predicting the future with code / ai × finance / ex xAI / tg: shmidtqq ads via @CELE8EX · https://xitter.twitterapi.xyz/t.co/wk6O0j61Sn
info ➜
Joined October 2021
- Tweets21.6K
- Following664
- Followers13.2K
- Likes81.5K
A 19-year-old from Miami mass-connected Claude Code to the Meta Ads API and mass-replaced a $4,000/month ad agency with a single terminal command
He doesn't touch Ads Manager. He doesn't design creatives. He doesn't write copy.
He opens Claude Code and types:
"Create 50 ad variations for this product. Hook, body, CTA. Test each at $2/day. Kill anything under 2% CTR after 48 hours. Scale winners to $50/day."
Claude Code writes the entire system from scratch:
- connects to Meta Marketing API
- generates 50 unique ad copies
- pushes them live through the API
- monitors performance every 6 hours
- kills losers, duplicates winners with new audiences
He set it up once. It took 11 minutes.
First client: a local gym owner spending $3,000/month on a media buyer who made 4 creatives per week
He showed him the dashboard: 50 creatives tested in 48 hours. 3 winners found. CPA dropped 41%.
He charges $2,500/month. His ad spend per client: $100 for testing. Claude Code cost: $0.
Second client came from a referral. Then a third. Then seven.
By Day 30 he had 9 clients at $1,500 to $3,500 each
Total monthly revenue: $22,400
Total time spent per client: one terminal command every Monday morning
The craziest part: his clients think he has a team of 5
He has Claude Code and a $20 API token
Agencies charge $4,000/month to do manually what he builds in 11 minutes
Full setup walkthrough is in the article below, bookmark it before every media buyer in your city finds it
shmidt retweeted
YOU'RE OVERPAYING FOR AI BY 1,000X.
xerj reader doesn't write answers.
it points at them.
you ask a question about your file.
it reads it once.
then it highlights the exact span that answers it.
or says "no answer" and stops.
it can't make up a word it never writes.
the only answer it can't give you is a wrong one.
and the math nobody wants to run is f**king brutal:
→ 1,050x cheaper than opus 5.5
→ 26x cheaper than haiku 5.5
→ roughly 4x cheaper than jev
same question. same file. three models.
opus writes you a paragraph.
haiku writes you a shorter paragraph.
xerj writes nothing.
it just underlines the sentence that was already there.
the problem was never the model.
you've been paying for an essay when all you needed was a highlighter.
watch all three race on one document ↓
I've built a new AI model that is not allowed to generate text. Instead, it produces a verifiable program that answers your prompt using the original reference.
XERJ Reader-1 is a neurosymbolic model for working with data. It points to the exact words in your document, writes a small program over them, and exact code computes the answer:
• 20-90x faster
• 10-1,000x cheaper (output tokens free)
• every answer is a program you can replay
The shortest path to automating data work.
shmidt retweeted
Claude Opus 5.5 just replaced my entire motion & 3D pipeline...
these are the 13 skills i actually use with Opus 5.5 to go from a blank prompt to a rendered 3D film, one skill per stage. 265K+ github stars between them.
