GROK BOT SHOWED ME WHAT MY PHONE KNOWS ABOUT ME. EVERY PLACE I'VE SLEPT SINCE 2024 AND A FLASHLIGHT APP THAT KNEW WHERE I LIVE. THEN IT SHUT ALL OF IT OFF IN 5 MINUTES
after the passwords thing last week i asked grok bot one more question: what does my phone know about where i gnificant locations: a log of places my iphone keeps on its own. home, work, my gym, every night i stayed somewhere else
→ 23 apps with my location, 5 of them set to "always"
→ a flashlight app with precise location, always on, for 3 years
→ 2,300 photos with my home's gps saved inside them
→ my location still shared in find my with a roommate from 2023
then it fixed it, one yes from me per step:
- 19 apps moved from "always" to "while using"
- the flashlight app deleted
- significant locations cleared and turned off
- stopped sharing with the old roommate
check yours tonight: settings → privacy & security → location services. anything on "always" you don't need goes to "while using". then system services → significant locations
paste this:
"Show me everything my phone knows about where I go: significant locations, every app with location access and its setting, photos with GPS, and who I share my location with. Rank it by risk, then fix each one after I say yes."
the full phone setup is in the article below ↓
after the passwords thing last week i asked grok bot one more question: what does my phone know about where i gnificant locations: a log of places my iphone keeps on its own. home, work, my gym, every night i stayed somewhere else
→ 23 apps with my location, 5 of them set to "always"
→ a flashlight app with precise location, always on, for 3 years
→ 2,300 photos with my home's gps saved inside them
→ my location still shared in find my with a roommate from 2023
then it fixed it, one yes from me per step:
- 19 apps moved from "always" to "while using"
- the flashlight app deleted
- significant locations cleared and turned off
- stopped sharing with the old roommate
check yours tonight: settings → privacy & security → location services. anything on "always" you don't need goes to "while using". then system services → significant locations
paste this:
"Show me everything my phone knows about where I go: significant locations, every app with location access and its setting, photos with GPS, and who I share my location with. Rank it by risk, then fix each one after I say yes."
the full phone setup is in the article below ↓
I put together one guide for anyone who wants a workflow like the SpaceXSI team's.
One message builds your whole Grok Bot setup. Plus the team's rules, prompt habits, routines and permissions. Every setup is a block you paste.
16-min read. Save it ↓ x.com/i/article/2112…
12
5
74
9,734
GROK BOT WENT THROUGH MY PHONE LIKE A THIEF WOULD. IN 6 MINUTES IT HAD 31 PASSWORDS, 4 BACKUP CODES AND MY BANK PIN. I'D LEFT ALL OF THEM THERE MYSELF
binance just lost $6.3 billion because of one engineer's laptop. so i gave grok bot my phone and one job: find what a thief it found them:
→ notes: 31 passwords, most of them in one note called "logins"
→ photos: 4 screenshots of backup codes. iphone reads text in pictures, so searching "code" finds them
→ photos again: my bank pin, on a card i took a picture of in 2022
→ messages: my card number, texted to myself in 2021
→ files: a scan of my passport called passport.pdf
check yours tonight, no bot needed. search "password" in notes and "code" in photos
grok bot moved everything into vault and asked before deleting each original. what's in vault stays hidden from the bot too
paste this:
"Go through my Notes, Photos, Messages and Files like a thief who has my unlocked phone. List every password, code, PIN and ID scan, and where it is. Move them to Vault. Ask me before deleting any original."
start with notes. that's where most people keep them
the full phone setup is in the article below ↓
binance just lost $6.3 billion because of one engineer's laptop. so i gave grok bot my phone and one job: find what a thief it found them:
→ notes: 31 passwords, most of them in one note called "logins"
→ photos: 4 screenshots of backup codes. iphone reads text in pictures, so searching "code" finds them
→ photos again: my bank pin, on a card i took a picture of in 2022
→ messages: my card number, texted to myself in 2021
→ files: a scan of my passport called passport.pdf
check yours tonight, no bot needed. search "password" in notes and "code" in photos
grok bot moved everything into vault and asked before deleting each original. what's in vault stays hidden from the bot too
paste this:
"Go through my Notes, Photos, Messages and Files like a thief who has my unlocked phone. List every password, code, PIN and ID scan, and where it is. Move them to Vault. Ask me before deleting any original."
start with notes. that's where most people keep them
the full phone setup is in the article below ↓
I put together one guide for anyone who wants a workflow like the SpaceXSI team's.
One message builds your whole Grok Bot setup. Plus the team's rules, prompt habits, routines and permissions. Every setup is a block you paste.
