a morning brief, checks your connected apps every few hours, takes on tasks you hand it by message, and keeps working in the cloud after you close the app. It asks for approval before sending, paying, signing, or deleting anything.
Not yet. During the research preview, Duet is limited to Claude accounts in the US. Users in the UK, Canada, Australia, and elsewhere can't get access for now, and Anthropic hasn't announced a date for other regions.
There's no separate charge during the preview, but there's no free version either. Duet is included with Claude Pro and Max plans and uses those plans' usage limits, including its background check-ins. You can join the waitlist at https://claude.com/product/claude-duet.
Both run on their own cloud computer and ask before taking important actions. Duet is included with Claude's $20 per month Pro plan and is built around professional and small business work. Dots is currently limited to ChatGPT's Pro tier, which starts at $100 per month.
Invitations go out in waves, so there's no set date. If you're still waiting, check that your waitlist email matches your Claude account and the Apple or Google account on your phone, because access is tied to that address. Team and Enterprise accounts, and anyone outside the US, can't get in yet.
Related posts

Video-to-Video AI Troubleshooting: Eight Common Problems and What to Change
Video-to-video AI troubleshooting: eight common problems, from drifting motion to soft output, each with its likely cause and a fix.
Read more
How Good Is Gemini 4 Carbon? Benchmarks, Access, and Real Cost
Gemini 4 Carbon ranks fourth of 80 models in independent tests, but it is slow, wordy, and its API price doubles on December 1. See why.
Read more
One Agent, Two Runs: One Large Model for All Steps, or One Chosen per Step
Two runs of the same agent show whether a model per step saves money without hurting speed or quality.
Read more