this is the pipeline, stage by stage:
№1 · motion-design-skill (2.9K ★)
Motion Principles. timing, easing, choreography, disney's animation rules
-> github.com/LottieFiles/motio…
№2 · anything2explainer (3.2K ★)
Script & Storyboard. topic in, narrated explainer video out
-> github.com/Vincentwei1021/an…
№3 · hallmark (53K ★)
Art Direction. design that refuses to look AI-made
-> github.com/Nutlope/hallmark
№4 · blender-skills (2.5K ★)
3D Blockout. 94 blender skills, from blockout to export
-> github.com/arjun988/blender-…
№5 · mcp-for-blender (34K ★)
3D Scene. claude drives blender and builds whole scenes
-> github.com/ahujasid/mcp-for-…
№6 · anatomy (26.6K ★)
3D Breakdowns. interactive isometric cutaways of real machines
-> github.com/wheresryan22/anat…
№7 · teardown (3.1K ★)
3D Simulators. any machine becomes a 3D simulator you can take apart
-> github.com/elvarofficial/tea…
№8 · threejs-skills (4.1K ★)
3D on the Web. claude writes three.js scenes the right way
-> github.com/CloudAI-X/threejs…
№9 · unity-mcp (15.9K ★)
Real-Time 3D. claude runs the unity editor for you
-> github.com/CoplayDev/unity-m…
№10 · after-effects-mcp (883 ★)
Compositing. claude builds comps, text and shapes inside after effects
-> github.com/Dakkshin/after-ef…
№11 · gsap-skills (18.3K ★)
Web Motion. official easing, timelines, scroll
-> github.com/greensock/gsap-sk…
№12 · video-shotcraft (18.6K ★)
Camera & Shots. 124 cinematic shot recipes for product promos
-> github.com/Vincentwei1021/vi…
№13 · hyperframes (82.5K ★)
Final Render. write HTML, get back a finished video
-> github.com/heygen-com/hyperf…
bonus · OpenMontage (79.6K ★)
the whole studio in one repo. 12 pipelines, 100+ tools
-> github.com/OpenMontageInc/Op…
at the end of the line: me, pressing render.
one prompt in. finished film out.
all open-source. save this, your next motion project starts at №1.
Claude Opus 5.5 just replaced my entire motion & 3D pipeline...
these are the 13 skills i actually use with Opus 5.5 to go from a blank prompt to a rendered 3D film, one skill per stage. 265K+ github stars between them.
this is the pipeline, stage by stage:
№1 · motion-design-skill (2.9K ★)
Motion Principles. timing, easing, choreography, disney's animation rules
-> github.com/LottieFiles/motio…
№2 · anything2explainer (3.2K ★)
Script & Storyboard. topic in, narrated explainer video out
-> github.com/Vincentwei1021/an…
№3 · hallmark (53K ★)
Art Direction. design that refuses to look AI-made
-> github.com/Nutlope/hallmark
№4 · blender-skills (2.5K ★)
3D Blockout. 94 blender skills, from blockout to export
-> github.com/arjun988/blender-…
№5 · mcp-for-blender (34K ★)
3D Scene. claude drives blender and builds whole scenes
-> github.com/ahujasid/mcp-for-…
№6 · anatomy (26.6K ★)
3D Breakdowns. interactive isometric cutaways of real machines
-> github.com/wheresryan22/anat…
№7 · teardown (3.1K ★)
3D Simulators. any machine becomes a 3D simulator you can take apart
-> github.com/elvarofficial/tea…
№8 · threejs-skills (4.1K ★)
3D on the Web. claude writes three.js scenes the right way
-> github.com/CloudAI-X/threejs…
№9 · unity-mcp (15.9K ★)
Real-Time 3D. claude runs the unity editor for you
-> github.com/CoplayDev/unity-m…
№10 · after-effects-mcp (883 ★)
Compositing. claude builds comps, text and shapes inside after effects
-> github.com/Dakkshin/after-ef…
№11 · gsap-skills (18.3K ★)
Web Motion. official easing, timelines, scroll
-> github.com/greensock/gsap-sk…
№12 · video-shotcraft (18.6K ★)
Camera & Shots. 124 cinematic shot recipes for product promos
-> github.com/Vincentwei1021/vi…
№13 · hyperframes (82.5K ★)
Final Render. write HTML, get back a finished video
-> github.com/heygen-com/hyperf…
bonus · OpenMontage (79.6K ★)
the whole studio in one repo. 12 pipelines, 100+ tools
-> github.com/OpenMontageInc/Op…
at the end of the line: me, pressing render.
one prompt in. finished film out.
all open-source. save this, your next motion project starts at №1.