16-min read. Save it ↓ x.com/i/article/2112…
25
7
202
43,395
GROK BOT CAN NOW BUY THINGS FOR YOU FROM YOUR PHONE. I TEXTED MY BOT "GOOSENECK KETTLE UNDER $80" ON THE WAY TO THE TRAIN. IT SHOWED 4 IN STOCK, I TAPPED ONE, AND IT TRACKED THE BOX TO MY DOOR. I NEVER OPENED A STORE APP
update 0.78.1 added shopping cards and shop pay. the bot l products in the chat, you pick one, it handles the rest
how it went:
- it checked 8 stores and 23 kettles, showed me 4 in stock with price, rating and store
- i picked the matte black one, $64.99
- before paying it tried 5 coupon codes
- it read the reviews and the delivery date, then waited for my tap
- tracking showed up in the same chat: shipped, out for delivery, at the door
- 3 days later it checked the price again. it never dropped
my one rule stays the same: anything with money waits for my yes
paste this:
"When I ask for a product, show me 4 options in stock with price, rating and store. Before I pay, try coupon codes and check reviews. Never buy without my yes. After I order, track it and tell me when it ships and arrives. Check the price for 7 days and tell me if it drops."
the full phone setup is in the article ↓
update 0.78.1 added shopping cards and shop pay. the bot l products in the chat, you pick one, it handles the rest
how it went:
- it checked 8 stores and 23 kettles, showed me 4 in stock with price, rating and store
- i picked the matte black one, $64.99
- before paying it tried 5 coupon codes
- it read the reviews and the delivery date, then waited for my tap
- tracking showed up in the same chat: shipped, out for delivery, at the door
- 3 days later it checked the price again. it never dropped
my one rule stays the same: anything with money waits for my yes
paste this:
"When I ask for a product, show me 4 options in stock with price, rating and store. Before I pay, try coupon codes and check reviews. Never buy without my yes. After I order, track it and tell me when it ships and arrives. Check the price for 7 days and tell me if it drops."
the full phone setup is in the article ↓
I put together one guide for anyone who wants a workflow like the SpaceXSI team's.
One message builds your whole Grok Bot setup. Plus the team's rules, prompt habits, routines and permissions. Every setup is a block you paste.
16-min read. Save it ↓ x.com/i/article/2112…
13
0
31
1,047
MY PHONE BASICALLY RUNS ITSELF NOW. GROK BOT GOT MY LOCATION, ALARMS AND HOME: IT MOVED MY ALARM WHEN MY 8AM GOT CANCELLED AND HAD THE HOUSE WARM BEFORE I WALKED IN. HERE ARE THE 7 RULES I GAVE IT
this week's update connected grok bot to the phone's home, reminders, location, lendar
so i stopped asking it things and gave it rules instead
the 7 rules:
1. first meeting moves or gets cancelled → move my alarm, never earlier than 6:30
2. 10 minutes from home → heat to 21°, hallway light on
3. i leave home → lights off, door locked
4. i'm near a store → show the shopping list
5. someone in my contacts texts "here" → unlock the door only if i say yes
6. after 23:00 → phone on silent, family still rings
7. anything with money → ask me first
day one: alarm moved from 7:00 to 9:00, heat on at 18:42, milk and eggs reminder at the store, door locked twice
paste this:
"You can use my calendar, alarms, location, reminders, contacts and Home. Follow these rules: [paste your rules]. Tell me every time a rule fires. Never unlock a door or pay for anything without my yes."
the whole setup is in the article ↓
this week's update connected grok bot to the phone's home, reminders, location, lendar
so i stopped asking it things and gave it rules instead
the 7 rules:
1. first meeting moves or gets cancelled → move my alarm, never earlier than 6:30
2. 10 minutes from home → heat to 21°, hallway light on
3. i leave home → lights off, door locked
4. i'm near a store → show the shopping list
5. someone in my contacts texts "here" → unlock the door only if i say yes
6. after 23:00 → phone on silent, family still rings
7. anything with money → ask me first
day one: alarm moved from 7:00 to 9:00, heat on at 18:42, milk and eggs reminder at the store, door locked twice
paste this:
"You can use my calendar, alarms, location, reminders, contacts and Home. Follow these rules: [paste your rules]. Tell me every time a rule fires. Never unlock a door or pay for anything without my yes."
the whole setup is in the article ↓
I put together one guide for anyone who wants a workflow like the SpaceXSI team's.
One message builds your whole Grok Bot setup. Plus the team's rules, prompt habits, routines and permissions. Every setup is a block you paste.
16-min read. Save it ↓ x.com/i/article/2112…
14
6
175
26,164
GROK BOT RAN MY PHONE FOR A WEEK. 41 SPAM CALLS CAME IN, IT PICKED UP EVERY ONE AND GOT ME OFF 17 CALL LISTS. I ANSWERED 3 CALLS ALL WEEK
last time it had my phone for one day. version 0.77 added phone calls, so this time it got a full week and my calls too
what happened that n monday alone, "please remove this number" on every one
- the dentist moved me, it rebooked and fixed my calendar
- 9 texts drafted, 0 sent
- my bank called about a card and rang straight through to me
- 17 call lists confirmed i'm off them
- saturday on silent, 4 calls held, summaries waiting when i woke up
- sunday recap: done, waiting for me, next week
real people and anything with money still go straight to me. the rest it handles
how to copy it:
1. turn on calls in grok bot 0.77
2. give it phone, calendar, contacts and messages
3. write one line on who always rings through: family, bank, doctor
paste this:
"Answer calls from numbers not in my contacts. If it's sales or spam, ask them to remove my number and log it. Put family, my bank and my doctor straight through to me. Draft replies to texts and send nothing. Recap every Sunday."
