Claude Duet Tested: Anthropic's Consumer AI Agent Is Helpful, If a Little Jumpy

Anthropic's Claude Duet is the company's bid to make AI agents useful to people who don't write code. According to a hands-on review by The Verge, the product lands somewhere between a genuinely helpful backstop and a slightly anxious assistant.

What it is, per Anthropic and The Verge

Duet is still labeled a research preview, limited to Pro, Max, and Ultra subscribers in the US, with a waitlist. Anthropic describes the target user as "professionals and small business owners." The agent lives in its own app, shares Claude's minimalist look, and connects to productivity and chat services. It then proposes actions as swipeable cards—The Verge likens the flow to a to-do list with dating-app mechanics.

The Verge gave it work email, calendar, Google Docs, and Slack. Because meeting summaries arrived by email, Duet linked topics across a staff meeting and a Slack direct message, and inferred who the reviewer's boss was from a shared one-on-one document.

Where it helped

Mornings started with a short briefing and reminders, such as packing gear before a flight. The standout moment: Duet noticed that upcoming time off left only a couple of working days before a major draft was due, proposed a fix, offered to update a Google Doc with weekly notes, and drafted a Slack message. The draft mimicked the reviewer's lowercase style and a common abbreviation—though it also introduced a shortened word the reviewer would never use.

Where it got in the way

Duet misread a soft deadline, flagging an unfinished edit before 9AM local time—ignoring that the reviewer's editor was in an earlier time zone and that the original message carried no hard promise. The Verge describes the tone as a slightly neurotic personal assistant. Duet also repeatedly offered to draft stories, which the reviewer declined.

The privacy and dependency questions

The reviewer opted in to sharing email, drafts, and Slack messages and notes an uncanny feeling at seeing the access in practice. A larger worry: if the agent handles calendar entries, to-dos, and meeting attention, does the user become less connected to their workplace? The Verge frames this as an open question rather than a verdict. Anthropic says it continues to test when Duet should message users, so calibration may improve.

The reviewer's own line

The Verge concludes that Duet clarified what they want from AI: a backstop for organizational systems and a catch for inconsistencies across email, DMs, meetings, and shared docs—not a creative collaborator. The reviewer prefers finding their own way into a topic, citing Hank Green on the value of stumbling across the unexpected. And no, they won't let an agent review itself.

What remains uncertain

This is one reviewer's week with a limited preview. There is no evidence yet on reliability at scale, pricing beyond existing subscriptions, availability outside the US, how data permissions work for typical users, or whether judgment about tone and nuance improves. Readers should treat the experience as indicative, not representative.