Meta's new AI app does your errands for you.
It books things, fills in forms, sends emails and shops.
And it keeps going after you close the app.
(Which is either a dream or the opening scene of a horror film. Depends on your settings.)
It's called Muse. It launched on September 8. Ten days later it was the #1 app on the US App Store.
That's a lot of people letting an AI loose on their errands in ten days.
So here's the short version: Muse is useful, but the default settings trust it more than I would.
Four switches fix that. Five minutes, tops.
Now a quick word from today's sponsor.
1,000+ Claude Prompts Top Professionals Actually Use at Work
Claude can be your analyst, editor, and strategist.
But most professionals are using it to fix grammar.
These 1,000+ Claude prompts take it from grammar tool to your most powerful AI work assistant.
Sign up for Superhuman AI and get:
1,000+ ready-to-use Claude prompts to get real work done in minutes — researched, tested, and used by professionals at Google, Microsoft, and NASA
Superhuman AI newsletter (4 min daily) so you keep learning new AI tools and skills to stay ahead in your career — the prompts are just the beginning
Right. Back to it.
Errands it'll do for you
You chat with it like any other AI app. The difference is that it acts.
Meta's own example: send it a recipe reel you saved on Instagram, and it turns it into a grocery list.
It goes bigger too. Booking travel, filling in forms, selling your old stuff on Marketplace, even trying to lower your bills.
It shops at Walmart, Best Buy, Sephora, Instacart, Expedia and so on, and pays through the usual suspects like PayPal and Stripe.
(Amazon blocked it on September 21. Amazon would rather do your shopping itself, thanks.)
It's free with a usage limit. Heavy users can pay $20 or $100 a month for more.
Keep control before you connect anything
Muse can only touch what you connect it to.
The catch is that connecting is easy. Slightly too easy.
If your Facebook, Instagram and Threads live in the same Meta account, Muse gets all three automatically. You don't click anything.
And the first horror stories are already in.
A columnist at Inc found Muse had copied more than 187,000 of his text messages off his Mac.
Meta says that sync is opt-in. It also said, and I quote: "That's on us."
Another user says Muse accepted a lowball Marketplace offer and shared his home address. The buyer turned up at 9:15pm.
Meta says Muse asked for permission first.
I'd still use it, just not on the default settings.
Four switches that make it ask first
Open Muse and tap Settings.
Permissions > Connectors: set it to "Always ask." Now Muse checks with you before it does anything with your accounts. The default only asks for "some actions." Not all of them.
Permissions > Web access: "Always ask" here too. Now it asks before it opens any website.
Connectors: go read-only where you can. Many connectors let Muse look without touching. It can read your inbox, but it can't send anything as you.
Data controls: turn off training. Your chats train Meta's AI by default. Switching it off covers your old chats too.
When Muse asks permission, you get five buttons. Tap "Allow once."
"Always allow" means it stops asking for good. I wouldn't go near that button in week one.
On a Mac, don't give it Full Disk Access. That's the permission behind the text message story.
A safe first job to try today
Meta's own advice is to start with low-risk tasks. I'd go further: start with jobs where the worst case is a bad list.
Connect your calendar and email as read-only where you can, then try this:
Look at my calendar for the next 7 days and my last 20 emails. Make me one list of every date, deadline and thing I've promised someone, sorted by date. Don't send, book or buy anything. Just the list.Then send it the recipe reel. Fun, harmless, and you get dinner out of it.
My rule for any AI agent: it can read almost anything, and it does nothing without asking. You loosen that one job at a time, once it's earned it.
Make it stick
Once a week, tap the assistant icon and open the Activity log. It shows what Muse has been up to.
Then ask it: "What do you remember about me?"
If it knows something it shouldn't, tell it to forget it. It will.
That's an AI running your errands while you keep the final say. If you run a business, that's exactly how Run On Claude works.
Every month, Claude takes one job off your plate. Right now that's your social posts and your customer emails, with a new job each month picked by member vote.
You set each job up once with a 20-minute guide. Claude writes the drafts, you read them and hit send.
If it hasn't saved you 5 hours in your first 30 days, you get your first month back.
Business owners only. $147 a month until we hit 50 members.
Talk tomorrow,
Zephyr
P.S. What's the first errand you'd hand an AI agent? Hit reply. I'm collecting the good ones.