Meet Duet, Claude’s Slightly Anxious New Assistant

Imagine an assistant who reads every email you send, knows what came up in every meeting, and taps you on the shoulder at 7AM about a deadline that isn’t really a deadline. Anthropic has built roughly that with Claude Duet, its new AI agent for consumers. The Verge AI tested it for a week during a busy stretch of travel and work, and its verdict is that Duet is “a little neurotic and surprisingly helpful.”

What Anthropic launched

Duet is Anthropic’s latest attempt to show that AI agents are useful for people who don’t write code. It’s a step closer to everyday users than Claude Cowork, which was already friendlier than Claude Code, the coding agent that kicked off this year’s agent hype.

It won’t order you burritos, though. Anthropic built Duet for “professionals and small business owners,” and it’s meant to use its own judgment about what needs your attention most.

How it works

  • A standalone app: Duet has its own app with Claude’s cheery, minimalist design, and it now fully supports the iPhone Duo.
  • Connected services: You hook it up to your email, calendar, Google Docs, and Slack.
  • Swipeable cards: Suggestions show up as cards. Swipe one way to dismiss a card and the other to put Duet to work on it. The Verge calls it “kind of a to-do list Tinder.”
  • Morning briefings: Every day starts with a short greeting and a few highlights, like a reminder to pack gear for a video shoot before a flight.
  • Drafting help: It offers to write messages, add agenda items, and update shared docs.

Where it shines

Duet is best at connecting the dots. The Verge’s reviewer gave it full access, and it soon linked a Gemini-generated meeting summary in her inbox to a related Slack DM. It also worked out who her boss was from a shared Google Doc they use for weekly 1:1s.

The standout moment came during planning. Duet noticed something she’d missed: upcoming time off would leave her only a couple of working days before a big draft was due. It suggested a fix, offered to update the weekly notes doc, and drafted a Slack message to her editor.

That message was almost too good. It copied her habit of writing in all lowercase (and pointed this out, a bit pleased with itself) and used the team’s shorthand for their CMS. But it also shortened “signature” to “sig,” which she’d never do. And nobody gets fooled anyway, since Duet’s messages carry a “sent with Claude” label.

Where it stumbles

The flip side is a nervous streak. The reviewer had told an interview subject that a story would “likely” go up on Tuesday. By Tuesday morning, Duet was alarmed that the draft was still waiting for an edit. It missed that the promise wasn’t a hard deadline, and that it was only 7AM where her editor lives.

Other caveats from the review:

  • It kept offering to write her drafts, which she firmly turned down every time.
  • At one point it offered to help review itself.
  • Watching an AI read every email, draft, and Slack message feels uncanny, even when you knowingly opted in.

How it compares

Agents for regular people have had a busy couple of months. Instinct released a strong text-based assistant, Meta launched Muse to help people shop, and OpenAI went after work tasks with Dots. Duet leans further into work than any of them.

Availability

Duet is still a “research preview.” It’s open only to Pro, Max, and Ultra subscribers in the US, and there’s a waitlist. So for now, it comes with a paid Claude plan rather than as a standalone product.

Why it matters

What stands out to me is that Duet’s biggest problem isn’t capability. It’s judgment. Reading your inbox and drafting a Slack reply is the easy part. Knowing when something actually matters, and when to stay quiet, is what makes a personal assistant worth having.

As The Verge points out, a human assistant reads tone, learns how people talk to each other, and knows how to be tactful. AI still struggles with all three. Anthropic says it’s still testing Duet’s “judgment about when to message you,” so it may calm down over time.

The bigger question is what happens to us if it does. If an agent tracks your calendar, your to-dos, and your meetings for you, you might lose touch with your own work. That’s the next problem these agents have to solve, and The Verge AI’s full review has more on what living with one is like.

Scroll to Top