Claude Duet Review: What Happened When I Set It Up

I expected Claude Duet to know me as soon as it imported my Claude account. It brought over my tools, projects, chats, memory, and skills, then still asked me to describe my work in my own words. That was the first surprise in setting up Anthropic’s personal AI agent, and it set the tone for the rest of the app: one assistant with a shared memory, separate channels for different parts of life, and a dedicated Tasks page for work that keeps moving.

I used Duet to organize a workout-coach channel, approve three research tasks for No Code MBA, and review the finished documents inside the app. My take: the interface makes ongoing work easier to follow than a pile of chat threads, and I like that it runs on Opus 5.5. The trade-off is fixed model choice. You can’t switch to another provider, and I still want to see how proactive Duet stays over time.

What surprised me during Claude Duet setup

Duet’s onboarding starts with continuity. It offers to import the tools, projects, chats, memory, and skills attached to your Claude account. It also gives Claude its own computer in the cloud, with an option to connect your computer so it can use sites where you’re already signed in. I skipped connecting apps at the start and kept the setup limited to what I needed to explore the interface.

Then Duet told me it had too little information to get a fair read on me and asked me to explain my work. I told it I run the No Code MBA YouTube channel and want to grow it. From there, it suggested work around my niche and video ideas. That extra question is small, but it’s worth knowing: importing account context didn’t remove the need to give Duet a plain-language description of what matters to you.

One agent, shared memory, separate channels

Duet gives you a main chat and channels that keep topics apart without creating a separate agent for each one. I made a channel called “Workout coach” so it wouldn’t get mixed into my work chat. Duet suggested tasks from the channel name alone, then I asked for a weekday 8 a.m. check-in. The experience made the structure easy to understand: different workspaces for different topics, but one assistant and one shared memory behind them.

That’s a useful middle ground if you want organization without managing a roster of agents. In my comparison, Grok Bot’s separate bots each have their own context, while Meta Muse is closer to Duet’s single-assistant approach with side chats. I’m describing how the products are structured here, not claiming a head-to-head performance test.

The Tasks page keeps background work visible

Duet suggested three tasks for my channel: find 10 trending no-code and AI video ideas, analyze what my best videos have in common, and size up competing channels. I approved each one from chat. All three showed up as active on the Tasks page, where I could open a task and see its plan and progress instead of guessing what the agent was doing.

The finished work arrived as documents inside Duet. I reviewed the video ideas and the analysis of my existing videos, then checked a task off to move it into Resolved. The useful part wasn’t just that it generated reports. Because the work stayed connected to the same assistant, I could return to the main chat and ask about a report it had made.

I also saw connector options for services including Slack, mail, Google Drive, Stripe, and 1Password. I didn’t connect those accounts in this session, so I’m not treating the video as a test of what Duet can do inside each service. Anthropic’s help page is the place to check current setup and eligibility before connecting your own accounts.

The fixed model choice is the main trade-off

Duet runs on Anthropic’s Opus 5.5, and the app doesn’t offer a model picker. If you want to route work to another provider or choose a different model for a task, Duet won’t do that. I like Opus 5.5, so this is a trade-off I’m comfortable with for now, not a reason to skip the app. If you’re deciding whether your Claude subscription is worth it, read the pricing guide linked below alongside Anthropic’s current plan details.

Is Claude Duet worth trying?

My first impression is that Duet is well executed. I like the combination of one assistant, shared memory, channels, and a separate Tasks page. It made the work I approved easy to follow, and the finished reports were available to review and discuss in the same place. If you’re already on a supported Claude plan and can get an invite, I think it’s worth downloading, especially if you like Opus 5.5.

The open question for me is proactivity: will Duet keep coming up with useful work as it learns my routines, or will most of the value come from tasks I explicitly approve? This first session shows the setup and task-review workflow, not a long-term reliability test. I’d keep using it, but I’d judge it over time by whether its suggestions are genuinely helpful and whether its background work consistently earns a place in my day.

Frequently Asked Questions