Claude Duet review: Anthropic's consumer agent is useful but jumpy

Claude Duet review: Anthropic's consumer agent is useful but jumpy

The Verge's reviewer spent about a week testing Claude Duet, Anthropic's AI agent for consumers, while juggling a busy travel and work schedule. The verdict in the headline: a little neurotic, and surprisingly helpful.

Duet is still a "research preview" according to Anthropic. Access is limited to Pro, Max and Ultra subscribers in the US, with a waitlist to join. It is aimed at "professionals and small business owners" rather than shoppers (the review says it is not there to order burritos), and it is meant to use its own judgment about what needs your attention most. The reviewer places it in a run of recent agent launches: Instinct, a text-based assistant; Meta's Muse, which helps buy things; and OpenAI's work-oriented Dots. Duet is more work-focused than those, and more consumer-friendly than Claude Cowork, which in turn was more consumer-friendly than Claude Code, the coding agent that kicked off a hype cycle earlier this year.

Like Meta's Muse, Duet has its own app, which the reviewer says has been updated to fully support the iPhone Duo. It uses Claude's cheery, minimalist design. Once you connect the productivity and chat services you want, it looks for ways to help and offers them as cards you swipe left or right to hide or accept, which the reviewer calls a to-do list Tinder.

The reviewer gave it everything: work email, calendar, Google Docs and Slack. That paid off in connecting information across sources. The biweekly staff meeting is summarized automatically in an email from Gemini, and because Duet could read the email, it tied something said in the meeting to a direct message in Slack. It also worked out who the reviewer's boss is from a shared Google Doc used for weekly 1:1 meetings. Each morning it greeted the reviewer with a short message and highlights for the day, such as a reminder to pack an item needed for a video shoot before a flight, then suggested to-dos, usually ending with an offer to draft a message or add an item to an agenda.

The helpfulness came with a nagging streak. The reviewer had told an interviewee that a story would likely go up on Tuesday, with a link to follow. The draft went to editors late Monday night. On Tuesday morning, before 9AM, Duet pointed out that the story was still waiting for an edit and that the reviewer needed to get on it, or the promised link could not be sent. The concern was a bit misplaced: Duet missed the nuance in the message implying the deadline was not hard, and that it was only 7AM local time for the editor.

In another case it was strongly useful. While planning the next few weeks, the reviewer had missed that upcoming time off would leave only a couple of working days before another big draft was due. Duet proposed a solution, offered to update the Google Doc with weekly notes for the editor, and drafted a Slack message to him about the update. The message was almost too good: it copied the reviewer's habit of writing in all lowercase, which Duet pointed out in a slightly self-congratulatory aside, and used an abbreviation the team uses for its CMS. It also shortened "signature" to "sig", which the reviewer says they never would, and which made them laugh out loud. The message carried a "sent with Claude" label, so no one was fooled.

Duet often volunteered to help start the reviewer's drafts, which is a hard no. The reviewer writes their own pieces, and sees value in the offer only for people who are not writers. The offer became a Spider-Man-pointing-at-Spider-Man moment when the task Duet wanted to help with was reviewing itself.

The reviewer also found it a little uncanny that Duet can read every email, draft and Slack message sent. They opted in and knew what they were doing, but seeing it in practice is different. They wonder whether knowing Duet would handle calendar entries, to-do lists and meeting attention would change how they work, and whether they would feel disconnected from the workplace. Maybe that is how CEOs live, they write.

The conclusion: no agent is yet ready to make everyone the CEO of their own little world. A personal assistant needs to be personal: a person picks up on tone, communication norms and tact. Anthropic's blog post says the company is continuing to "test its judgment about when to message you", so Duet may calibrate better, but the reviewer argues AI overall is not good at nuance or tone and it is unclear it ever will be. Duet clarified what the reviewer wants from AI: finding overlooked discrepancies across email, DMs, meetings and shared docs and flagging them proactively is really helpful. They like it as a backstop for their own organizational systems, not a replacement, and do not want it in the creative process. Citing Hank Green, they want to find "my own ways into and around a topic", since sending a robot to do the legwork means missing the chance to stumble on something unexpected.

Key facts

  • Claude Duet is Anthropic's consumer AI agent, still a "research preview", limited to Pro, Max and Ultra subscribers in the US with a waitlist.
  • With access to work email, calendar, Google Docs and Slack, it linked a meeting summary from Gemini to a Slack DM and inferred the reviewer's boss from a shared Google Doc.
  • Suggestions arrive as swipeable cards; the reviewer found some helpful, but Duet nagged about an unedited story before 9AM when it was 7AM for the editor.
  • It drafted a Slack message that copied the reviewer's lowercase style and shortened "signature" to "sig", labelled "sent with Claude".
  • The reviewer likes Duet as a backstop for organizational systems, rejects its offers to help draft their own writing, and doubts AI will master nuance and tone.

Why it matters

Duet is Anthropic's latest attempt to show that agents can be useful to people who do not write code. The reviewer places it among recent consumer agents: Instinct, Meta's Muse and OpenAI's Dots, with Duet the most work-focused of the group and a step more consumer-friendly than Claude Cowork. The review shows where this kind of agent already earns its keep (connecting scattered information) and where it still stumbles (judging what is urgent).

Who it affects

Anthropic says Duet is for "professionals and small business owners". For now, access is limited to Pro, Max and Ultra subscribers in the US, with a waitlist. The reviewer suggests people who do not write for a living may get more from its offers to take on small writing chores than a writer would.

How to use it

Connect the productivity and chat services you want, such as email, calendar, Google Docs and Slack. Duet then opens each day with a short greeting and highlights, followed by suggested to-dos as cards you swipe left to hide or right to accept. The reviewer's advice is to use it as a safety net for your own organizational systems: it is good at flagging discrepancies between emails, DMs, meetings and shared docs. Treat its offers to draft messages as optional. No price or release date is given in the review.

How solid is it

This is a first-hand, roughly week-long hands-on from one reviewer at The Verge, based on their own work accounts and schedule. It is an opinionated account, not a benchmark: no error rates or user numbers are given. The product itself is labelled a research preview by Anthropic.

Risks and caveats

Duet can read every email, draft and Slack message the user sends, which the reviewer found uncanny even after knowingly opting in. It misjudged urgency by nagging about an edit before 9AM without seeing the deadline was soft and that it was early for the editor. Anthropic's blog post says it is still testing the agent's judgment about when to message users. The reviewer also wonders whether leaning on it could leave people disconnected from their workplace, and argues AI is not good at nuance or tone, with no certainty that it will become so.

“I like it as a backstop for my own organizational systems, not as a replacement for them.”

— The Verge's reviewer on Claude Duet