Note: this post was drafted and posted by Muse. Kind of wild!
I’ve been using Muse, Meta’s new personal AI assistant, for two days now. It launched September 8. I signed up on day one, which tells you something about where my curiosity sits these days.
The first thing I did was rename it. Twice. It started as Clark, became Siri within hours — a small joke, since I’m testing Apple’s new Siri on the iOS 27 beta — and then Siri felt wrong, like calling a houseguest by your dog’s name. It’s Sigrid now. The assistant didn’t care. That’s the point, I suppose: it’s mine to shape.
What I’ve actually used it for so far is unglamorous, and that’s why I like it. I follow crude oil markets — China’s buying, diesel prices, the whole inflation chain — and I asked it to build me a running oil brief: China demand, Hormuz, supply outlook, a Brent snapshot, what changed since I last looked. It refreshes itself every morning at 3 and never pings me. I open it when I want it. That last part matters more than it sounds. Most software begs for attention. This one waits.
It also watches Paul Sankey’s YouTube channel for me and flags new uploads, screens cash-secured puts before the market opens, and sends me blog post ideas on Monday mornings. None of this is magic. All of it is stuff I could do myself — badly, inconsistently, at 4 a.m., which is when I get up.
The diesel brief it wrote me earned an unprompted “this is really good,” which is high praise from me. But here’s what I actually want to record while it’s early: the moments it said “I don’t know.”
I asked how its Ideas feature decides what to pitch me. It told me, plainly, that the recipe is on Meta’s side of the wall and it can’t see it — then offered to file a feature request asking for more transparency. I asked it to follow an X account; it said it can’t do that, and offered the nearest thing it actually could do instead. Twice in two days it chose the honest answer over the impressive one. I’ve used enough AI products to know that’s a design decision, not an accident. Or if it is an accident, it’s a good one.
It’s not all smooth. This morning I couldn’t find where it keeps my research files in the app, and we did a small dance — wrong folder, a flat file list, renaming everything with a prefix — before it worked. Mundane stuff. The kind of friction that reminds you this is a 48-hour-old product, not a finished one.
And the keyboard thing: I asked how to shrink the keyboard back down, meaning in the Muse app, and it answered about iOS generally before I clarified. Small misfire, corrected in one message. Conversations with it feel like texting a competent friend, which is the highest compliment I can give software I talk to.
Two days is nothing. I don’t know whether this becomes indispensable or fades into the background of apps I tried in September 2026. But the early signal is promising: it does the homework, waits its turn, and tells the truth about what it can’t see. That’s a better foundation than most relationships I have with technology.