YOU'RE OVERPAYING FOR AI BY 1,000X.
xerj reader doesn't write answers.
it points at them.
you ask a question about your file.
it reads it once.
then it highlights the exact span that answers it.
or says "no answer" and stops.
it can't make up a word it never writes.
the only answer it can't give you is a wrong one.
and the math nobody wants to run is f**king brutal:
→ 1,050x cheaper than opus 5.5
→ 26x cheaper than haiku 5.5
→ roughly 4x cheaper than jev
same question. same file. three models.
opus writes you a paragraph.
haiku writes you a shorter paragraph.
xerj writes nothing.
it just underlines the sentence that was already there.
the problem was never the model.
you've been paying for an essay when all you needed was a highlighter.
watch all three race on one document ↓
I've built a new AI model that is not allowed to generate text. Instead, it produces a verifiable program that answers your prompt using the original reference.
XERJ Reader-1 is a neurosymbolic model for working with data. It points to the exact words in your document, writes a small program over them, and exact code computes the answer:
• 20-90x faster
• 10-1,000x cheaper (output tokens free)
• every answer is a program you can replay
The shortest path to automating data work.
shmidt retweeted
holy f*ck. switching between ai models cost this guy $100,000 in 21 months...
his card said $658/mo for ai. the real number was $4,794. 47 times a day he stopped to pick a model. 3.2 min each. 55 hours a month. 86% of that $100k never showed up on a statement.
here's the alpha he used to stop the bleeding:
/1. statement test:
next to every ai sub, write the last real thing you used it for. can't fill it in → cancel.
/2. sticky note test:
tomorrow, one tally mark every time you hesitate between models. you'll be shocked by lunch.
/3. gear rule:
shift up only when a better answer is worth more than the heavier gear.
→ summaries, tables, pdf figures: !fast
→ emails, first drafts: default
→ contracts, prod bugs, research you bet on: !deep
he runs it all through one ai employee, viktor in slack.
[4. checkup]
once a month ask: "where are my credits going?"
his answer: 12% burned on slack summaries that should've been !fast.
$658 → $263 on the card. 55h → 3.5h of choosing. same output.
the subscriptions were never the thief. the choosing was.
save this, full audit in the video ↓
holy f*ck. switching between ai models cost this guy $100,000 in 21 months...
his card said $658/mo for ai. the real number was $4,794. 47 times a day he stopped to pick a model. 3.2 min each. 55 hours a month. 86% of that $100k never showed up on a statement.
here's the alpha he used to stop the bleeding:
/1. statement test:
next to every ai sub, write the last real thing you used it for. can't fill it in → cancel.
/2. sticky note test:
tomorrow, one tally mark every time you hesitate between models. you'll be shocked by lunch.
/3. gear rule:
shift up only when a better answer is worth more than the heavier gear.
→ summaries, tables, pdf figures: !fast
→ emails, first drafts: default
→ contracts, prod bugs, research you bet on: !deep
he runs it all through one ai employee, viktor in slack.
[4. checkup]
once a month ask: "where are my credits going?"
his answer: 12% burned on slack summaries that should've been !fast.
$658 → $263 on the card. 55h → 3.5h of choosing. same output.
the subscriptions were never the thief. the choosing was.
save this, full audit in the video ↓
shmidt retweeted
In 2015, I built an AI that detects early cancer.
Today, we’re introducing ARC by Rayline: the world’s first unified model for routing AI.
> 97.5% of Opus 5.5 quality
> Up to 1/10th of the cost
> Fully open-source
It picks the right model and the right amount of thinking for every step of your agent’s work.
Try it here and read our technical report below:
rayline.ai
THE SCARIEST AI BUG ISN'T THE ONE THAT CRASHES.
IT'S THE ONE THAT HAPPENED ONCE AND NEVER AGAIN.
A customer sends a screenshot. Your agent did something unhinged.
You replay the chat. Perfect.
You run it 20 more times. Perfect every time.
Because the world it broke in is gone.
The order moved. The record changed. That exact moment is over.
What a ghost bug costs you:
> days of an engineer reading logs
> the customer who took the screenshot
> everyone they showed it to
> Plurai Simulation: free
@pluraiAI just launched it, and the replay is the part that got me. Pull the production trace, rebuild the whole world as it was at that step, and replay from the exact moment it went wrong.
