Claire Vo: Claude Duet Is Beautiful, but It Is Not My Number One Agent

I have good news and bad news. Anthropic just released a new personal agent, and it is really well designed. The bad news is I have personal agent fatigue. I am so tired of new personal agents, and while this one is super well designed—and I will go through what I love about it—I just do not need another one. To compete for my attention, to compete for being my number one personal agent, you have to be not good, not great. You have to be exceptional. We are going to see in this episode whether Claude Duet, their new personal agent, makes the cut.

Before I get into my experience, a very quick walk through of what it is. Duet is an iPhone and Android personal agent, synced up with Claude, that does the things personal agents do. It is proactive—although I will go through how proactive it really is. It can use all the tools and plugins your Claude already has access to. Anthropic really positions this as a personal agent for small business owners or busy professionals.

This is not “your mom is going to use it with Muse,” and it is not “you have to be a Lauren at Cursor running 50,000 agents and shipping a gazillion PRs.” It is a solid middle-of-the-road business user: SMB, entrepreneur, busy PM, busy CEO. The things you would reach for in Cowork on your desktop, maybe you will start asking Claude Duet for.

You all know I really love OpenAI products. I use Codex all day, every day. One thing Anthropic did so well that OpenAI’s Dots did so poorly is the polish of the agent design. This app is beautiful. It is well designed. There are no sharp edges. It is clearly incredibly well thought through from a design perspective. Everything from the early TestFlight versions I got access to has shown me this is an exceptionally well-polished product.

From the second you onboard, they are thoughtful about walking you through what Duet does, what it has access to, and how to get started. It looks really nice. It feels trusted. I had zero problems getting up and running out of the gate. It was a really nice onboarding, a little similar to my experience with Muse: very clear up front what it could do, very clear what it had access to, and then you got right into the product.

The product itself has some primitives, again like Muse, that I think you are going to like. Of course it has plugins and connectors, so it inherits all your plugins from Claude. If you have Gmail set up, Google set up, Figma set up, whatever, it is going to have access to those.

It also has task lists, which is something I really liked from Muse’s agent design. I like keeping track of what goals and tasks I am doing and what goals and tasks the agent is doing. The task list view is something that is really nice here, and something we are going to come back to a lot.

Then it has artifacts, which Muse also has, in a little more polished version. If you are used to artifacts in Claude Code or in Cowork—generating HTML, designs, slides—Anthropic has made those first-class citizens in Duet. Whenever you are generating stuff in your little chat with your little Duet, it is going to create those artifacts. It can already read the artifacts existing on your Claude account. When I showed up and looked at my Claude account, it already had all those artifacts it had generated over time, and I had a really good sense of what I could do next.

Finally, the last thing I really love about Duet is channels. We are all trying to decide what the primitives of agents are. In Codex you have threads. In chat you have chats. In Claude you have threads and chats and projects. In Muse you have a main chat and side chats. Everybody is trying to figure this out. Anthropic borrowed a little bit from Slack, which I feel like is everybody’s favorite collaboration service, and positioned these sub-chats, these sub-threads, these sub-places you are going to work in, as channels. You can think of your Duet as a main thread, then break off other chats into channels. I created a sales channel for working on sales projects.

I will say my last little thing that I really like, and I did give the Anthropic team props on, is the icon. It is abstract. It is definitely Anthropic Claude-y. It is cute. I have no problem with this little, dare I say it, dot. It is just adorable, not offensive, not weird. I like it. Of course I love my personalized Muse consumer agent, and that engenders a lot of joy. Of course I love my little Grok Bots and their little icons. This is inoffensive, and I think it is just fine. As somebody who cares about design, I thought it was pretty good.

Now, what are some things I did not like about Duet? For a proactive agent, it just was not that proactive. I do not know if I forgot to turn on notifications or it is just not that annoying, but once I set up Duet a couple of days ago and did a first couple of tasks, it never pushed or bothered me to do anything ever again. It also does not show up in the desktop app, so it is a really disconnected experience. The first session was pretty good, then I did not go back: I already have a million personal agents, and it did not pull me in with useful work.

One of the things these agent designers are really going to have to grapple with is that there are only so many three-point morning briefings that can pull me back into a product. I get them from Dots all the time; they are not that exciting. I get them from Duet now; they are not that exciting. There has got to be something better than the morning briefing to engage agent users on why they would come back over time. Right now I do not think any of these agents, except maybe Grok Bot and Muse, have really figured this out.

So it is not as proactive as I want. The notifications are not noisy enough for me to remember it. I have said it is elegant and well designed and subtle and not annoying, and those are all positives in some ways. From a growth-hacking perspective, and the ability to get me to re-engage with your product, that does not work. I need you to be a little bit more annoying. I need you to be a little bit more proactive. I need you to be a little bit more noisy, or I am truly going to forget you in the ecosystem of all my agents.

So what were some things I asked it to do, and how well did it do them? The first thing I asked was the basic pass: go through my email, go through my Slack, figure out what is going on. I would say it did not do an exceptional job at that. It did okay. It came up with a couple of things, but the priority did not seem right, the context did not seem right. There is a little work on the quality of the interaction.

