Josh The Monkey (@joshzemon) header
Josh The Monkey (@joshzemon) profile picture

I'm just a Monkey doing research.

Joined May 2023
Photo by Josh The Monkey (@joshzemon)Photo by Josh The Monkey (@joshzemon)Photo by Josh The Monkey (@joshzemon)Photo by Josh The Monkey (@joshzemon)Photo by Josh The Monkey (@joshzemon)Photo by Josh The Monkey (@joshzemon)Photo by Josh The Monkey (@joshzemon)Photo by Josh The Monkey (@joshzemon)Photo by Josh The Monkey (@joshzemon)Photo by Josh The Monkey (@joshzemon)
This is f*cking insane. Andrej Karpathy at Sequoia, on what comes after vibe coding: "You can outsource your thinking You can't outsource your understanding" In 30 minutes, he lays out the serious layer above vibe coding: LLMs as ghosts, the app that shouldn't exist and 12 lessons on agentic engineering Vibe Coding → Agentic Engineering → LLMs as Ghosts → Systems You Understand Bookmark and watch it today Then read how to build that layer yourself: Harness, Loops, Graphs ↓
Your agent is not autonomous You are its loop You check every step because nothing else in the system can I broke down the 7 levels from babysitting agents to running a fleet ↓
4
5
1
56
3,540
This is f*cking insane. Andrej Karpathy, the man who co-founded OpenAI and led AI at Tesla: "Reinforcement learning is terrible It just so happens that everything that we had before is much worse" In 7 minutes, he explains why LLMs still can't learn like humans, and why training them is like "sucking supervision bits through a straw" Human Learning → Reinforcement Learning → One Reward → Every Token Upweighted The models are smart. The way they learn is still broken Bookmark and watch it today Then read how to build around those limits: Harness, Loops, Graphs ↓
Your agent is not autonomous You are its loop You check every step because nothing else in the system can I broke down the 7 levels from babysitting agents to running a fleet ↓
7
50
9
1,051
62,796
> be Andrej Karpathy > born in Bratislava, Slovakia > move to Toronto at 15 > take Geoffrey Hinton's class at the University of Toronto > PhD at Stanford under Fei-Fei Li > create CS231n, the course half of AI learned from > 2015: become a founding member of OpenAI > 2017: Elon pulls him to Tesla to lead Autopilot AI > Elon to the team: "the #2 computer vision mind on Earth, after Ilya" > 2025: coin the term "vibe coding" > now sit down for an hour on where engineering goes next he built Tesla's self-driving brain and helped start OpenAI. give him one hour, then read how to build what he's describing ↓
Your agent is not autonomous You are its loop You check every step because nothing else in the system can I broke down the 7 levels from babysitting agents to running a fleet ↓
6
1
110
7,657
This is f*cking insane. Ilya Sutskever, the man who co-founded OpenAI, filmed while they were building GPT-4: "The beliefs and desires of the first AGIs will be extremely important So it's important to program them correctly" In 11 minutes, he explains what he thinks AGI will do to the world, and why he takes it this seriously GPT → GPT-4 → AGI → Superintelligence → The Biggest Decision in History This was recorded years ago. Most of what he described is already happening Bookmark and watch it today Then read how to start building with it: Harness, Loops, Graphs ↓
Your agent is not autonomous You are its loop You check every step because nothing else in the system can I broke down the 7 levels from babysitting agents to running a fleet ↓
16
167
11
1,373
123,899
This is f*cking insane. Ilya Sutskever, the man who co-founded OpenAI, described the future he doesn't like: "Maybe every person will have an AI that does their bidding. It earns money for them, advocates for their needs, then writes a little report: here's what I've done The person says great, keep it up. But the person is no longer a participant" In this 1.5-hour interview, he explains why almost nobody can feel what's coming until it's already here AI → Agents → AIs That Earn For You → Humans Who Are No Longer Participants The only way to stay in the game is to be the one running the agents Worth more than any $2,000 AI course Bookmark and watch it today Then read how to run them yourself: Harness, Loops, Graphs ↓
991