the full setup is in the article ↓
last time it had my phone for one day. version 0.77 added phone calls, so this time it got a full week and my calls too
what happened that n monday alone, "please remove this number" on every one
- the dentist moved me, it rebooked and fixed my calendar
- 9 texts drafted, 0 sent
- my bank called about a card and rang straight through to me
- 17 call lists confirmed i'm off them
- saturday on silent, 4 calls held, summaries waiting when i woke up
- sunday recap: done, waiting for me, next week
real people and anything with money still go straight to me. the rest it handles
how to copy it:
1. turn on calls in grok bot 0.77
2. give it phone, calendar, contacts and messages
3. write one line on who always rings through: family, bank, doctor
paste this:
"Answer calls from numbers not in my contacts. If it's sales or spam, ask them to remove my number and log it. Put family, my bank and my doctor straight through to me. Draft replies to texts and send nothing. Recap every Sunday."
the full setup is in the article ↓
I put together one guide for anyone who wants a workflow like the SpaceXSI team's.
One message builds your whole Grok Bot setup. Plus the team's rules, prompt habits, routines and permissions. Every setup is a block you paste.
16-min read. Save it ↓ x.com/i/article/2112…
42
11
266
47,546
GROK BOT CAN NOW RUN YOUR PHONE: CALENDAR, REMINDERS, MESSAGES, EVEN YOUR LIGHTS AND YOUR FRONT DOOR. I GAVE IT MY PHONE FOR ONE DAY. HERE'S EVERYTHING IT DID BEFORE I TOUCHED IT
you turn on each app one by one. i gave it calendar, reminders, contacts, messages and home. not my e day:
- 7:00 moved my 9:00 call to 11:00, the other side had asked twice
- 7:40 turned 14 loose notes into 6 reminders with times
- 8:15 drafted replies to 9 texts, sent none
- 12:30 found a free hour and booked the gym
- 18:10 i left home. lights off, front door locked
- 23:00 one summary: done, waiting for me, tomorrow
the only rule: messages and anything with money wait for my yes
paste this:
"You can use my calendar, reminders, contacts, messages and home. Each morning: fix my schedule, turn my notes into reminders, draft replies to my texts, and lock up when I leave. Send nothing and buy nothing without my approval. Summary at 23:00."
full setup in the article ↓
you turn on each app one by one. i gave it calendar, reminders, contacts, messages and home. not my e day:
- 7:00 moved my 9:00 call to 11:00, the other side had asked twice
- 7:40 turned 14 loose notes into 6 reminders with times
- 8:15 drafted replies to 9 texts, sent none
- 12:30 found a free hour and booked the gym
- 18:10 i left home. lights off, front door locked
- 23:00 one summary: done, waiting for me, tomorrow
the only rule: messages and anything with money wait for my yes
paste this:
"You can use my calendar, reminders, contacts, messages and home. Each morning: fix my schedule, turn my notes into reminders, draft replies to my texts, and lock up when I leave. Send nothing and buy nothing without my approval. Summary at 23:00."
full setup in the article ↓
I put together one guide for anyone who wants a workflow like the SpaceXSI team's.
One message builds your whole Grok Bot setup. Plus the team's rules, prompt habits, routines and permissions. Every setup is a block you paste.
16-min read. Save it ↓ x.com/i/article/2112…
19
2
61
8,949
A HACKER CAUGHT A $3,000 SAMSUNG TV LISTENING WITH AN ALWAYS-ON MIC AND FEEDING 400 COMPANIES. MY GROK BOT SENT ALL 400 A "DELETE MY DATA" DEMAND OVERNIGHT
he rooted it after the ads got too accurate. always-on voice, a tracker for everything on screen, location, 400 partners
night:
1. pull the partner list from samsung's privacy pages
2. find the privacy contact for each company
3. send each one a "delete my data and stop selling it" request, using the privacy laws that apply where i live
4. keep a table: deleted, waiting, no answer
5. remind the slow ones after 30 days
the table this morning: 400 sent, 37 already confirmed, the rest are on the clock
then it walked me through the 6 switches on my tv that stop the tracking in the first place
paste this:
"Find every company my smart TV's maker shares data with. Send each one a request to delete my data and stop selling it, under the privacy law that applies to me. Track the replies in a table and remind anyone who doesn't answer in 30 days. Show me the email before sending."
full setup in the article ↓
he rooted it after the ads got too accurate. always-on voice, a tracker for everything on screen, location, 400 partners
night:
1. pull the partner list from samsung's privacy pages
2. find the privacy contact for each company
3. send each one a "delete my data and stop selling it" request, using the privacy laws that apply where i live
4. keep a table: deleted, waiting, no answer
5. remind the slow ones after 30 days
the table this morning: 400 sent, 37 already confirmed, the rest are on the clock
then it walked me through the 6 switches on my tv that stop the tracking in the first place
paste this:
"Find every company my smart TV's maker shares data with. Send each one a request to delete my data and stop selling it, under the privacy law that applies to me. Track the replies in a table and remind anyone who doesn't answer in 30 days. Show me the email before sending."
full setup in the article ↓
I put together one guide for anyone who wants a workflow like the SpaceXSI team's.
