Anthropic launches Claude Duet AI agent in research preview

Anthropic has announced Claude Duet, a new AI agent aimed at professionals and small business owners, according to The Verge. The product matters because it moves Claude beyond a chat window and toward a cross-platform assistant that can watch connected apps, summarize what needs attention, and take on work with user approval.

The company is positioning Duet against other agent products, including Meta’s Muse and OpenAI’s Dots, The Verge reports. It is launching as a research preview, which is a signal that Anthropic is still testing how much responsibility users should hand over to an AI assistant that can act across tools.

What Claude Duet is designed to do

Claude Duet is built around Claude Opus 5.5, according to Anthropic. The agent can send a short morning brief based on important items across connected apps, then continue watching those apps during the day and alert the user when something needs attention.

Anthropic says users can assign Duet a range of tasks. The examples given span routine admin work, such as scheduling meetings, as well as more involved projects like creating dashboards across many tools, making financial or sales forecasts, prototyping, and building web apps.

For creators, marketers, freelancers, and small businesses, the most important shift is not just that Duet can answer questions. It is designed to sit closer to the flow of work: inboxes, messages, apps, and browser-based tasks. That could make it useful for people who spend much of the day triaging requests, checking updates, or coordinating client work across several platforms.

A standalone app with Claude context

Duet is available as a standalone app on iOS and Android and through Claude on the web, according to The Verge. It also pulls in memory, skills, and connected apps from Claude, which means Anthropic is tying the new agent to the broader Claude environment rather than presenting it as a completely separate product.

Anthropic says Duet has a virtual machine with a browser that it can use to complete work. That browser-based setup is important because many small-business workflows still happen inside web apps rather than through formal software integrations.

The practical implication is that Duet may be most useful where a task requires multiple small steps: reading context, moving between tools, preparing an output, and checking back with the user. It also puts Claude closer to the same broad agent trend we have seen in workplace AI, including Google’s move to bring a Gemini agent across apps in the enterprise market (read our coverage).

Guardrails and approval steps

Because Duet may handle sensitive information, Anthropic says it asks for permission before major actions. The examples in The Verge’s report include sending a payment or deleting something.

The company also says Duet has a dedicated classifier and safety prompt intended to block requests to share the most sensitive tier of information, including credit card numbers. For a work assistant that monitors apps and messages, that safety layer is central: users may want speed, but they also need to know when the system will stop and ask.

For small teams, this is where the product’s usefulness will likely be judged. An agent that can draft, summarize, schedule, and build is valuable only if its approval points are clear enough for everyday use. Too many interruptions reduce the benefit; too few create risk. Anthropic’s own description suggests it is trying to keep high-impact actions under human control.

This also fits with the broader direction of Anthropic’s Claude product line, where the company has been expanding from chat into more structured work patterns, including dynamic workflows for parallel agents (read more).

Who can try it, and what to watch

Claude Duet is available through a waitlist, but The Verge reports that users need to be Pro or Max subscribers in the US to use it. Anthropic describes the current version as early and says it may still misunderstand a thread or misjudge what is urgent.

The company also says Duet is not as strong at personal errands such as booking tickets. That is a useful boundary for buyers: Duet appears to be aimed more at work coordination and business tasks than consumer concierge chores.

For Emonarc readers, the near-term question is whether Duet can reduce coordination overhead without adding supervision work. Freelancers may care about client follow-ups and scheduling. Small businesses may look at dashboards, forecasts, and admin tasks. Marketers may be interested in daily briefs across campaign tools and messages.

But because this is a research preview, it is too early to treat Duet as a replacement for established workflows. The sensible approach is to test it on low-risk tasks first, watch how often it asks for approval, and check whether its summaries and urgency judgments match how your business actually works.