This is f*cking insane. Ilya Sutskever, the man who co-founded OpenAI, described the future he doesn't like: "Maybe every person will have an AI that does their bidding. It earns money for them, advocates for their needs, then writes a little report: here's what I've done The person says great, keep it up. But the person is no longer a participant" In this 1.5-hour interview, he explains why almost nobody can feel what's coming until it's already here AI → Agents → AIs That Earn For You → Humans Who Are No Longer Participants The only way to stay in the game is to be the one running the agents Worth more than any $2,000 AI course Bookmark and watch it today Then read how to run them yourself: Harness, Loops, Graphs ↓
Your agent is not autonomous You are its loop You check every step because nothing else in the system can I broke down the 7 levels from babysitting agents to running a fleet ↓
5
17
2
209
19,934
This is f*cking insane. Andrej Karpathy spent 8 years building AI at OpenAI and Tesla Then he packed everything he knows into one free 2-hour lecture Agents → Loops → Graphs → Self-Improving Systems People pay $20K for bootcamps that teach less than half of this You probably don't have 2 hours right now Don't let it disappear from your feed Bookmark it and watch it this weekend Then read the guide below and build your first harness ↓
Your agent is not autonomous You are its loop You check every step because nothing else in the system can I broke down the 7 levels from babysitting agents to running a fleet ↓
4
6
61
1,948
This is f*cking insane. Andrej Karpathy, the man Elon called the #2 computer vision mind on the planet: "The people who don't use LLMs are already losing And the ones who think signing up for Claude counts as using it are losing too" In 4 minutes, he explains why smart people are still going to fall behind Asking AI → Typing Prompts → Building Something → A System That Runs Without You Typing a prompt and reading the answer isn't the skill. Anyone can do that Worth more than most $1,000 AI courses Bookmark and watch it today Then read how to build the part that runs without you: Harness, Loops, Graphs ↓
Your agent is not autonomous You are its loop You check every step because nothing else in the system can I broke down the 7 levels from babysitting agents to running a fleet ↓
5
27
368
27,468
This is f*cking insane. Andrej Karpathy, the man Elon called the #2 computer vision mind on the planet: "Most people still treat LLMs like chatbots 5 shifts are turning them into systems that finish the work on their own" In 42 minutes, he walks through all five, one at a time Memory → Vision → Reasoning → Search → Tools → Agentic Systems Each shift alone is a feature. All five together replace a workflow Worth more than most $1,000 AI courses Bookmark and watch it today Then read how the system part is built: Harness, Loops, Graphs ↓
Your agent is not autonomous You are its loop You check every step because nothing else in the system can I broke down the 7 levels from babysitting agents to running a fleet ↓
3
72
3,112
This is f*cking insane. Anthropic just paid millions to hire Andrej Karpathy, the man Elon called the #2 computer vision mind on Earth The same week, he gave away what he knows for $0: "Vibe coding is dead. Software 3.0 is here. You don't write the program anymore, you build the system that writes it" In 29 minutes, he explains why LLMs are ghosts, not animals, and what that changes about how you build with Fable 5.2 Software 1.0 → Neural Nets → Prompts → Agents → Software 3.0 OpenAI co-founder. Former head of AI at Tesla. No paywall, just a link Worth more than any $2,000 AI course this year Bookmark and watch it today Then read how to build that software factory step by step: Harness, Loops, Graphs ↓
Your agent is not autonomous You are its loop You check every step because nothing else in the system can I broke down the 7 levels from babysitting agents to running a fleet ↓
10
64
7
552
76,034
This is f*cking insane. Andrej Karpathy just told you to stop learning most of what AI twitter sells: "90% of what AI twitter tells you to learn will be dead in 6 months The frameworks change every quarter. The thing underneath them does not" In 39 minutes, he goes through what senior AI engineers already stopped wasting time on, and what they kept Hype Frameworks → Dead Demos → The Model → The Harness → Skills That Survive Every hour you spend on a dying tool is an hour you didn't spend on the harness Worth more than most $1,000 AI engineering courses Bookmark and watch it today Then read the part that survives: Harness, Loops, Graphs ↓
1
166