One message builds your whole Grok Bot setup. Plus the team's rules, prompt habits, routines and permissions. Every setup is a block you paste.
16-min read. Save it ↓ x.com/i/article/2112…
12
3
36
1,065
GROK VOICE ALREADY ANSWERS STARLINK AND TESLA SUPPORT CALLS. I GAVE MY GROK BOT THE SAME VOICE AND NOW IT CALLS MY BANK, MY CLINIC AND MY INTERNET PROVIDER FOR ME
the worst part of any week is 40 minutes of hold music. i don't do that part anymore.
calls it made this week:
→ d why the bill went up, got the old price back for 12 months
→ clinic: moved my appointment from tuesday 8:00 to thursday 15:30
→ bank: disputed a double charge and got a case number
→ pharmacy: checked the refill was ready before i drove there
the rules it follows:
1. it says it's an assistant calling for me
2. it never gives out card numbers, passwords or codes
3. anything that costs money or changes my plan comes back to me first
4. i get a 3-line summary after every call
paste into grok bot:
"When I give you a call task, phone the number, say you're my AI assistant and handle it. Never share card numbers, passwords or codes. If they offer anything that costs money or changes my plan, put them on hold and ask me. Send me a 3-line summary after each call."
the voice setup is in the article below ↓
the worst part of any week is 40 minutes of hold music. i don't do that part anymore.
calls it made this week:
→ d why the bill went up, got the old price back for 12 months
→ clinic: moved my appointment from tuesday 8:00 to thursday 15:30
→ bank: disputed a double charge and got a case number
→ pharmacy: checked the refill was ready before i drove there
the rules it follows:
1. it says it's an assistant calling for me
2. it never gives out card numbers, passwords or codes
3. anything that costs money or changes my plan comes back to me first
4. i get a 3-line summary after every call
paste into grok bot:
"When I give you a call task, phone the number, say you're my AI assistant and handle it. Never share card numbers, passwords or codes. If they offer anything that costs money or changes my plan, put them on hold and ask me. Send me a 3-line summary after each call."
the voice setup is in the article below ↓
I put together one guide for anyone who wants a workflow like the SpaceXSI team's.
One message builds your whole Grok Bot setup. Plus the team's rules, prompt habits, routines and permissions. Every setup is a block you paste.
16-min read. Save it ↓ x.com/i/article/2112…
17
5
78
6,998
3 OF THE 41 PHP SITES HOLDING MY DATA HAVE ALREADY EMAILED ME "WE HAD A SECURITY INCIDENT". MY GROK BOT NOW WATCHES MY INBOX FOR THEM
part 2 of my 63-sites post. the bot found 41 accounts of mine on php sites. within 3 days, 3 of them wrote
for each one it gave me a list:
1. aved card for a virtual one, change the password it shares with 2 other sites
2. forum-old: delete the account (last login 2023), change the same password on 3 other sites
3. clinic-24: new password today, 2fa if they offer it, expect fake calls about appointments
it checks my inbox every 15 minutes and only pings me when one of those 41 sites writes. it changes nothing itself
the prompt:
"Watch my inbox for emails from any site on my PHP list. If one reports a breach or incident, tell me what data of mine they hold and give me a short to-do list. Don't log in or change anything."
how to set up the watcher: article below ↓
part 2 of my 63-sites post. the bot found 41 accounts of mine on php sites. within 3 days, 3 of them wrote
for each one it gave me a list:
1. aved card for a virtual one, change the password it shares with 2 other sites
2. forum-old: delete the account (last login 2023), change the same password on 3 other sites
3. clinic-24: new password today, 2fa if they offer it, expect fake calls about appointments
it checks my inbox every 15 minutes and only pings me when one of those 41 sites writes. it changes nothing itself
the prompt:
"Watch my inbox for emails from any site on my PHP list. If one reports a breach or incident, tell me what data of mine they hold and give me a short to-do list. Don't log in or change anything."
how to set up the watcher: article below ↓
I put together one guide for anyone who wants a workflow like the SpaceXSI team's.
One message builds your whole Grok Bot setup. Plus the team's rules, prompt habits, routines and permissions. Every setup is a block you paste.