Then Plurai traces it to the line of code, writes the fix and re-runs it 3 times. Inside Claude Code, Codex or Copilot.
One screenshot + a rebuilt world + a replay = the ghost, caught on camera.
🤝 Paid partnership
Big day for us at @pluraiAI ! Plurai Simulation is live and free for every AI engineer.
AI agents look great in demos. Until real users break them.
So we built a world where your agent can fail, learn, and get fixed before your users ever see it.
→ Digital twin: Your real agent, its tools, and environment recreated from your code
→ Simulated users & tasks: Realistic users, tasks, and workflows that push your agent into the edge cases that matter
→ Auto evals: Check what your agent actually did, not just what it said
→ Verified fixes: Every failure comes with the run that exposed it and a fix you can verify
→ Replays: Turn any production failure into a simulation and replay it against every new version of your agent
All inside the coding agent you already use: @claudeai code, @OpenAICodexCli , or @GitHubCopilot .
Start simulating for free: simulation.plurai.ai/connect
shmidt retweeted
A 22-year-old college dropout typed 3 messages to Claude and closed $8,740 in 14 days
He had no portfolio, no followers, no website
He typed one message:
"I can write and I understand short-form video. Give me a service I can sell in 48 hours for $500-$1,500/month. No audience needed. Ask me questions until you're 95% confident in the offer"
Claude asked him 9 questions back. Picked one niche:
done-for-you short-form video scripts for small business owners on Instagram
Second prompt:
"Find the single most successful person doing this exact thing. Pretend you are them. Write me a 14-day plan that lands my first paying client by tomorrow"
Claude searched the web. Built his entire career on Instagram DM outreach and retainer pricing.
Then it handed him the whole playbook: who to DM, the exact first message, how to price the ladder from $500 to $1,500, when to follow up
Third prompt:
"Now be my business coach. Ask me one question at a time until I've done everything on the list"
Claude stopped giving answers. It started holding him accountable.
Day 1: first $500 client from a cold DM
Day 7: raised to $800 after Claude rewrote his pitch
Day 14: 11 clients. $8,740 collected. Zero ad spend.
Total cost: $20 Claude subscription + his thumbs
The craziest part: the prompt that built the entire business is 3 sentences long
Full prompt in the article below, save it before your competition finds it
A 22-year-old college dropout typed 3 messages to Claude and closed $8,740 in 14 days
He had no portfolio, no followers, no website
He typed one message:
"I can write and I understand short-form video. Give me a service I can sell in 48 hours for $500-$1,500/month. No audience needed. Ask me questions until you're 95% confident in the offer"
Claude asked him 9 questions back. Picked one niche:
done-for-you short-form video scripts for small business owners on Instagram
Second prompt:
"Find the single most successful person doing this exact thing. Pretend you are them. Write me a 14-day plan that lands my first paying client by tomorrow"
Claude searched the web. Built his entire career on Instagram DM outreach and retainer pricing.
Then it handed him the whole playbook: who to DM, the exact first message, how to price the ladder from $500 to $1,500, when to follow up
Third prompt:
"Now be my business coach. Ask me one question at a time until I've done everything on the list"
Claude stopped giving answers. It started holding him accountable.
Day 1: first $500 client from a cold DM
Day 7: raised to $800 after Claude rewrote his pitch
Day 14: 11 clients. $8,740 collected. Zero ad spend.