This is f*cking insane. Andrej Karpathy just told you to stop learning most of what AI twitter sells: "90% of what AI twitter tells you to learn will be dead in 6 months The frameworks change every quarter. The thing underneath them does not" In 39 minutes, he goes through what senior AI engineers already stopped wasting time on, and what they kept Hype Frameworks → Dead Demos → The Model → The Harness → Skills That Survive Every hour you spend on a dying tool is an hour you didn't spend on the harness Worth more than most $1,000 AI engineering courses Bookmark and watch it today Then read the part that survives: Harness, Loops, Graphs ↓
Your agent is not autonomous You are its loop You check every step because nothing else in the system can I broke down the 7 levels from babysitting agents to running a fleet ↓
2
1
2
49
1,644
Josh The Monkey retweeted
This is f*cking insane. Andrej Karpathy just named the biggest mistake in AI right now: "People are forcing agents to work before they master the harness You building agents right now are at the forefront. Not OpenAI. Not DeepMind. You" In under 6 minutes, he explains why demos are easy and products take a decade Model → Harness → Agents → Loops → Products That Actually Ship The agent was never the product. The foundation under it is Worth more than most $1,000 agent courses Bookmark and watch it today Then read what that foundation looks like: Harness, Loops, Graphs ↓
Your agent is not autonomous You are its loop You check every step because nothing else in the system can I broke down the 7 levels from babysitting agents to running a fleet ↓
7
59
4,007
This is f*cking insane. Andrej Karpathy just named the biggest mistake in AI right now: "People are forcing agents to work before they master the harness You building agents right now are at the forefront. Not OpenAI. Not DeepMind. You" In under 6 minutes, he explains why demos are easy and products take a decade Model → Harness → Agents → Loops → Products That Actually Ship The agent was never the product. The foundation under it is Worth more than most $1,000 agent courses Bookmark and watch it today Then read what that foundation looks like: Harness, Loops, Graphs ↓
Your agent is not autonomous You are its loop You check every step because nothing else in the system can I broke down the 7 levels from babysitting agents to running a fleet ↓
7
59
4,007
Josh The Monkey retweeted
This is f*cking insane. Anthropic just paid millions to hire Andrej Karpathy, the man Elon called the #2 computer vision mind on Earth The same week, he gave away what he knows for $0: "Vibe coding is dead. Software 3.0 is here. You don't write the program anymore, you build the system that writes it" In 29 minutes, he explains why LLMs are ghosts, not animals, and what that changes about how you build with Fable 5.2 Software 1.0 → Neural Nets → Prompts → Agents → Software 3.0 OpenAI co-founder. Former head of AI at Tesla. No paywall, just a link Worth more than any $2,000 AI course this year Bookmark and watch it today Then read how to build that software factory step by step: Harness, Loops, Graphs ↓
Your agent is not autonomous You are its loop You check every step because nothing else in the system can I broke down the 7 levels from babysitting agents to running a fleet ↓
10
64
7
552
76,034
This is f*cking insane. Elon Musk emailed his team: "Andrej is the #2 computer vision mind on Earth, after Ilya Sutskever" At 15, Karpathy left Slovakia for Toronto. He studied under Geoffrey Hinton, did his PhD under Fei-Fei Li, and spent 17 years teaching machines to see In 2012 he wrote that computer vision was "really, really far away" Slovakia → Toronto → Hinton → Stanford → OpenAI → Tesla → The Man Everyone Learns From 8 minutes on how one person got there Bookmark it Then read the "Harness, Loops, Graphs" guide below ↓
Your agent is not autonomous You are its loop You check every step because nothing else in the system can I broke down the 7 levels from babysitting agents to running a fleet ↓
6
85
5
1,069
86,284
This is f*cking insane. Anthropic just paid millions to hire Andrej Karpathy, the man Elon called the #2 computer vision mind on Earth The same week, he gave away what he knows for $0: "Vibe coding is dead. Software 3.0 is here. You don't write the program anymore, you build the system that writes it" In 29 minutes, he explains why LLMs are ghosts, not animals, and what that changes about how you build with Fable 5.2 Software 1.0 → Neural Nets → Prompts → Agents → Software 3.0 OpenAI co-founder. Former head of AI at Tesla. No paywall, just a link Worth more than any $2,000 AI course this year Bookmark and watch it today Then read how to build that software factory step by step: Harness, Loops, Graphs ↓
33