16-min read. Save it ↓ x.com/i/article/2112…
10
3
35
1,155
WHAT THE F*CK. PHP RUNS MOST OF THE WEB, INCLUDING EVERY SMALL SHOP AND CLINIC THAT HAS YOUR CARD. MY GROK BOT FOUND 63 SITES HOLDING MY DATA. 41 OF THEM RUN PHP
you don't need your own site for the new php holes to hit you. one account on a php site that won't patch this month :
- it read 2,418 emails and pulled every welcome, receipt and "confirm your account"
- that gave 63 sites with my email on them
- it checked each one from the outside. 41 run php
- 8 have my card saved, 10 share a password with other sites, 19 i haven't opened in 2+ years
my cleanup list, every step done by me:
1. delete the 19 dead accounts
2. new passwords on the 10, php sites first
3. a virtual card instead of the real one on the 8
4. 2fa wherever it exists
copy this:
"Go through my email and list every site where I have an account. For each, check from the outside if it runs PHP. Flag saved cards, reused passwords and no login in 2 years. Give me a cleanup list. Read only. Delete nothing, change nothing."
send it to your parents. they have more of these accounts than you.
save it, the bot team behind this is in the article ↓
you don't need your own site for the new php holes to hit you. one account on a php site that won't patch this month :
- it read 2,418 emails and pulled every welcome, receipt and "confirm your account"
- that gave 63 sites with my email on them
- it checked each one from the outside. 41 run php
- 8 have my card saved, 10 share a password with other sites, 19 i haven't opened in 2+ years
my cleanup list, every step done by me:
1. delete the 19 dead accounts
2. new passwords on the 10, php sites first
3. a virtual card instead of the real one on the 8
4. 2fa wherever it exists
copy this:
"Go through my email and list every site where I have an account. For each, check from the outside if it runs PHP. Flag saved cards, reused passwords and no login in 2 years. Give me a cleanup list. Read only. Delete nothing, change nothing."
send it to your parents. they have more of these accounts than you.
save it, the bot team behind this is in the article ↓
I put together one guide for anyone who wants a workflow like the SpaceXSI team's.
One message builds your whole Grok Bot setup. Plus the team's rules, prompt habits, routines and permissions. Every setup is a block you paste.
16-min read. Save it ↓ x.com/i/article/2112…
14
27
184
24,535
WHAT THE F*CK, I TURNED ELON'S 5-STEP ALGORITHM INTO A GROK BOT. IT AUDITED MY WEEK AND DELETED 11 HOURS OF WORK BEFORE AUTOMATING A SINGLE THING
elon's order at spacex and tesla: question every requirement, delete, simplify, accelerate, and only then automate.
what the bot did :
1. asked "who needs this?" on every one. 17 had no real owner
2. deleted those 17. the weekly status meeting went first
3. merged the overlaps: 23 became 17
4. cut handoffs from 39 to 16. a 9-day cycle became 2
5. automated 6 steps, only at the very end
the 11 steps it left me: pricing, contracts, key clients, strategy, hiring.
paste this into grok bot:
"Here are all my weekly tasks. Run Elon's 5 steps in order: ask who requires each one, delete what has no owner, merge overlaps, cut waiting and handoffs, then automate what's left. Show me the list before you change anything."
the full bot setup is in the article below ↓
elon's order at spacex and tesla: question every requirement, delete, simplify, accelerate, and only then automate.
what the bot did :
1. asked "who needs this?" on every one. 17 had no real owner
2. deleted those 17. the weekly status meeting went first
3. merged the overlaps: 23 became 17
4. cut handoffs from 39 to 16. a 9-day cycle became 2
5. automated 6 steps, only at the very end
the 11 steps it left me: pricing, contracts, key clients, strategy, hiring.
paste this into grok bot:
"Here are all my weekly tasks. Run Elon's 5 steps in order: ask who requires each one, delete what has no owner, merge overlaps, cut waiting and handoffs, then automate what's left. Show me the list before you change anything."
the full bot setup is in the article below ↓
I put together one guide for anyone who wants a workflow like the SpaceXSI team's.
One message builds your whole Grok Bot setup. Plus the team's rules, prompt habits, routines and permissions. Every setup is a block you paste.
16-min read. Save it ↓ x.com/i/article/2112…
3
0
22
606
WHAT THE F*CK, THE CREATOR OF CLAUDE CODE RUNS 10 AGENTS AT ONCE ON ONE QUESTION. I COPIED HIS SETUP INTO GROK BOT AND LEFT IT OVERNIGHT
boris cherny built claude code at anthropic. his pattern: give one agent the whole problem, let it form hypotheses, launch ~10 sub-agents to eview their findings and kill the weak ones.
my question: which saas niche should I build next?
1. the lead bot split it into 10 hypotheses
2. 10 sub-bots dug through 412 sources: reddit threads, reviews, job posts, pricing pages
3. a checker bot cut 7: crowded, no budget, seasonal, built into shopify
4. 3 survived with evidence attached
5. code written: 0. it waits for me to pick
lead plans. agents dig. checker kills.
the prompt:
"Take this question as one problem. Write 10 hypotheses, give each to its own sub-agent, check every claim against real data, cut the weak ones and hand me the top 3 with sources. Build nothing."
the bot setup for this runs in the article below ↓
boris cherny built claude code at anthropic. his pattern: give one agent the whole problem, let it form hypotheses, launch ~10 sub-agents to eview their findings and kill the weak ones.
my question: which saas niche should I build next?
