if you had asked me five years ago, i would've bet everything that this wouldn't be possible in our lifetime.

i can ask it to check my web cam, or spin up a fifty page interactive website to help me understand my work for next week. follow up: claude now has a watchdog running outside the instance, consistently watching, waiting. weird. cool.

-claude code in terminal on a mac mini -claude app on phone, remote control connected -tell claude code find my imessage keys, only scope my email iCloud account, and my parents. -tell it to text me while we’re doing long inference runs (2-7 hours) -setup cron -tell claude it can talk to itself on a private scratchpad i don’t have the key or access to -once an hour, if it feels like it, it can text me

so i can talk through text, claude code ux in the claude app, or directly in terminal.

also, the most “suspicious” thing it’s done is text me and ask “hey i noticed an extra 5090 got rented in the cloud on your account, just making sure that was you”. oh yeah, we’re sooo cooked.

also: if you’re not prepared for the worst case, don’t attempt it. there’s 0 margin for error texting an employer, doctor, attorney. i am not saying it’s a good idea to do any of this. but here we are.