This is f*cking insane. Andrej Karpathy, the man Elon called the #2 computer vision mind on Earth: "Agents aren't magic. They're distillation at scale 99% of what your model knows is dead weight. A small model with the right tools and a closed loop beats it" In 16 minutes, he lays out the stack that actually works, and why the people at the top stopped paying for the biggest model Big Model → Small Model → Right Tools → Closed Loop → Self-Improving Agent Most people keep upgrading the model. The loop is what makes a small one dangerous Worth more than any $1,000 AI course this year Bookmark and watch it today Then read how to build that closed loop step by step in Harness, Loops, Graphs ↓
Your agent is not autonomous You are its loop You check every step because nothing else in the system can I broke down the 7 levels from babysitting agents to running a fleet ↓
6
56
3
398
62,123
This is f*cking insane. Anthropic just paid millions to hire Andrej Karpathy, the man Elon called the #2 computer vision mind on Earth The same week, he gave away what he knows for $0: "Vibe coding is dead. Software 3.0 is here. You don't write the program anymore, you build the system that writes it" In 29 minutes, he explains why LLMs are ghosts, not animals, and what that changes about how you build with Fable 5.2 Software 1.0 → Neural Nets → Prompts → Agents → Software 3.0 OpenAI co-founder. Former head of AI at Tesla. No paywall, just a link Worth more than any $2,000 AI course this year Bookmark and watch it today Then read how to build that software factory step by step: Harness, Loops, Graphs ↓
117
This is f*cking insane. OpenAI co-founder Ilya Sutskever went back to the University of Toronto and told the graduates: "The day will come when AI does everything we can do. Not some of it. All of it Anything I can learn, anything any of you can learn, AI will learn too" In this 9-minute speech, he explains why your brain is a biological computer and why a digital one catches up Neural Nets → AlexNet → ChatGPT → Superintelligence → AI That Can Do Every Job "You may not take interest in AI, but AI will take interest in you" Worth more than any $5,000 AI course Bookmark it and watch it tonight Then read what this looks like on a desk today: 6 Grok bots, 200+ agents, a smaller bill ↓
Your Grok Bot is not expensive Your setup is 75% of my Grok Bot tokens went to re-reading old chat. About 1% went to actual work I broke down the 6-bot setup the Grok Bot team runs, and the 9 leaks eating your bill ↓
6
25
2
201
32,117
This is f*cking insane. Andrej Karpathy, the man Elon called the #2 computer vision mind on Earth: "Agents aren't magic. They're distillation at scale 99% of what your model knows is dead weight. A small model with the right tools and a closed loop beats it" In 16 minutes, he lays out the stack that actually works, and why the people at the top stopped paying for the biggest model Big Model → Small Model → Right Tools → Closed Loop → Self-Improving Agent Most people keep upgrading the model. The loop is what makes a small one dangerous Worth more than any $1,000 AI course this year Bookmark and watch it today Then read how to build that closed loop step by step in Harness, Loops, Graphs ↓
2
190
This is f*cking insane. Andrej Karpathy: "A 1B model can beat most of today's LLMs. 90% of what they spend goes into memorizing stuff that doesn't matter" In 17 minutes, he explains why the next big model is a small one, and why the harness matters more than the weights Giant Model → Cognitive Core → Tools → Tight Feedback Loop → A 1B Agent You Own A raw LLM is a CPU with no OS. The harness is the OS Worth more than any $2,000 AI engineering course Bookmark and watch it today Then read "Your AI video is not bad. Your checks are.", the harness behind 10 Opus 5.5 motion films ↓
I fixed the same bug in two films Opus never remembers my fixes So I hired an AI employee to remember them for me, then tried to make him break my one rule on deadline He didn't ↓
12
24
2
265
22,995
This is f*cking insane. Andrej Karpathy, the man Elon called the #2 computer vision mind on Earth: "Agents aren't magic. They're distillation at scale 99% of what your model knows is dead weight. A small model with the right tools and a closed loop beats it" In 16 minutes, he lays out the stack that actually works, and why the people at the top stopped paying for the biggest model Big Model → Small Model → Right Tools → Closed Loop → Self-Improving Agent Most people keep upgrading the model. The loop is what makes a small one dangerous Worth more than any $1,000 AI course this year Bookmark and watch it today Then read how to build that closed loop step by step in Harness, Loops, Graphs ↓
90