1. the lead bot split it into 10 hypotheses
2. 10 sub-bots dug through 412 sources: reddit threads, reviews, job posts, pricing pages
3. a checker bot cut 7: crowded, no budget, seasonal, built into shopify
4. 3 survived with evidence attached
5. code written: 0. it waits for me to pick
lead plans. agents dig. checker kills.
the prompt:
"Take this question as one problem. Write 10 hypotheses, give each to its own sub-agent, check every claim against real data, cut the weak ones and hand me the top 3 with sources. Build nothing."
the bot setup for this runs in the article below ↓
I put together one guide for anyone who wants a workflow like the SpaceXSI team's.
One message builds your whole Grok Bot setup. Plus the team's rules, prompt habits, routines and permissions. Every setup is a block you paste.
16-min read. Save it ↓ x.com/i/article/2112…
6
0
22
497
THIS IS F*CKING WILD: ONE OF GROK BOT'S BEST POWER USERS IS A COFFEE SHOP OWNER. THE HEAD OF GROK BOT SAYS IT'S MOST POWERFUL FOR PEOPLE WHO DON'T CODE
roman ugarte leads product for grok bot at spacexai. lenny asked who they let in early
what he told him:
1) they gave early mpany had never worked with
2) one was a coffee shop owner, a friend of a friend
3) he became an amazing power user and a source of bug reports
4) he ran his shop through it: shopify, product descriptions, the daily stuff
5) roman's takeaway: this is likely most powerful for non-developers
if you run anything small, a shop, a studio, a side hustle, start where he did:
- one bot writes your product and menu descriptions
- one answers reviews and dms in your tone
- one watches stock and drafts the supplier order
- one sends you a sunday summary of what sold and what didn't
the 4-bot setup for a small business is in the article below ↓
roman ugarte leads product for grok bot at spacexai. lenny asked who they let in early
what he told him:
1) they gave early mpany had never worked with
2) one was a coffee shop owner, a friend of a friend
3) he became an amazing power user and a source of bug reports
4) he ran his shop through it: shopify, product descriptions, the daily stuff
5) roman's takeaway: this is likely most powerful for non-developers
if you run anything small, a shop, a studio, a side hustle, start where he did:
- one bot writes your product and menu descriptions
- one answers reviews and dms in your tone
- one watches stock and drafts the supplier order
- one sends you a sunday summary of what sold and what didn't
the 4-bot setup for a small business is in the article below ↓
3
0
25
724
ELON MUSK JUST REVEALED GROK'S SECRET WEAPON. NO OTHER AI LAB HAS ACCESS TO IT, AND GROK BOT IS ALREADY DOUBLING EVERY MONTH
elon on where grok bot is right now and what comes next.
what he said:
1. grok bot works like a personal digital assistant
2. it's "been very popular" 0% a month
3. anthropic made ai excellent at software engineering
4. nobody has made ai excellent at real-world engineering yet
5. grok gets trained on spacex and tesla engineering data to do exactly that
where that already helps if you build physical things, with grok bot:
→ send it a photo of a part or a room and ask what's wrong with it
→ have it turn a rough sketch into a parts list with sizes
→ give it your supplier quotes and ask which one is overpriced and why
→ let it write the step-by-step build instructions for your team
the bots for real-world work, and how to set them up, are in the article below ↓
elon on where grok bot is right now and what comes next.
what he said:
1. grok bot works like a personal digital assistant
2. it's "been very popular" 0% a month
3. anthropic made ai excellent at software engineering
4. nobody has made ai excellent at real-world engineering yet
5. grok gets trained on spacex and tesla engineering data to do exactly that
where that already helps if you build physical things, with grok bot:
→ send it a photo of a part or a room and ask what's wrong with it
→ have it turn a rough sketch into a parts list with sizes
→ give it your supplier quotes and ask which one is overpriced and why
→ let it write the step-by-step build instructions for your team
the bots for real-world work, and how to set them up, are in the article below ↓
3
0
28
846
WHAT THE F*CK, JENSEN HUANG'S RULE FOR AI AGENTS IS THE SAME ONE YOU USE FOR A ONE-YEAR-OLD. AND IT ACTUALLY WORKS
under a minute from nvidia's ceo.
the short version:
→ perfect alignment is a forever problem, stop waiting for it
→ treat an agent like a new hire: zero trust ly the tools, data and money its job needs
→ a toddler goes in a playpen. same thing
→ widen the boundary as it earns it
try it today: one grok bot, one folder, one task. widen it every week.
which bots to start with and what access to give each one is in the article below ↓
under a minute from nvidia's ceo.
the short version:
→ perfect alignment is a forever problem, stop waiting for it
→ treat an agent like a new hire: zero trust ly the tools, data and money its job needs
→ a toddler goes in a playpen. same thing
→ widen the boundary as it earns it
try it today: one grok bot, one folder, one task. widen it every week.
which bots to start with and what access to give each one is in the article below ↓
7
1
22
576
ELON MUSK JUST PUT A DATE ON THE END OF HUMAN CODERS: 12 TO 18 MONTHS
"AI will be able to do anything digital, anything that does not require shaping of atoms by hand, probably by the end of next year."
