An AI assistant that can read across your work tools may catch things that slip through the cracks. In a week of testing, Claude Duet spotted scheduling conflicts and proposed practical follow-ups, but it also treated a flexible deadline as urgent and repeatedly offered to help with writing. The experience shows both the promise and the limits of an assistant that acts on context from across a workplace.
One view across scattered work
Claude Duet is Anthropic’s consumer AI agent aimed at professionals and small business owners. It is a research preview, with access limited to Pro, Max, and Ultra subscribers in the US, and a waitlist to join. The service is designed to decide what may need attention and offer help, rather than simply wait for a specific question.
Users connect the productivity and chat services they want Duet to use. It then presents possible tasks as cards that can be accepted or dismissed. In the test described, those services included work email, a calendar, Google Docs, and Slack.
That combination let Duet connect information that lived in different places. A staff meeting summary arrived by email from Gemini, while a related conversation was in Slack. Duet linked the two. It also inferred who the writer’s boss was from a shared Google Doc used for weekly one-on-one meetings.
This cross-service view is the agent’s clearest advantage: a detail in one tool can matter because of something recorded elsewhere. Instead of relying on a person to remember every message, note, and meeting, Duet can look for relationships among them and turn those into a suggested next step.
Useful reminders can become overbearing
Duet began each morning with a short greeting and highlights for the day. One reminder was to pack an item needed for a video shoot before a flight. It also suggested tasks and offered to draft a message or add something to an agenda.
Those prompts can serve as a practical backstop for someone juggling many responsibilities. A missed item or an overlooked conflict can be surfaced before it causes a problem. The test also found a planning benefit: Duet noticed that upcoming time off left only a couple of working days before another major draft was due.
It proposed a way to manage that squeeze, offered to update a Google Doc with weekly notes for an editor, and drafted a Slack message explaining the update. The draft picked up some workplace habits, including lowercase writing and a familiar abbreviation, but also used “sig” for “signature,” a choice the writer would not have made. The message was labeled “sent with Claude.”
At other times, the assistant’s sense of urgency ran ahead of the situation. After a writer told an interviewee that an article was likely to appear on Tuesday, Duet flagged the story as still awaiting an edit early Tuesday morning and warned that the promised link might be delayed. The draft had already gone to editors late Monday. Duet missed that the timing was not a firm deadline and that it was only 7AM local time for the editor.
Access brings a different kind of unease
To make those connections, Duet needs access to a wide range of personal work information. The test involved sharing email, calendar entries, documents, and Slack messages. Even when a user chooses to connect those sources knowingly, watching an agent draw on them can make the breadth of access feel newly immediate.
That raises a question beyond whether a reminder is correct: how might work habits change if an assistant is expected to keep track of meetings, tasks, and follow-ups? The writer wondered whether relying on Duet could make someone less engaged with the small organizational routines that keep them connected to colleagues and projects.
The value of a personal assistant also depends on tact. People can often tell whether a commitment is tentative, whether a message needs an immediate response, or whether a nudge would be unwelcome. Duet’s example suggests that pulling together more information does not automatically give an AI the judgment to interpret tone or workplace norms.
A backstop, not a stand-in
The test points to a useful role for AI agents: checking across scattered tools for overlooked details and bringing them to a person’s attention. That can support someone’s own planning without taking over the responsibility for it.
Creative work is a different matter for the writer. Duet often offered to help start drafts, but those offers were unwanted. Writing can involve finding an unexpected angle along the way; delegating the early work may remove part of that process. The writer also found it uncomfortable when the agent seemed ready to help review a piece about the agent itself.
Duet may improve its judgment about when to send a message. For now, its strengths lie in noticing connections and proposing practical actions, while its weak spots include urgency, nuance, and tone. Whether that balance feels helpful depends not just on what the assistant can access, but on which parts of work a person wants to keep doing themselves.