THE GROK BOT TEAM ONBOARDED THEIR FIRST 300 USERS BY HAND. THIS IS THE 20-MINUTE SETUP THEY WALKED EVERY ONE OF THEM THROUGH.
for about two weeks the core team sat on calls with every new user
"sit on a call for 20 minutes when the computer isn't spinning up"
every rough call turned into a fix: "that can never happen again. we need to solve this tomorrow"
what roman tells every new user to do first, as a 20-minute setup:
→ minute 0-5: connect slack or messages, email and the files you use every day
→ minute 5-10: tell it your role and what you care about, like onboarding a new hire
→ minute 10-15: ask it to go through everything and suggest 5 things it can take off your plate
→ minute 15-20: keep the best 2, one bot per job
the prompt for minute 10:
"go through my email, messages and files. suggest 5 things you could take off my plate, and what access you'd need for each."
minute 0 is where you hand over your logins. lock them first, article below ↓
2
26
399
GROK BOT'S PRODUCT LEAD SAYS YOU SHOULD ALMOST NEVER TAKE OVER YOUR BOT'S COMPUTER. THE 4-LINE BRIEF THAT MEANS I DON'T HAVE TO.
his comparison is a coworker. how often do you grab their mouse? "hopefully it's pretty close to zero"
"you should never be clicking into a remote virtual machine"
when i had to take over, it was always my fault: the bot didn't know the goal, the login, or when to stop
so every job now starts with 4 lines:
1. goal: what done looks like
2. access: which logins it can use, and only those
3. stop: when to pause and ask me
4. proof: a screenshot of the result
the prompt:
"goal: [done looks like]. access: [logins]. stop and ask me if [condition]. when you're done, send me a screenshot of the result."
i haven't grabbed the mouse in 3 weeks. how i keep those logins safe, article below ↓
GROK BOT'S PRODUCT LEAD STOPPED OPENING NEW CHATS. HE GIVES EACH BOT ONE ROLE AND IT GETS SMARTER EVERY WEEK. MY 6 ROLES.
his problem with chats: "I find myself copying and pasting between chats all the time"
his fix is roles, like on a real team. "these are long lived agents. these are not individual one-off sessions"
a role keeps the memory. a chat starts from zero every time
my 6:
1. money: bills, refunds, subscriptions
2. inbox: triage and drafts
3. calendar: scheduling and reminders
4. research: anything i'd google for more than 10 minutes
5. home: deliveries, repairs, appointments
6. security: logins, alerts, leaked passwords
the prompt to set one up:
"you are my [role] bot from now on. keep notes on everything we do together and use them next time. at the end of each week, tell me one thing you learned about how i like it done."
the role i'd set up first is security. how mine works, article below ↓
2
22
385
SPACEXAI'S SALES TEAM HAD A TASK FAILING FOR 7 DAYS UNTIL GROK BOT GOT ITS OWN COMPUTER. 5 JOBS WITH NO API I HAND MINE NOW.
the tools their sales team uses "do not have well-supported MCPs or APIs". other ai tools got stuck halfway
grok bot worked like a new hire: "they gave it a laptop and it could just run"
the fix was mouse control and seeing the pixels on screen. next day the team heard that "this workflow that was failing the last seven days finally works"
jobs with no api i give my bot:
1. pulls invoices from 4 supplier portals that only have a login page
2. updates my listings by clicking through a marketplace form
3. downloads my bank statements every month and renames them
4. books appointments on a clinic site with a calendar widget
5. copies new leads from a crm screen into my sheet
the prompt:
"use your own computer like i would. log in, click through and do [task] every [day]. if a page changes or you get stuck, send me a screenshot and ask."
a bot that logs in for you needs those logins locked. how i set that up, article below ↓
1
24
596
SPACEXAI'S RECRUITING BOT READS NEW AI PAPERS EVERY MORNING, FINDS THE AUTHORS AND ASKS FOR AN INTRO. NOBODY TOUCHES IT.
the setup, in his words:
"I want you every morning to go to the conference website, download the PDFs"
then the bot:
→ finds every new author name they haven't tracked
→ adds it to a spreadsheet
→ researches the person
→ checks who at SpaceXAI is connected to them
→ sends that person a Slack message asking for an intro
"now it's the type of thing AI is superhuman at"
the same loop works for anything: every morning, one source → new names → sheet → one action.
the prompt to start yours:
"every morning at 8, check [source] for new [people/companies]. add new ones to my sheet with one line on why they matter. if someone in my contacts knows them, draft an intro request for me to approve."
