Three Days With Muse
I started trying Muse a few days ago. I tried it out of curiosity, expecting a chatbot. What I got feels closer to a colleague.
If you haven't met it: Muse is Meta's personal AI assistant. It chats like you'd expect, but it also has its own computer — it can write code, browse the web, and connect to everyday services.
We've mostly worked on my blog — I describe what I want in plain words, it does the building, and I review everything before it goes live. My passwords stay mine; it never asked. When something broke, I pasted the error and we debugged it together until it worked.
The fun part surprised me. I asked it to design a share image, and I turned into an art director: several rejected drafts later, we landed on one we both liked. It never got tired or offended. Then I asked for something random from my music — it dug through my playlists and picked a song I'd forgotten I loved. It couldn't actually press play, so we just sat with the recommendation.
Three days in, my rules are simple: describe the outcome, show the raw error, and give away as little access as needed. It does the tedious parts and shows its work. What it can touch is limited — and I can take it back anytime.