Claude Duet review: Anthropic's AI agent tested in daily work

Anthropic launches a consumer agent

Anthropic has released Claude Duet, an AI agent aimed at ordinary consumers rather than programmers. The Verge reports that a writer tested it for a week alongside a busy travel and work schedule. The publication describes the experience as sometimes genuinely useful, occasionally irritating, and marked by an unusual eagerness to help write about itself.

Duet is Anthropic's latest step in a broader industry push to make AI agents useful to people who do not write code. According to The Verge, it follows other recent consumer agents, including Meta's Muse and OpenAI's Dots, and sits between those products and more work-focused tools in complexity.

How Duet works day to day

Duet has its own app, which The Verge notes has been updated to fully support the iPhone Duo. The design follows Claude's usual minimal style. After connecting services such as email, calendar, Google Docs and Slack, Duet suggests ways to help through cards that can be accepted or swiped away, an approach The Verge compares to a to-do list version of a dating app.

In testing, Duet proved most useful when it could draw on lots of connected data. Because a staff meeting was automatically summarised by email, the agent could link a comment from that meeting to a Slack message. It also worked out who the reviewer's boss was from a shared weekly meeting document.

Each morning, Duet greeted the user 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. It then suggested tasks, often ending with an offer to draft a message or add an agenda item.

Eager but tone-deaf

The Verge's writer found Duet could behave like an overanxious assistant. In one case, the agent complained before 9AM that a story was still awaiting an edit, worried that a promised link could not be sent. The concern was misplaced: the deadline was soft, and it was only 7AM in the editor's local time zone. Duet missed both cues.

There were brighter moments. When the reviewer overlooked that upcoming leave left only a couple of working days before a draft was due, Duet proposed a solution, offered to update a shared Google Doc and drafted a Slack message. That message copied the writer's lowercase style and used an in-house abbreviation, but it also shortened a word in a way the writer never would. Anthropic's blog says the company is continuing to test Duet's judgement about when to message users.

Privacy and the nagging question of reliance

The reviewer granted Duet access to work email, drafts and Slack, and describes the experience of an assistant that can read everything as uncanny in practice, even when the data sharing was deliberate. The Verge also raises a longer-term question: whether people would start relying on an agent for calendars, to-do lists and meeting notes, and whether that would leave them feeling detached from their workplace.

These observations come from one reviewer's experience rather than a formal assessment, and Anthropic has not commented on the tone issues through The Verge's piece. The article frames them as open questions for AI agents generally, not confirmed defects.

What it means for users

The Verge's conclusion is that Duet clarifies where AI agents help and where they do not. Spotting overlooked discrepancies across email, messages, meetings and shared documents is described as genuinely valuable, positioning Duet as a backstop for a person's own organisational systems rather than a replacement for them.

Creative work is a different matter. The reviewer declined Duet's repeated offers to start drafts, arguing that handing over the legwork removes the chance of stumbling across unexpected ideas. The wider point, according to The Verge, is that AI still struggles with tone and nuance, and it is unclear whether that will change.