Pinned Tweet
this guy used grok bot to quit his 9-5 and start his first business in 3 days
he just leaked the entire system he used in GROK BOT to build it
steal this
grok bot is stupidly cracked at designing hardware.....
i turned mine into a hardware lab. it’s now working on an ESP32 board with:
> ESP32-WROOM-32 and GPIO breakout headers
> USB-C power/data, a USB-UART bridge, and auto-reset
> AMS1117 3.3V power, EN/BOOT buttons, and LEDs
here’s the exact team, in build order:
1. Archie: architecture + antenna
x.ai/bot/To_CQRg_q2PYHBjemNM…
2. Volt: power
x.ai/bot/wvrapeX6Qpeg8Pa3-QR…
3. Pinout: connectors + programming
x.ai/bot/jy4bAXa_Jv5vWAlUVAm…
4. Ohm: passives + electrical rules checks
x.ai/bot/NbT2NbpQCV1uyvh1oad…
5. Trace: routing + design rules checks
x.ai/bot/Z43ZFpOw4aGO_UqeVPe…
6. Gerber: fabrication files + order draft
x.ai/bot/JrPTyHpK_E_BSptXWmm…
7. Parts Buyer: sourcing + cart
x.ai/bot/5J8Ruh6O3pHCbUUv7WE…
steal this.
on monday i wrote about stranded GPUs. idle in one place, unavailable in another.
today @edsgoode and @shadeformai launch Flex VMs: one command to get a bare-metal GPU server onto the marketplace within hours.
more chips isn't the only answer. making existing ones usable is.
🤝 Paid partnership
every “world model” that’s secretly a video model must be sweating...
one model can now create the character, animate it, and build the world around it, in actual 3D.
Viggle’s JST-2 is a full-stack, physics-native world foundation model.
it works with 3D atoms, each with a position, physical attributes and dynamics, so the scene can be more than a stream of pixels.
→ one image: a 3D character with motion built in
→ text or video: motion
→ a photo: a 3D scene you can keep expanding
characters, movement, and worlds from one model.
don’t just make the shot. make the world behind it.
watch JST-2 bring one to life ↓
🤝 Paid partnership
Today we are releasing JST-2, the world's first full-stack physics-native world foundation model.
Many world models learn in pixel/video space. We went back to first principles and built the entire model stack in 3D.
Generate a 3D scene in seconds and expand it in real time. Create characters with motion built in, all through one general architecture.
"H100 rentals went up, not down"
that line broke my brain a little...
an older chip should be getting cheaper to rent every quarter. it isn't.
and it's not because GPUs don't exist. they're STRANDED:
> locked inside long contracts
> spread across 30+ clouds
> sitting dark while another team can't get one machine
that's the part nobody warns you about as a builder... compute stops being an engineering problem and quietly turns into a finance problem
shadeform's bet: one marketplace so you can launch anywhere, and idle GPUs actually get used
@shadeformai
🤝 Paid partnership
The compute market has been a wild ride in 2026.
Prices have skyrocketed, minimum contract durations have tripled, and near term capacity is few and far between.
At the same time, agents are becoming a primary user of compute themselves, and not just GPU compute.
It’s clear that the existing rental models and platforms across the industry no longer fit the reality we see today.
That’s why this week we’re announcing six new features for the Shadeform platform 👇️
everyone's predicting the next token...skyfall dreams the whole object first
SKY-001 is what they're calling the first engineering world model. on CAD and BIM tasks vs GPT-6 Astra and every frontier model they tested:
> 51-125x faster
> 150-210x cheaper
> higher IOU
it simulates the thing before anything gets built. that's how an engineer actually thinks.
18 months of work from the team out of Maluuba's deep learning lab (acquired by Microsoft)
design today. autonomous manufacturing is the bet.
@skyfallai @spisallyouneed
🤝 Paid partnership
Introducing SKY-001, the first Engineering World Model.
On CAD and BIM tasks, it beats GPT-6 Astra and every frontier model we tested:
-> 51-125x faster
- > 150-210x cheaper
-> Higher IOU than all frontier models
SKY-001 lets you dream a physical object before you build it.
18 months of work. Built by the team out of Maluuba's deep learning lab (acq. by @MSRResearch).
The first ‘AI General Engineer’ will be the backbone for Autonomous Manufacturing.
