A gym in January is the busiest place on earth. Doors swinging, treadmills full, everyone with a plan. Come back in March and you can have the whole row of bikes to yourself. The sign-up sheet was never the hard part.
I keep thinking about that while reading the numbers on Muse, Meta’s new personal AI agent. By the download counts, it has had one of the loudest openings an AI app has had. App-intelligence firm Appfigures estimates Muse pulled 2.8 million global installs in its first 12 days, with 359,000 daily active users on iOS. A separate report put an early download figure at 730,000. And Apptopia, running a matched U.S.-and-Canada iOS comparison, found Muse outpacing ChatGPT’s early mobile download pace at the same point in its life.
That is a genuinely unusual result. ChatGPT’s mobile launch is the benchmark everyone measures against, the thing other companies quietly hope to get within shouting distance of. Muse cleared it.
What the numbers actually say
Before anyone gets carried away, two bits of translation for folks who do not spend their days reading app charts.
First, these are estimates. Appfigures and Apptopia do not get Meta’s internal dashboards. They model app performance from outside signals, which is useful and directionally reliable, and also why you will see different numbers in different write-ups. The 730,000 figure and the 2.8 million figure are not contradicting each other so much as measuring different windows and different scopes. Treat all of them as well-informed approximations, not receipts.
Second, downloads and daily active users are two very different things, and the gap between them is where the real story lives. An install means someone tapped a button once. A daily active user means someone came back. Muse’s 359,000 daily actives on iOS sit against millions of installs, which is normal for a brand-new app but also the number I would watch obsessively if I worked at Meta.
Why a fast start is easier for Meta than for anyone else
Here is the part that makes me cautious about reading Muse’s opening as a verdict on quality. Meta is not a startup asking strangers to discover a new app. It owns some of the most-used surfaces on the internet, and it has spent years getting very good at pointing attention from one product toward another. A company with that kind of built-in audience should beat a cold-start launch. The surprise would be if it did not.
ChatGPT, by contrast, had to be explained to people before they downloaded it. It grew mostly because friends told friends, journalists could not stop writing about it, and a text box that answered questions felt like a small magic trick. Distribution came after the curiosity, not before it.
So comparing the two launches is a little like comparing a first-time novelist to a sequel. Both can sell a lot of copies. Only one of them started from zero.
What an AI agent has to earn
If you are new to the term, an AI agent is a step past a chatbot. A chatbot answers. An agent is meant to act on your behalf, taking a request and carrying it through a few steps without you supervising each one. That framing is why Muse’s early numbers matter more than a typical app launch.
Agents live or die on trust, and trust is built through repetition. You do not hand over your calendar, your inbox, or your shopping to a tool you tried once on a Tuesday. You hand it over after the tool has quietly gotten small things right for a few weeks and never embarrassed you.
That is a slower kind of win than a download chart can capture. The metrics that will tell us whether Muse is a hit are the unglamorous ones:
- Retention past the first month. Who is still opening it when the novelty wears off?
- Session depth. Are people asking it to do real work, or poking at it for fun?
- The daily-active-to-install ratio over time. Does it climb, hold, or quietly sink?
Why I think this is still a big deal
None of that skepticism erases what happened. For a couple of years, the assumption in this space was that OpenAI had a lead in consumer AI that would be difficult to close, partly because ChatGPT became a verb before competitors had shipped anything comparable. Muse’s opening suggests that lead is contestable, at least at the top of the funnel.
For the rest of us, competition is the good outcome. When two well-funded companies are fighting over who gets to be the assistant on your phone, the pressure lands on being useful, cheap, and clear about what it does with your data. That is a better arrangement than one winner setting the terms.
For now, Muse has had a genuinely strong first 12 days, measured by outside estimates, with Meta’s distribution muscle behind it. Whether it becomes something people rely on is a question the January gym crowd answers every March.
🕒 Published: