Anthropic launches Claude Duet, an AI agent for professionals

Anthropic has introduced Claude Duet, an AI agent built for professionals and small business owners, launching as a research preview. The move puts Anthropic directly against Meta’s Muse and OpenAI’s Dots, as major AI labs race to ship agents that can act on a user’s behalf rather than just answer questions.

Claude Duet is available as a standalone app on iOS and Android, and through Claude on the web. It is built around Claude Opus 5.5, Anthropic’s model. According to Anthropic, the agent sends a short brief each morning covering the most important items across a user’s connected apps. During the day, it keeps watch on those same apps and only messages the user when something needs attention, rather than sending constant updates.

Anthropic says the agent can also take on direct requests, ranging from simple administrative work like scheduling meetings to more involved jobs such as building dashboards that pull from dozens of tools, producing financial or sales forecasts, prototyping, and building web apps. To do this, Claude Duet draws on memory, skills, and connected apps already tied to a user’s Claude account, and it runs a virtual machine with its own browser to carry out tasks.

Guardrails before sensitive actions

Because the agent can take real-world actions, Anthropic has built in an approval step. Before anything significant, such as sending a payment or deleting a file, Claude Duet messages the user and waits for a go-ahead rather than acting on its own. Anthropic also says it added a dedicated classifier, a filter that screens requests, along with a new safety system prompt, instructions that guide the model’s behavior, both designed to reject requests to share what Anthropic calls the strictest tier of information, including credit card numbers.

Anthropic is upfront that Claude Duet is an early version. The company says the agent still might misread a thread or miscalibrate what counts as urgent. It also says Duet is not as strong yet at personal errands, such as booking tickets, which suggests the current focus is squarely on work tasks rather than personal life admin.

Access is limited for now. Anthropic is running a waitlist for people who want to try Claude Duet, and the agent is restricted to Pro or Max subscribers in the US. Anthropic has not said when or whether that access will widen to other regions or subscription tiers.

The launch frames Claude Duet as Anthropic’s answer to a growing field of always-on agents from rival labs, each betting that professionals and small business owners will hand over day-to-day digital busywork to software that can watch, decide, and act with limited human sign-off. What remains to be tested is how Claude Duet performs outside controlled demonstrations, particularly around the misread threads and urgency misjudgments Anthropic itself flagged, and whether the approval step holds up once the agent is handling real payments and real deletions at scale.