A week with Claude Duet: good at catching what falls between apps, shaky on tone and urgency

Anthropic launched Duet in October as its agent for professionals and small business owners (see our launch coverage). It remains a research preview limited to U.S. subscribers on the Pro, Max and Ultra plans, with a waitlist. For this hands-on, The Verge’s reviewer connected it to work email, calendar, Google Docs and Slack and used it for a week.

That gives one of the clearer pictures yet of how a personal agent behaves for someone who doesn’t write code. Duet is designed as the most consumer-friendly of Anthropic’s three agent products, alongside Claude Cowork and Claude Code, with its own app and swipeable suggestion cards.

The standout moment was a scheduling catch. Duet noticed that upcoming time off would leave only a couple of working days before a big deadline, proposed a fix, offered to update the shared document and drafted a message to the editor. Nobody had asked it to look. That is the case for a proactive backstop: the cross-checking that slips through the cracks when work is spread across five apps. It also connected a meeting summary to a related Slack message, the kind of link a busy person misses.

This lines up with what Anthropic is building more broadly. The same week, it launched Claude Meetings, Sheets and Whiteboards to pull non-developers into Claude (our analysis of the office-suite push). Duet is the consumer edge of that strategy.

The misses are just as instructive. Duet treated a soft promise to share a link as a firm deadline and nudged before the editor’s workday had even started. It mimicked the reviewer’s all-lowercase Slack style convincingly but did not really grasp tone or what mattered. It also volunteered to draft creative work that the reviewer wanted to write herself. Anthropic says it is still testing when the agent should speak up. The reviewer called it slightly neurotic, which is fair. An over-eager agent spends attention instead of saving it.

Duet sees every email, draft and Slack message because the user opted in. The review describes the difference between knowing that in principle and watching the agent work out who your boss is from a shared document. Scale that across a workplace and the harder questions involve the colleagues and clients whose messages an agent now reads without their consent, and how much an employer or vendor could infer from all that context. Meta faced similar trust questions when it launched Muse (Will consumers trust it?).

One design choice deserves credit: messages sent through Duet carry a label saying they were sent with Claude. As agents draft more everyday communication, people should know when they are reading a person and when they are reading a machine’s draft.

The reviewer wonders whether leaning on Duet would leave her disconnected from her own work. That is the right worry. Agents that catch gaps make people better at their jobs. Agents that take over the noticing, remembering and writing could make them worse. Builders should design for the first. More on the category in our agents coverage and Anthropic coverage.