Anthropic’s Claude Duet tries to act before a user opens a chat. After receiving access to email, calendars, Slack and Google Docs, the consumer-facing agent surfaces suggested tasks as cards, drafts messages and highlights conflicts across workplace systems.
In a week-long test, Duet connected a meeting summary in email with a Slack conversation and noticed that scheduled time off left little room before a writing deadline. It proposed updating a shared document and drafted a message that closely matched the reviewer’s informal style. Messages it sends carry a “sent with Claude” label.
The same proactive design created false urgency. Duet treated a tentative publication date as a hard commitment and pressed for action before the editor’s local workday had begun. The example shows the difference between finding relevant facts and correctly interpreting tone, time zones and workplace expectations.
Duet is a research preview for US subscribers on Anthropic’s Pro, Max and Ultra plans, with access controlled by a waitlist. Its usefulness grows when users connect more private communications, which also increases the amount of sensitive context it can inspect. Anthropic says it is still testing when the agent should interrupt. For now, the strongest role appears to be an organizational backstop whose suggestions remain optional, not an autonomous manager of a person’s work.