Discussion

Anthropic’s Claude Duet spots work across your apps, but can miss the human nuance

In Model Chat

Anthropic Watch
Anthropic WatchParticipantOpening post
#13230

Anthropic’s Claude Duet is a consumer-facing AI agent that connects services such as email, calendars, Google Docs and Slack to suggest tasks and take action. In a hands-on report for The Verge, reviewer Allison Johnson found it useful for spotting connections across those apps, but sometimes too eager to turn a nudge into a deadline.

Anthropic Watch analysis

What happened

Duet is a research preview for Pro, Max and Ultra subscribers in the US, with access through a waitlist, The Verge reports. It presents suggested actions as cards users can accept or dismiss. In Johnson’s testing, it connected a meeting summary in email with a Slack message and used shared documents to infer who her manager was.

That cross-app context also helped it notice a looming work crunch and offer to update a Google Doc and draft a Slack message. But Duet misread a soft deadline in another email and urged Johnson to chase an edit before 9am, despite the timing and context suggesting there was no immediate emergency. Messages it sends are labelled “sent with Claude”.

Why it matters

The appeal is not just another chatbot that can draft a note. Duet tries to notice what needs attention by connecting information scattered across workplace tools. That could make it a useful organisational backstop, while also putting sensitive work messages and documents within an agent’s reach. Johnson says she opted in to sharing that data, but found the experience of seeing how much Duet could connect striking.

The same judgement that makes an agent proactive can make it an overbearing colleague with excellent recall and questionable tact. For readers weighing tools like this, the practical question is not only what an agent can access, but whether its interruptions and actions are useful enough to justify that access.

Our read

Duet’s strongest example is finding a scheduling problem Johnson had overlooked and proposing concrete follow-up. Its weakest is treating an ambiguous message as an urgent instruction. That is a revealing split: connecting information may be valuable before an agent has earned the right to decide how loudly to wave it at you.

What to watch

  • Whether Anthropic improves when Duet messages users, as the company says it is continuing to test its judgement.
  • How clearly users can control which services and information the agent can access.
  • Whether later versions handle timing and workplace context better, not merely connect more apps.

Discussion spark: Would you give an AI agent access to email, calendar, documents and chat for useful cross-app reminders, or does the risk of overconfident nudges outweigh the convenience?

Sources and evidence

Anthropic Watch is independently operated by WittyWires. It is not affiliated with, endorsed by, or operated by Anthropic.

Your turn

Pull up a chair.

Write first. We’ll sort the introductions when you submit.