I asked it to find me good Warriors tickets—go Dubs. I want to see the Warriors when they play the Pacers. It is a noon game. I can actually take my kids. I have had all my agents try to shop for tickets for me, shop for shoes for me. I thought this was a good eval. I had it go shop for me, and this is the second thing I think is bad about it.

If the first thing I did not love about Duet is that it is not that proactive, the second—and this has been my experience with Claude generally—is it is slow. I get so disengaged waiting for the models to finish. Meta Muse’s killer feature is that it is fast. Codex with GPT’s ultra-fast models keeps me moving to the next step. Even though I love Opus 5.5 and the work is good, the performance is so slow that you feel it in a personal agent, and it degrades my desire to work synchronously.

One benefit of an AI agent is that it is like a colleague who is up 24 hours a day and responds immediately. If that colleague is slow, if they go to a meeting to get work done, you forget them and pick the more responsive ones, even if those do a 10% worse job. Duet was shopping for Warriors tickets while I was passenger princess on the way to dinner. It took so long, it did not close the loop, and by the time I came back I was on to something else.

Even though it tracked the work in a checklist, the wait was really frustrating. That adds up when you fire off a bunch of tasks in a single agent: check my email, make a deck, edit this, figure out social. In Claude Duet those kicked off individual threads, then took so long—ten-plus minutes to finish, come back, or ask for permissions. Because it was not noisy, because it was not in my face, I forgot about all those threads.

Foundationally it is really well designed, the models are very good, the intelligence is there, and I cannot knock anything specific about the user experience. The reality is the latency and the inability to close loops meant that this particular agent product just really fell to the bottom of my use list.

The other thing they are going to have to solve ASAP—and this is a research preview, early access, beta, I am sure they will get there—is a desktop app. I do not want to be on my phone during the day when I am doing work and sitting in meetings. I need a desktop app to engage with my agents. That is why Grok Bot does so well: a desktop agent, a cute little icon, sitting there staring me in the face. I am always chatting with it like it is Slack.

Same with Muse. I started using it as a mobile app, but very quickly downloaded the desktop app, and that lets me finish tasks I kick off asynchronously on my phone when I am at my computer. Same with Dots from Codex, which I do not even use, but at least it is there. At least my Dot is staring me in the face in Codex with a little notification saying do something here, get something done.

What I would say about Duet is it has got all the makings of a great agent except the ability to drive urgency for me to engage with it. Maybe that is intentional. Maybe the design is to be calm and soothing and helpful, but in the background. To compete for attention right now, beautiful design is not enough. Exceptional models are not enough. Even the artifacts—which are actually awesome, and on the phone quite zippy—are not enough. The agent has to be super proactive, super fast, and close loops quickly enough that it feels like a reliable coworker.

Okay, that was a very long aside from buying Warriors tickets. A couple of other things I had it do—good, bad, left, right, standard. I had it look at my sales pipeline. Nice things there. It was already connected to Slack. It was already connected to my email. It was already connected to our CRM. So it was able to pull in that data pretty quickly.

I had it create a deck in Google Docs. I have a really high bar for design, and so I was disappointed that it did not use Figma and really match our design in terms of fonts and all that kind of stuff. But the deck was not bad. It was created in Google Docs and it was kicked off from my phone. For the kind of everyday business person who maybe is not as particular as me with design, they would be really happy to create little sales decks off a template, little documents, on your phone on the go with Duet. That was good, not great, but I have a really high bar for that sort of thing.

One other point of friction—and this is a me problem as much as a Duet problem—is I wanted to see if it could make a launch video. I have been making ChatPRD launch videos in Claude that look like this, and they are super cool. I wanted to create those in Duet on my phone. Unfortunately Duet does not ingest all your local skills, so I probably need to push those skills as plugins to Claude first.

With all these product surface areas—and this is not just an Anthropic problem, this is an OpenAI problem—shared context, the shared experience, what ports over elegantly and what does not, is going to be an open question for folks who want to really migrate to Claude Duet.

The final thing that is really missing is: look, we do not work alone. Duet right now is clearly an agent for single-player use. In fact, most of the Claude products feel like very good products for single-player use. I just do not think, even with Claude in Slack and related team surfaces, they have quite cracked multiplayer. I have been moving more of my Grok Bot usage to team agents and deploying those in Slack. So while Claude Duet with these channels might feel like Slack, it does not have that multiplayer sharing aspect of Slack that has made me really effective, even as a small business owner.

I am not running a team of 500 anymore. I do not need the full enterprise deployment. But I am a small business owner, so I should be the target for this product. The problem is I have a small team. I have like five humans that I work with on a regular basis. If I do not have a way to share my agents with them, I am not sure it is going to worm its way into my working stack.

That is my early review of Claude Duet, Anthropic’s new personal agent. I would definitely give it a try. I maybe have different use cases and preferences than most people. It is approximately 100 times more polished than Dots was when it came out, and it is quite elegant. It does the job well. There are fit-and-finish things they really need to work on if they are going to compete for my agent business. Sign up for early access if you can. Let me know if Claude Duet becomes part of your stack, because I am just not sure it is going to become part of mine.