Anthropic has launched Claude Duet, a proactive agent that works from a user’s connected apps rather than waiting to be prompted. Duet sends a short brief each morning on items that need attention, checks the connected apps every few hours, and messages the user only when something requires a decision. It runs on its own virtual machine with a dedicated browser, so it keeps working in the background without a local computer or an open Claude session.
The task range Anthropic claims runs from scheduling meetings and drafting replies to building dashboards, financial and sales forecasts, prototypes and web apps. Duet inherits the memory, skills and app connections already stored in a Claude account, works in a single ongoing thread, and connects to the Claude Marketplace’s catalogue of tools and plugins, which Inc42 put at more than 6,000, including Gmail, Slack, QuickBooks, Stripe and DocuSign.
The guardrails are part of the pitch. Duet asks for approval before sending a message, making a payment, signing a document or deleting a file, and Anthropic says it does not see passwords or logins when given account access, with a dedicated classifier checking messages against leaks of sensitive information. Access is a research preview, by waitlist, for Pro and Max subscribers in the United States, on the web and through standalone iOS and Android apps, and background checks count against existing plan limits.
The launch completes a three way race. OpenAI’s Dots arrived several weeks ago and Meta’s Muse has grown quickly, reaching the download milestone covered on this site yesterday, the same day its sharing feature was paused after researchers enumerated 6,800 shared chats through public DNS. Anthropic is entering that market with the approval step as its headline feature, a direct response to the trust problem the Muse episode exposed.