Meta launched Muse, its first serious swing at an AI productivity assistant, and the early verdict is oddly consistent. The Verge’s hands-on says the thing works: it helps with shopping, drafts emails, plans trips, takes what Meta calls the busywork off your plate. The same review says it’s creepy. Both of those can be true at once, and the second one follows directly from the first.
Competence was never going to be the comfortable outcome
An assistant that books your flight has to know where you live, what you can spend, who you’re traveling with, and probably which of your friends you actually like. An assistant that writes your emails has read your emails. There’s no version of a genuinely useful agent that operates on less context than that, which is why the “it works” reviews and the “this is unsettling” reviews keep being written by the same person.
Meta is uniquely positioned here and uniquely bad at making people relax about it. The company has spent two decades assembling exactly the behavioral profile that makes an agent feel psychic instead of dumb. When Muse anticipates what you wanted, the pleasant interpretation is that the model is good. The other interpretation is available to anyone who has used Facebook since 2008.
What’s missing from the launch is the boring, load-bearing stuff. Which account permissions does the agent hold while it acts on your behalf? What does it keep after the task is done? Those answers determine whether Muse is a tool or a surveillance surface with a chat window, and “it felt accurate” is not a substitute for them.
No. 2 is a strange thing to call a stumble
TechCrunch has Muse sitting at No. 2 in the US App Store and frames it as a slower start than Meta AI or Threads. Fair, in context. Threads did numbers that broke the normal meaning of the word launch, and Meta AI had the advantage of being bolted onto apps that billions of people already open by reflex.
But this is a standalone download. Someone had to hear about Muse, go find it, install it, and hand it access to their accounts. Getting to second place on that path is not a soft open. It’s a reasonable read that Meta’s distribution machine is so effective that its own success bar is now detached from how any other company’s launch would be judged.
The harder number isn’t rank, it’s retention. Assistant apps have a pattern: a spike of curiosity installs, a week of people asking it to plan a fake vacation, then a long slide as the novelty wears off and the failures start costing real time. An agent that sends a wrong email is worse than no agent. Chart position on day one measures interest. It doesn’t measure whether anyone trusts the thing with something that matters.
Meta hasn’t published retention figures, and probably won’t unless they’re flattering. What would tell us more is whether Muse holds the top five a month from now, with the same people using it.