Velo retweeted
Elon Musk's SpaceXAI Grok Bot just got a 6-bot support team that cuts a $95,400-a-year AI bill to $9,272, and this is a f*cking goldmine
100,000 support messages a month run through 6 Grok Bots, and not one of them pays Claude Fable 5.1's $50 per million output tokens
bot 1 → Jev, built by ex-OpenAI researchers at a16z-backed TypeSafe AI, answers 5 questions per message in ~100 ms. the whole month costs $3.15
bot 2 → Order DB handles where is my order. 45,000 messages a month for $0
bot 3 → Templates sends a saved answer only when Jev is 80%+ sure. another 25,000 messages for $0
bot 4 → Kimi K3, Moonshot AI's 2.8 trillion parameter open model, writes the hard 30% at $15 per 1M output vs Fable's $50
bot 5 → Human desk catches chargebacks, lawyer threats and safety issues before any bot says a word
bot 6 → Spam drop kills SEO pitches before they burn a single token
Metaview put Jev into every agent on its platform in one weekend, and one demo sorted 1,018 research papers for 8 cents
the top row of the video is live pricing pages from Kimi, Vercel and Claude. every number is real
save this before you hand $95,400 a year to a premium model for what 6 Grok Bots do for $9,272
Boris Cherny, the creator of Claude Code, just published how he actually uses it...
i turned his workflow into 1 prompt that runs a whole marketing agency. no team, no VA
here's the full setup step by step:
step 1 → paste the prompt from the images into Claude Code
step 2 → it builds a folder for every client with rules, data, drafts and a log
step 3 → 7:00 every morning it checks spend and flags anything off by 10%
step 4 → friday 8:00 it drafts a 3-page report with revenue pulled from Shopify
step 5 → once a month it kills the bottom 50% of ads and writes new tests
the catch: nothing goes to a client and no budget moves without your yes. it gathers, checks and drafts. you decide
week 1 it only drafts. by month 2 it makes the small changes on its own
1 person. 0 employees. the agency runs while you sleep
the full prompt is in the images, the rest of it in the comments ↓
I made a 25 minute animated episode with Claude, no AI video generator.
I gave it the idea for the show and for every episode. Claude directed everything else and wrote the whole animation engine in code. Some props it built in Blender through Blender MCP.
One of the main characters is a Claude Code terminal turned into a robot that keeps asking permission. The inventor hits yes every time without reading. (WHICH IS ME)
It had rendered over 37,000 frames and kept redoing shots on its own until the scenes worked.
The actual prompt is in the first reply.
The first ever CLAUDE MADE ANIME is out!!
It's called Gus & Spark and 25 minute episode. Opus 5.5 MAX worked on it for over 15 hours, and wrote it code by code with 0 AI generated video.
- 34K lines of code
- 986M tokens, about $370 at API prices
I believe Fable 5.5 should be able to produce even higher quality anime!
In the future all 3D animated videos can be created by Claude without any AI generated video.
Velo retweeted
AMD'S LISA SU'S $400 GPU FROM 2020 JUST MASS-KILLED $900/MO OF AI SUBSCRIPTIONS → CURSOR, CHATGPT, PERPLEXITY, GROK → ALL REPLACED FOR $8 A MONTH OF ELECTRICITY
$400 once → $8/mo of power → that's the entire bill → a used rx 6800 xt from 2020 now does the job of four subscriptions from four different companies charging $200-300 a month each
cancelled → replaced with:
cursor ultra ($200/mo) → devstral small 2
12.8gb → built by mistral for multi-file agentic coding → cursor charges $200/mo for the same workflow → mistral gave it away for $0
chatgpt pro ($200/mo) → qwen3 14b
9gb → openai charges $200 for gpt-5.5 → alibaba gave away a model that matches it → for free → forever
perplexity max ($200/mo) → deepseek r1 14b
the model that mass-embarrassed openai on every reasoning benchmark → runs research loops on a $400 card that perplexity charges $200/mo for
supergrok heavy ($300/mo) → phi-4 14b
microsoft open-sourced it → 14b parameters → long docs, analysis, code review → xai charges $300/mo for this → microsoft charges $0
what i was paying:
cursor ultra .............. $200
chatgpt pro ............... $200
perplexity max ............ $200
supergrok heavy ........... $300
total ..................... $900/mo
what i pay now:
used rx 6800 xt ........... $400 once
power ..................... ~$8/mo
nvidia fans paid $500 for a 3090. amd did the same thing for $400
lisa su never held it up on a stage → never gave a keynote about it → the card just quietly works and nobody's talking about it
that's the part that should scare cursor, openai, perplexity and xai the most
save this → models in the comments
haggle bot is the most insanely powerful grok bot spacexai has ever shipped...
in its first week it found $100k+ in wasted spend
i'm turning that idea into an entire department: the Department of Grok Efficiency
here's how to build it, step by step:
Velo retweeted
this guy used grok bot to quit his 9-5 and start his first business in 3 days
he just leaked the entire system he used in GROK BOT to build it
steal this
Velo retweeted
THIS PERSON LITERALLY BUILT An AGENTIC BROWSER USING GROK @bot
>he was complaining about his browser being slow
>so grok @bot made him his own
follow the guide below to build anything with grok @bot
xitter.cf/Av1dlive/status/211041…
Velo retweeted
if you are young and ambitious. I'm begging you to read this.
grok Bot is literally a 24/7 team of agents who will help
you build and launch your first business with a matter of prompts.
if you are young and ambitious. I'm begging you to read this.
grok Bot is literally a 24/7 team of agents who will help
you build and launch your first business with a matter of prompts.






























