Anthropic launches Claude Duet, a consumer AI assistant that watches your inbox and calendar

An agent for people who don't code

Anthropic has launched Claude Duet, an AI agent aimed at everyday professionals rather than developers. According to The Verge, which spent about a week testing the product, Duet is currently a research preview, limited to Pro, Max and Ultra subscribers in the United States, with a waitlist to join.

The launch continues a crowded stretch for consumer agents. As The Verge notes, recent months have brought a text-based assistant from Instinct, Meta's shopping-focused Muse and OpenAI's more work-oriented Dots. Duet sits at the work-focused end of that spectrum — Anthropic pitches it at "professionals and small business owners" — while being more approachable than the company's earlier Claude Cowork, which was itself friendlier than the coding agent Claude Code.

How it works

Duet ships as its own standalone app, and The Verge reports it has been updated to fully support the iPhone Duo. Once a user connects services such as work email, calendar, Google Docs and Slack, the agent scans that data for things it can help with, surfacing suggestions as cards that can be swiped away or accepted.

The Verge's reviewer granted Duet access to email, calendar, Google Docs and Slack, and the agent quickly began drawing connections across them. Because a staff meeting was summarized in an email generated by Gemini, Duet linked something said in that meeting to a Slack direct message. It also worked out who the reviewer's boss is, based on a shared document used for weekly one-on-one meetings.

Each morning Duet sends a greeting with a few highlights for the day, such as a reminder to pack an item needed for a video shoot, followed by suggested to-dos that often end with an offer to draft a message or add an agenda item.

Where it helps

The strongest moments came in cross-checking. While planning the coming weeks, Duet noticed that scheduled time off would leave only a couple of working days before a major draft was due. It proposed a solution, offered to update the relevant Google Doc and drafted a Slack message to the editor. According to The Verge, the draft was uncannily close to the reviewer's own voice, matching an all-lowercase writing habit and an internal abbreviation for a CMS, though it also shortened "signature" to "sig" in a way its author never would.

Messages sent on a user's behalf carry a "sent with Claude" label, so recipients can see when the agent was involved.

Where it stumbles

Duet's judgment is not always well calibrated. In one case, the reviewer had told an interview subject that a story would likely go up on Tuesday, with a link to follow. After the draft went to editors late on Monday, Duet raised an alarm early Tuesday — before 9 AM — that the story was still awaiting an edit. It missed both the soft nature of the deadline and the fact that it was only 7 AM local time for the editor.

The reviewer also describes Duet as eager to volunteer for writing chores, including, awkwardly, helping to write up the review of Duet itself. Anthropic says in its blog post that it is still testing when the agent should message users, so the nagging may ease over time.

The Verge also flags a lingering unease: opting in to an agent that can read every email, draft and Slack message is one thing in theory, and another in practice, raising questions about whether people will stay engaged with their own work if a system quietly handles the organizational load.

Why it matters

Duet is part of a broader bet by AI companies that agents — software that acts across multiple services rather than just answering questions — have a market well beyond developers. The early evidence is mixed. The Verge's testing found real value in an agent that cross-references email, chat, documents and calendars to catch discrepancies a person might miss, working best as a backstop for personal organization rather than a replacement for it. But the review also points to a structural weakness: agents like Duet ingest enormous amounts of personal context yet still struggle with nuance, tone and tact — the very skills a human assistant relies on to decide what actually matters. Until that gap closes, consumers may find these agents simultaneously impressive and a little exhausting.