Anthropic takes on Muse with Claude Duet, its new AI agent
Oct 21, 2026, 12:53 PM · The Verge
Anthropic's answer to Muse is pitched at people running small businesses, and its most interesting feature is how often it promises to stop and ask.
Why it matters
Anthropic launched Claude Duet, an agent built on Claude Opus 5.5 for professionals and small business owners, as a research preview on iOS, Android and the web. It sends a morning brief across connected apps, watches them during the day, and takes on tasks from scheduling to forecasts, dashboards and simple web apps, using a virtual machine with a browser.
With Meta's Muse and OpenAI's Dots already out, the major labs now each have a personal agent in market. The competition is no longer whether agents work, but whose design people will trust with their accounts.
From the desk
We think Anthropic picked a smart lane. Muse is leaning into consumer shopping. Duet is aimed at work: the inbox triage, the calendar shuffle, the sales forecast a small business owner never has time for. That's where an agent can save real hours, and it's a place where carefulness is a selling point rather than a drag.
The guardrails are the story. Duet messages the user for approval before big steps like sending a payment or deleting something. Anthropic also says a new dedicated classifier and safety system prompt are designed to reject requests to share the most sensitive tier of information, such as credit card numbers. That's the right instinct for an agent sitting on top of email and business tools, where one misread message could leak data or move money.
Anthropic is also unusually candid. It says Duet is an early version that still might misread a thread or misjudge urgency, and that it's weaker at personal errands like booking tickets. I'd rather hear that up front than discover it later. But candor isn't a control. An agent that misjudges urgency can bury the one email that mattered, and approval prompts only help if people actually read them instead of tapping yes.
Access is narrow for now: a waitlist, US only, Pro or Max subscribers. That limits harm while the product matures, and it also means early impressions will come from power users, not the small shops it's ultimately pitched to. The real test is whether Duet stays cautious once it's competing on convenience against agents that ask fewer questions.
Context
Duet draws on memory, skills and connected apps from the main Claude product. Anthropic describes it as messaging users only when something needs attention, rather than streaming constant updates. The launch lands as Meta reports more than 10 million Muse users and OpenAI pushes its own Dots agent.
Who feels it
- Small business owners
- A potential daily assistant for briefings, scheduling and forecasting, with approval gates on payments and deletions. Still a preview that can misread priorities.
- Enterprises and IT
- Another agent with broad app connections arriving through personal subscriptions, which raises questions about data access policies at work.
- Rival agent makers
- Duet sets an expectation that agents ask before consequential actions and refuse to share top-tier secrets, which could become a baseline for the category.
What to watch
- How fast Anthropic opens the waitlist and expands beyond the US
- Reports of Duet misjudging urgency or acting without clear approval
- Whether its sensitive-data classifier holds up under adversarial testing
- Pricing or usage limits once it leaves research preview