a bot that sends messages for you needs locked accounts. how i set that up, article below ↓
2
28
865
ELON MUSK SAYS BY THE END OF THIS YEAR AI WILL DO ANYTHING A HUMAN AT A COMPUTER CAN DO. THEN COMES WHAT HE CALLS "THE INFINITE MONEY GLITCH"
his words: "I'd be surprised by the end of this year if digital human emulation has not been solved."
what that means in plain terms:
→ anything a person does on a computer, AI does
→ he calls it a "digital Optimus"
→ that's the ceiling until robots arrive
then the robots:
"I call Optimus the infinite money glitch"
robots get better on 3 curves at once: AI, chips, hands
"then the robot can start making the robot"
"this is supernova"
if AI can do anything at a computer, it can also log into yours. how i lock mine down, article below ↓
24
471
ELON MUSK JUST ADMITTED GROK IS WORSE THAN CLAUDE OPUS 5.5. THEN HE GAVE A DATE.
"it's not as good as, say, Opus 5.5 that just got released."
catch up to the frontier "sometime next year, most likely."
then he split the market himself:
anthropic made AI great at software engineering. nobody has done it yet for real-world engineering. grok trains on spacex and tesla data to take that spot.
what i do with that today:
1. code goes to claude
2. jobs that live inside x and my phone go to grok bot
3. any new task: same prompt to both, keep the winner for a week
4. retest every new model release, the order changes every 1-2 months
no loyalty to a model. use whichever wins on your task.
grok bot is "growing about 100% a month". mine guards my logins, setup in the article below ↓
ELON MUSK SAYS AI AGENTS ALREADY USE THE INTERNET MORE THAN HUMANS DO. IN 5 YEARS: 1,000X MORE
ron baron brought it up first. starlink was planned at 20,000 satellites. now it's 100,000.
the reason is bots, not people.
elon on stage:
1. agent traffic already passed human traffic, going by cloudflare's numbers
2. cloudflare's forecast: 1,000x human traffic in 5 years
3. his own take: "I think that is underestimating"
20,000 → 100,000 satellites → most of the traffic isn't human
more bots online than people means more bots trying your logins. how i lock mine down, article below ↓
3
20
533
7 MONTHS BEFORE HE RESIGNED, JENSEN HUANG PROVED AI CREATES JOBS. HIS EXAMPLE: RADIOLOGISTS.
he said: "it turns out they were absolutely right. a hundred percent right."
why it matters: AI really does read scans better than doctors now.
he said: "however, radiologists demand has gone up."
why it matters: cheaper scans, so hospitals run way more of them.
he said: "a job has a purpose and a task."
why it matters: AI took the task. the purpose stayed with people.
he said: "I see nobody saying, you know what? I was wrong. I'm sorry."
AI won't take my job. it does guard my accounts. how i set it up, article below ↓
8
20
378
ELON MUSK: NOT EVEN BILLIONAIRES WILL CONTROL AI. "I THINK THEY'LL CONTROL THEMSELVES."
he was asked how to stop AI money from piling up with a few rich people.
his answer skipped the rich people completely.
"I don't think even so-called wealthy people will ultimately control the AIs."
then he explained his own money.
going from $1B to $10B doesn't make you 10x happier.
his trillion is mostly shares he needs to stay in charge of spacex and tesla.
"there's not some magical way to avoid taxes. If there were, I would know about it."
mine already controls one thing: my account security. article below ↓
7
24
297
ELON MUSK: COMPANIES RUN PURELY BY AI AND ROBOTS WILL "VASTLY OUTPERFORM" ANY COMPANY WITH HUMANS IN THE LOOP
His proof: "computer" used to be a human job. Whole skyscrapers of people doing math. Now it's one laptop.
He says up front it's going to sound doomerish
"Computer used to be a job that humans had."
20 to 30 floors of people, doing calculations by hand
Then one laptop with a spreadsheet did more than the whole building
His test: imagine some of the cells in your spreadsheet were still calculated by humans. It would be worse, not better
Same thing for companies, he says. Pure AI and robotics beat anything with people in the loop
His last words in the clip: "this will happen very quickly."
Want one working for you? It's all in the article below
8
22
289
Elon Musk says AI will add $30 trillion a year to the world economy. Then he called it "not very much."
What comes after that: money stops mattering.
His math goes in two steps.
Step one is digital superintelligence. A 20 to 30% bigger world economy.
"It's like $30 trillion a year incremental. It's not very much really. It's not transformative."
Step two is robots plus space. A mass driver on the moon. An economy he puts at a million times the size of Earth's.
At that size, he says, giving every human whatever they want is trivial.
And the dollar? Replaced. Currency gets measured in power and mass. Wattage and tonnage.
AI → +30% → robots → x1,000,000 → money becomes irrelevant.
112 seconds of Elon walking through it, number by number.
Save this one and watch it tonight ↓
5
23
347


