Muse, Dots, and what design does when agents reach everyone

Meta released Muse on September 8, and within ten days it was the top free app on the US App Store. Muse gives every user an agent with its own virtual machine, and the agent keeps going after you close the app. You can hand it a task before bed and it chips away at it while your phone sits on the nightstand. Three weeks after Muse came out, on September 29, OpenAI announced Dots at DevDay, an agent that does much the same thing inside ChatGPT.

When the news first broke, I didn't think much of it. I was already doing this. I hand off my notes and scheduled tasks to Claude Code, and anything that needs a logged-in website to an AI browser. Feature for feature, neither Muse nor Dots did anything my setup couldn't.

A few days later it dawned on me that I'd been asking the wrong question. I had only asked whether agents could do something new. The real question was who could use them now. My setup works for people who are at home in a terminal. Muse just asks you to download an app.

There were phones before the iPhone

Phones could already do plenty in 2006. You could check email on a BlackBerry, browse the web on a Nokia, and listen to music on an iPod. But getting work email onto a BlackBerry took some setting up, and browsing on a tiny screen meant a lot of fiddly button presses. So these features mostly belonged to office workers and gadget lovers.

The iPhone didn't invent any of it. Email, the web, and music were all there already. What the iPhone did was boil them down to swiping and tapping. People who would never have gone near a settings menu started reading email and browsing on their phones. That's why we remember the iPhone as the start of the smartphone.

Agents are in roughly the same spot today. The features are there. But using them means opening a terminal, getting an API key, and putting up with things breaking. My own setup only keeps running because I keep tinkering with it. Most people won't go to that trouble, and they shouldn't have to.

Where Muse and Dots put their effort

Using Muse, I never got the sense that the model under the hood was especially smart. What stands out is that the agent comes across as a character. Someone clearly sweated the details of how it moves and what the screen shows while it works. That may sound like a small thing, but it goes a long way for a first-time user. When you can see what the agent is up to, it's a lot less nerve-racking to hand something off and wait. I think that had a lot to do with Muse making it to number one.

What sold me on Dots is that you can call it. I've hooked it up to my LLM wiki, a collection of my notes and the papers I've read, and I use phone calls to practice job interviews in English. In a chat window you can rewrite an answer before you hit send. On a call you have to answer out loud, on the spot, the way you would in a real interview. Since my notes are connected, it can also quiz me on my own research. My research is about people practicing conversations with AI, so every call sends me back to my own work.

Both products put their effort into how you use the agent more than into new features. Muse puts the agent front and center the moment you open the app, and with Dots you can just pick up the phone and talk to it. There is next to nothing for a new user to learn.

When a technology goes mainstream

Once a technology can do a job well enough for ordinary people, what it can do starts to matter less than what it's like to use. That's what happened with phones. For years after the iPhone, phone makers competed on how smoothly the screen scrolled and whether you could pick the thing up and use it without a manual. From then on, designers made as much of a difference as engineers.

Muse and Dots suggest agents have hit that point. Both were built from day one for people who will never read a changelog. For those people, model names and benchmark scores matter less than two simple things: knowing what to do when they first open the app, and being able to see how the task they handed off is coming along.

When the agent becomes someone you deal with

This shift matters to me because of my research. I study AI as the other party in an interaction: someone a person talks to, practices with, and shares judgment with. Once an agent has a character, as Muse's does, people start treating it less like a tool and more like someone. Nobody thanks a calculator, but when something has a name and a voice, a "thanks" slips out on its own.

That changes what designers have to think about. When should this other party speak up, and when should it keep its head down and work? A notification in the middle of a meeting is an interruption, and making a big call on its own is a much bigger problem. How much of what it did on my behalf should it show me? Show everything and I'm buried in detail. Show nothing and it's hard to trust. When should it check with me before acting? Drafting an email on its own is probably fine, but it should ask before it hits send.

A better model won't answer these questions. You can only answer them if you know what the person on the other end wants. The same feature has to work differently for a busy office worker and for a parent trying an agent for the first time. I expect agent products to win or lose on questions like these over the next few years.

Where the phone comparison falls short

The phone comparison only goes so far. A phone did what you pressed and nothing more. An agent can send email and make payments on your behalf while you sleep. However friendly the interface, if you can't trust what it did while your back was turned, you won't hand it anything that matters. So putting agents in everyone's hands will be harder than it was with the iPhone.

That makes the problem design has to solve less about good-looking screens and more about trust. You should be able to go back and check what the agent did, step by step, and it should stop and ask before anything that can't be undone. Getting those pieces right could take a while. For all their popularity, Muse and Dots may end up remembered less like the iPhone and more like the touchscreen phones that came before it.

These thoughts come from using both products myself. The more I use them, the clearer it seems that agents now compete on what they're like to use, not just on what they can do.

Sources: MacRumors on the Dots launch · Tom's Guide comparing Muse and Dots