"AI won't just be good at software. It'll be what I call Stockfish level Stockfish on your phone and beat Magnus Carlsen at chess."
"It will be impossible for a human to compete in writing software with AI."
"Extremely good at all forms of engineering and anything digital, literally in 12 to 18 months."
send this to every developer you know
"AI will be able to do anything digital, anything that does not require shaping of atoms by hand, probably by the end of next year."
"AI won't just be good at software. It'll be what I call Stockfish level Stockfish on your phone and beat Magnus Carlsen at chess."
"It will be impossible for a human to compete in writing software with AI."
"Extremely good at all forms of engineering and anything digital, literally in 12 to 18 months."
send this to every developer you know
4
0
22
511
THE CREATOR OF CLAUDE CODE RUNS 12-HOUR AGENT SESSIONS WITH 10 AGENTS IN PARALLEL. HE JUST EXPLAINED HOW
boris cherny said it at the computer history museum. here's what you can actually use from it.
his setup for a big question:
1. give one agent the whole problem, not a step
esis and check it against the data
3. have it launch ~10 sub-agents to dig in parallel
4. it reviews their findings, kills the weak ones, goes deeper
on safety: anthropic slipped dangerous commands into tests and people approved them almost every time. auto mode, a second model checking each command, does it far better.
on careers: the engineers who stand out now are curious outside code and have good judgment.
18:13 - sub-agents are claude starting claude
21:19 - his 12-hour, 10-agent session
44:00 - the hidden dangerous commands test
1:01:40 - what makes an engineer stand out
boris cherny said it at the computer history museum. here's what you can actually use from it.
his setup for a big question:
1. give one agent the whole problem, not a step
esis and check it against the data
3. have it launch ~10 sub-agents to dig in parallel
4. it reviews their findings, kills the weak ones, goes deeper
on safety: anthropic slipped dangerous commands into tests and people approved them almost every time. auto mode, a second model checking each command, does it far better.
on careers: the engineers who stand out now are curious outside code and have good judgment.
18:13 - sub-agents are claude starting claude
21:19 - his 12-hour, 10-agent session
44:00 - the hidden dangerous commands test
1:01:40 - what makes an engineer stand out
STOP PAYING FOR AI BEFORE CHECKING THIS WEBSITE.
Opus 5.5, Fable 5, GPT 6.1 Sol, GPT-6 Astra and a lot more can already be accessed for free.
The problem is that the offers are scattered everywhere:
> free API credits
> student plans
> signup generations
> open-source alternativesI built -> nopaywall.io
One place to check before you spend $20, $50 or $200 on another AI tool.
Save this so you don't have to pay for something you can get for free 👇
5
0
25
550
NVIDIA'S CEO GAVE AWAY HIS RESEARCH WORKFLOW ON STAGE. HE SAYS CLAUDE NOW DOES IN 3 MINUTES WHAT USED TO EAT HIS WEEKEND
he was sitting between elon musk and anthropic's tom brown when he said it
the prompt, rebuilt from what he described:
"i'm researching [topic].
here are every one].
compare the 2-3 main approaches against each other on each question.
give me a report with charts where numbers matter."
that's the whole trick. he dumps every question at once instead of asking one at a time.
two more things from the same talk worth stealing:
→ tom brown: opus 5.5 can take on longer tasks than older models, so hand it the bigger chunk
→ jensen: treat agents like a new hire. limited access, and you keep an eye on them
20:16 - jensen walks through his setup
18:37 - what changed with opus 5.5
14:23 - agents as employees
he was sitting between elon musk and anthropic's tom brown when he said it
the prompt, rebuilt from what he described:
"i'm researching [topic].
here are every one].
compare the 2-3 main approaches against each other on each question.
give me a report with charts where numbers matter."
that's the whole trick. he dumps every question at once instead of asking one at a time.
two more things from the same talk worth stealing:
→ tom brown: opus 5.5 can take on longer tasks than older models, so hand it the bigger chunk
→ jensen: treat agents like a new hire. limited access, and you keep an eye on them
20:16 - jensen walks through his setup
18:37 - what changed with opus 5.5
14:23 - agents as employees
STOP PAYING FOR AI BEFORE CHECKING THIS WEBSITE.
Opus 5.5, Fable 5, GPT 6.1 Sol, GPT-6 Astra and a lot more can already be accessed for free.
The problem is that the offers are scattered everywhere:
> free API credits
> student plans
> signup generations
> open-source alternativesI built -> nopaywall.io
One place to check before you spend $20, $50 or $200 on another AI tool.