Total cost: $20 Claude subscription + his thumbs
The craziest part: the prompt that built the entire business is 3 sentences long
Full prompt in the article below, save it before your competition finds it
Claude Opus 5.5 just built me a 12-floor studio HQ...
these are the 12 skills i actually use with Opus 5.5 to run an entire creative agency, every department staffed by one skill. 965K+ github stars between them.
this is who moved in, floor by floor:
#01 · motion-design (2.7K ★)
Animation Director. lottie animations from a text prompt
github.com/LottieFiles/motio…
#02 · remotion skills (5.5K ★)
React Video Engineer. the official one, videos written in react
github.com/remotion-dev/skil…
#03 · unity-mcp (15K ★)
Game Developer. claude runs the unity editor for you
github.com/CoplayDev/unity-m…
#04 · figma-context mcp (16K ★)
Design Handoff. claude reads your figma file before writing a line
github.com/GLips/Figma-Conte…
#05 · gsap-skills (17K ★)
Motion Engineer. official easing, timelines, scroll
github.com/greensock/gsap-sk…
#06 · blender-mcp (33K ★)
3D Artist. claude drives blender, builds entire scenes
github.com/ahujasid/blender-…
#07 · playwright-mcp (39K ★)
QA Tester. tests the thing claude just built
github.com/microsoft/playwri…
#08 · hyperframes (81K ★)
Video Editor. write HTML, get back a finished video
github.com/heygen-com/hyperf…
#09 · impeccable (107K ★)
Design Critic. /polish /bolder /critique, murders AI slop
github.com/pbakaus/impeccabl…
#10 · ui-ux-pro-max (146K ★)
Product Designer. thinks like a senior at a top agency
github.com/nextlevelbuilder/…
#11 · anthropic skills (187K ★)
Creative Director. frontend-design, canvas, algorithmic art
github.com/anthropics/skills
#12 · superpowers (314K ★)
Chief of Staff. 100+ skills, keeps the whole studio running
github.com/obra/superpowers
on the rooftop: me. the only human in the building.
salary $0. hours 24/7. sick days 0. headcount, still 1.
all open-source. the tower in the video is mine, built in three.js, every floor is one real hire.
BARRY GOT RECAST.
Same shot. Same camera. Same voice.
He has not been told yet.
Sound on ↓
🤝 Paid partnership
16
83
8,330
And characters are just the start.
Product swap and location swap are coming next. Same mechanic.
Read that again.
One good take. Change who's in it. Then what's in it. Then where it is.
One shot + one picture = a new cast without a new shoot.
1
2
76
Try Recast in ElevenCreative: try.elevenlabs.io/shmidt
Barry is currently between roles.
Who should replace him next? Reply below.
@elevenlabsio #ElevenCreativePartner
2
70
shmidt retweeted
A 24-year-old girl from Austin asked Claude one prompt and made $11,340 in 20 days
She was broke, laid off, no audience, no savings
She typed one message to Claude:
"Give me 2 ways to make $1,000 in 30 days using skills I already have, no audience, sellable in 48 hours. Ask me questions until you are 95% confident"
Claude asked her 12 questions back, picked one niche:
writing cold-email scripts for small B2B SaaS
Second prompt:
"Pick the single most successful person in this niche, pretend you are them, write a 20-day plan under $1000 that gets me a paying customer by Day 1"
Claude handed her the whole thing: who to DM, the exact script, the price ladder, follow-up timing
Day 1 she closed her first $200 client
By Day 20 she had 18 clients at $500 to $1,200 each
Total spent: $84
The craziest part: the prompt that cracked it is 3 sentences long
Full prompt in the article below, copy it before everyone has it
A 24-year-old girl from Austin asked Claude one prompt and made $11,340 in 20 days
She was broke, laid off, no audience, no savings
She typed one message to Claude:
"Give me 2 ways to make $1,000 in 30 days using skills I already have, no audience, sellable in 48 hours. Ask me questions until you are 95% confident"
Claude asked her 12 questions back, picked one niche:
writing cold-email scripts for small B2B SaaS
Second prompt:
"Pick the single most successful person in this niche, pretend you are them, write a 20-day plan under $1000 that gets me a paying customer by Day 1"
Claude handed her the whole thing: who to DM, the exact script, the price ladder, follow-up timing
Day 1 she closed her first $200 client
By Day 20 she had 18 clients at $500 to $1,200 each
Total spent: $84
The craziest part: the prompt that cracked it is 3 sentences long
Full prompt in the article below, copy it before everyone has it