Claude Duet Explained: How the AI Agent Works and Who Can Join

I didn't expect Anthropic to launch a personal AI agent. While Meta, xAI, and OpenAI were releasing consumer agents like Muse, Grok Bot, and ChatGPT's agent tools, Anthropic stayed focused on coding and developer workflows. That just changed with Claude Duet.

Duet is a standalone app built on Claude Opus 5.5 that automatically pulls in your existing Claude memory, skills, and connectors. Instead of burying ongoing work inside a chat thread, it puts a dedicated Tasks tab right in the center of the navigation bar so you can see active jobs, upcoming runs, and tasks it suggests from your history.

What is Claude Duet?

Claude Duet is Anthropic's proactive personal agent, currently rolling out in a research preview for US Pro and Max subscribers. The biggest practical advantage if you already use Claude is that Duet connects directly to your existing account. Your Claude memory, skills, and app connectors carry over automatically, so you don't have to spend an hour configuring a new tool before it becomes useful.

According to Anthropic, Duet can surface tasks on its own, work across your connected apps, and keep running in the cloud after you close the app. It also pauses to ask for your approval before taking high-stakes actions like sending a message, making a payment, signing a document, or deleting files.

Inside the app: Chat, Tasks, and Make

Small layout choices change how you actually use an AI agent. In Anthropic's preview screens, Duet splits the bottom navigation bar into three tabs: Chat, Tasks, and Make.

Giving Tasks the center button on the navigation bar is the smartest design choice in the app. When you open Tasks, you see what Duet is actively working on, what is lined up next, tasks saved for later, and suggested tasks based on your history that you can kick off with one click. Make sits on the right side of the bar and collects everything Duet builds for you, along with starter templates.

How to join the Claude Duet waitlist

Access is opening in waves right now, so it is worth getting on the waitlist early if you want to try it:

1. Go to claude.com, open the Product menu at the top, and select Claude Duet.
2. Click Join the waitlist and enter your name and the email tied to your Claude account (Anthropic notes this should match the Google or Apple email on your phone).
3. Select your Claude plan and phone type, then describe your work and the first task you want to hand off.
4. Submit the form and watch for an email invite as Anthropic opens more spots.

Right now, Anthropic limits the preview to invited Pro and Max subscribers in the US, and Team or Enterprise accounts aren't eligible yet. In my experience with waitlists like this, writing a specific, concrete use case in the form gives you a better shot at getting in early than leaving a generic one-liner.

Claude Duet vs. Meta Muse, Grok Bot, and ChatGPT

Looking at the app structure, Duet sits right between Meta Muse and Grok Bot. Muse feels built for everyday consumer tasks, while Grok Bot is geared toward setting up multiple specialized bots for business workflows. Duet gives you a clean, proactive task manager in one place without making you manage a fleet of separate bots. I also prefer that Anthropic made Duet a separate app rather than baking it into the main chat interface the way ChatGPT did, which makes ChatGPT's agent harder to navigate.

The biggest trade-off is model lock-in. Because Duet comes from Anthropic, the preview runs strictly on Claude Opus 5.5 with no option to switch models. Meta Muse similarly sticks to Meta's models, while Grok Bot claims it routes tasks to the best model available (even if you still can't pick the model yourself). I really like Opus 5.5, so running on it by default is a strong starting point, even without multi-model flexibility.

Is Claude Duet worth trying?

Since Duet is still rolling out by waitlist, I'm judging the interface walkthrough and setup rather than live task reliability. Even from the preview screens, though, two things make it stand out: the center Tasks tab makes proactive agent work easy to see and start, and pulling in your existing Claude memory, skills, and connectors removes the cold-start problem. Once my invite lands, the real test will be how reliably it finishes background tasks across connected apps.