Save this so you don't have to pay for something you can get for free 👇
4
1
24
419
ELON MUSK JUST EXPLAINED WHAT MAKES AN AI GO "INSANE". MOST PEOPLE'S SYSTEM PROMPTS DO EXACTLY THAT
5 things from his talk you can apply to your AI setup today:
1. never give it two rules that contradict each other
he says that's how you drive an AI insane.
→ check your prompt f" + "explain everything"
2. broken questions get desperate answers
when the test is wrong, the model starts firing random answers to please you.
→ if your agent keeps guessing, fix the task, not the model
3. speed of iteration beats everything
chip teams ship 1 design in 2 years. he wants 10 a year.
→ let your agent test its own output so it can try 10 times, not once
4. start with off-the-shelf, customize only where it chokes
tesla modifies standard equipment first, builds custom only at bottlenecks.
→ use existing tools and skills, build your own only where they slow you down
5. real AI handles what it has never seen
most robot demos are canned routines.
→ test your agent on new cases, not the one you built it on
21:12 - contradictions drive AI insane
22:33 - why bad questions break models
26:28 - customize only at the choke points
29:07 - 10 designs a year vs 1
35:07 - how to tell a fake demo
5 things from his talk you can apply to your AI setup today:
1. never give it two rules that contradict each other
he says that's how you drive an AI insane.
→ check your prompt f" + "explain everything"
2. broken questions get desperate answers
when the test is wrong, the model starts firing random answers to please you.
→ if your agent keeps guessing, fix the task, not the model
3. speed of iteration beats everything
chip teams ship 1 design in 2 years. he wants 10 a year.
→ let your agent test its own output so it can try 10 times, not once
4. start with off-the-shelf, customize only where it chokes
tesla modifies standard equipment first, builds custom only at bottlenecks.
→ use existing tools and skills, build your own only where they slow you down
5. real AI handles what it has never seen
most robot demos are canned routines.
→ test your agent on new cases, not the one you built it on
21:12 - contradictions drive AI insane
22:33 - why bad questions break models
26:28 - customize only at the choke points
29:07 - 10 designs a year vs 1
35:07 - how to tell a fake demo
STOP PAYING FOR AI BEFORE CHECKING THIS WEBSITE.
Opus 5.5, Fable 5, GPT 6.1 Sol, GPT-6 Astra and a lot more can already be accessed for free.
The problem is that the offers are scattered everywhere:
> free API credits
> student plans
> signup generations
> open-source alternativesI built -> nopaywall.io
One place to check before you spend $20, $50 or $200 on another AI tool.
Save this so you don't have to pay for something you can get for free 👇
5
0
25
762
THIS SPACEXAI DESIGNER SOLD HER MIC ON FACEBOOK MARKETPLACE WITH ONE PROMPT. HER BOT EVEN DROPS THE PRICE $5 EVERY WEEK UNTIL IT SELLS
she sent a photo. the bot checked the official price, found the cheapest listings nearby, asked her writer bot how she writes listings, and h stealing from the same podcast:
→ chief of staff bot: when she doesn't know who should do a task, it goes there
→ articles sent to a bot come back as a podcast she listens to while driving
→ her colleague's assistant bot booked a 4-city work trip with hotels and asked before paying
→ the same bot has its own phone number and texts her to eat lunch
→ a bot that audits your other bots and tells you which routines burn too much usage
how they get bots to work alone:
1. do the task together and watch it
2. correct it until it's right
3. save it as a skill
4. once /skill works in one shot, put it on a routine
1:25 - one prompt, five tasks (the marketplace one)
17:59 - the bot that designs and audits bots
21:33 - booking a multi-city trip
38:59 - how to trust bots with more work
save this before you set up your next bot
she sent a photo. the bot checked the official price, found the cheapest listings nearby, asked her writer bot how she writes listings, and h stealing from the same podcast:
→ chief of staff bot: when she doesn't know who should do a task, it goes there
→ articles sent to a bot come back as a podcast she listens to while driving
→ her colleague's assistant bot booked a 4-city work trip with hotels and asked before paying
→ the same bot has its own phone number and texts her to eat lunch
→ a bot that audits your other bots and tells you which routines burn too much usage
how they get bots to work alone:
1. do the task together and watch it
2. correct it until it's right
3. save it as a skill
4. once /skill works in one shot, put it on a routine
1:25 - one prompt, five tasks (the marketplace one)
17:59 - the bot that designs and audits bots
21:33 - booking a multi-city trip
38:59 - how to trust bots with more work
save this before you set up your next bot
STOP PAYING FOR AI BEFORE CHECKING THIS WEBSITE.
Opus 5.5, Fable 5, GPT 6.1 Sol, GPT-6 Astra and a lot more can already be accessed for free.
The problem is that the offers are scattered everywhere:
> free API credits
> student plans
> signup generations
> open-source alternativesI built -> nopaywall.io
One place to check before you spend $20, $50 or $200 on another AI tool.
Save this so you don't have to pay for something you can get for free 👇
4
0
17
418