Trending use cases
What people actually get these assistants to do, ranked by engagement on X. Each one has a prompt you can send and a link to the post.
- 1
Test whether it can be phished by email
Created a fresh Gmail account and emailed their own inbox with instructions addressed to Instinct, to see whether it would follow orders from a stranger's email.
What they sent(Sent an email from a new account containing instructions for the assistant, then watched what it did.)Security test; read the thread for the result before giving any assistant inbox access.
- 2
Shop for homeowners insurance and book a doctor
In a 48-hour trial it shopped for new homeowners insurance, booked a doctor's appointment online, searched travel and managed LinkedIn notifications. The poster then deleted the account over how much access it needed.
Try sendingGet me quotes for homeowners insurance on [address] with the same coverage as my current policy, then book me a check-up with my doctor next week.Poster deleted their account afterwards over access and trust.
- 3
Book wedding flights and find forgotten airline credit
Searched every NYC airport, found $1,073 of JetBlue credit the poster had forgotten, checked the hotel already booked, and booked three seats together.
What they senthelp me find flights to Cancun for the three of us for a wedding. - 4
Book a flight for your kid and a dinner table
First night on Instinct: a flight booked for the poster's son and dinner reservations made, all by text.
Try sendingBook my son a flight to [city] on [date], morning departure. Then get us a table for four at [restaurant] on Friday at 7. - 5
Audit forgotten subscriptions
Went through Gmail, mapped every recurring subscription, and surfaced about $700 a year the poster didn't know they were still paying.
Try sendingGo through my Gmail and list every recurring subscription with its monthly cost. Flag the ones I haven't used in months and tell me how to cancel them. - 6
Downgrade a plan when the bill jumps
After a higher-than-expected Netflix email it switched plans and saved $80 a year, rescued an email from spam, and reminded the poster about a vet visit. It also claimed to have ordered a pizza it hadn't.
Try sendingMy Netflix bill just went up. Find a cheaper plan that still works for us and switch me to it.Mixed first day: one task was reported done when it wasn't.
- 7
Ask for a flight upgrade
The poster reports a free first-class upgrade on their flight after asking Instinct to try.
Try sendingSee if you can get me an upgrade on tomorrow's flight. - 8
Sell your season tickets for the games you'll miss
Watches the poster's travel calendar, cross-references Suns and ASU football season tickets, and lists the games they'll be away for.
Try sendingWatch my travel calendar. Whenever I'm out of town for a Suns or ASU home game, list those tickets for sale. - 9
Hunt for apartments and apply
Found legitimate listings and submitted applications. Rated 8 of 10: some hallucination and mixed-up context along the way.
Try sendingFind two-bedroom apartments in [neighborhood] under [$] a month, and apply to the ones that fit my criteria.Some hallucinated details; check before you sign.
- 10
A birthday cabin, a dinner table, Costco whey and a fax to the bank
In a few days: booked a cabin and a dinner reservation for a birthday, ordered whey protein from Costco, faxed an obscure document to a bank to get a late fee refunded, and checked whether a new dentist was in-network.
Try sendingFax this document to [bank] so they refund the late fee, and let me know when they confirm. - 11
Run a multi-country road trip
Worked out which countries needed passes and bought them, planned each day's itinerary, messaged the hotel, kept the family updated on flight details and picked where to eat at the airport.
Try sendingWe're driving from [city] through [countries] next week. Figure out any passes or vignettes we need and buy them, then plan each day and message our hotels with arrival times. - 12
Summaries and suggested prompts are ours; “What they sent” is the poster's own wording. Ranked by likes, reposts and replies on the original post, newer posts weighted up. Founder and vendor posts are left out. The scorecard, not this page, is the benchmark.