Anthropic is introducing Claude Duet, an AI agent designed to handle work across connected apps. Aimed at professionals and small business owners, it is launching as a research preview and is available through a standalone app on iOS and Android, as well as Claude on the web.
A daily brief and help with longer tasks
Claude Duet is built around Claude Opus 5.5. Anthropic says it can send a short morning brief covering important items across connected apps, then keep watch during the day and message the user when something needs attention.
Users can also assign it work, from scheduling meetings to building dashboards that use dozens of tools. Anthropic lists financial or sales forecasts, prototyping, and building web apps among the more involved tasks it can handle. The pitch is that an agent can move between routine administration and larger projects while keeping an eye on incoming information.
That could make the agent useful when work is spread across multiple services. A brief can bring relevant items together, while ongoing monitoring may help surface a thread that needs a response. More complex assignments suggest Claude Duet is intended to do work, not only answer questions.
Connections, memory, and safeguards
Claude Duet can use memory, skills, and connected apps from Claude. It also has a virtual machine with a browser for completing tasks, giving it tools to interact with information and services as it works.
Some actions require a pause for user approval. Anthropic says the agent will ask before major steps such as sending a payment or deleting something. The approval prompt matters because the agent may encounter sensitive information or make changes with real consequences.
Anthropic says Claude Duet has a dedicated classifier and safety system prompt intended to reject requests to share the strictest tier of information, including credit card numbers. These protections are part of the product’s design, though the company also describes the current release as an early version.
Early access comes with limits
Anthropic cautions that Claude Duet may misread a thread or misjudge what is urgent. Those are meaningful limitations for an agent that monitors connected apps and decides when to interrupt a user. Its usefulness depends not just on completing a task, but on understanding the surrounding context well enough to act at the right time.
The company also says the agent is not as strong at personal errands such as booking tickets. That distinction suggests its initial focus is workplace support, rather than acting as a general assistant for every part of a user’s life.
Claude Duet is available by waitlist, and access requires a Pro or Max subscription in the US. As a research preview, it offers a look at how Anthropic wants AI agents to combine app monitoring, administrative work, and more elaborate projects, while asking users to approve consequential steps.