AI Trailblazer Calls AI Agents “Slop,” Then Hands the World a $100 Key to Build Its Own AI
In a fiery shake-up, AI pioneer Andrej Karpathy tears into the industry’s hype—then surprises everyone by releasing a $100 blueprint that lets anyone build their own AI.
PALO ALTO, Calif. — In a place where optimism is practically the air people breathe, one of artificial intelligence’s brightest minds just threw a thunderbolt through Silicon Valley’s sky. Andrej Karpathy—the man who helped build OpenAI and led Tesla’s AI team—has a brutally honest word for the current state of AI Agents.
It’s “slop.”
That single, cutting word, dropped casually in a recent interview, rippled across the tech world that’s obsessed with autonomous AI agents. “The industry is making too big a leap and pretending this is amazing,” Karpathy said. “It’s not—it’s slop.”
But this isn’t your typical tale of a genius renouncing his invention. Far from it. Moments after torching the AI hype, Karpathy handed the public a gift. He released nanochat, a free, open-source project that lets anyone build a ChatGPT-style AI from scratch—for just $100 and a weekend of tinkering.
Picture that: with one hand, he’s dismantling the AI hype machine, and with the other, he’s giving you the blueprints to build your own engine. What’s he up to? According to those close to him, Karpathy isn’t worried about a Terminator-style robot rebellion. His real fear is far more human—a future where we become like the cushioned, idle characters from WALL·E, numbed into irrelevance by comfort and automation.
His message is simple but sharp: if you don’t want to be a passive consumer of AI “slop,” you’ve got to roll up your sleeves and learn how it works yourself.
The Ghost in the Machine
So, what’s wrong with modern AI? Karpathy doesn’t sugarcoat it. He says today’s models are like “ghosts, not animals.” They mimic intelligence brilliantly but don’t truly understand. These systems feed on snapshots of the internet—static data frozen in time—without the grounding or real-world learning that defines living creatures.
“We’re building cortex without the older brain circuits,” he explained. In plain English, that means we’re creating clever sleepwalkers. They respond well, they sound social, but at their core, they’re not awake.
And that’s why AI agents still stumble when you push them beyond simple tasks. They’re brittle. They improvise poorly. Ask one to get creative, and nine times out of ten, you’ll get a bland blog post that could’ve come from a corporate template.
“This isn’t the ‘Year of Agents,’” Karpathy quipped, poking fun at tech marketing. “It’s the ‘Decade of Agents.’” Moving from 99% reliable to 99.9% reliable isn’t just a small step—it’s a mountain climb. And most people, he warns, underestimate that climb entirely.
A $100 Reality Check
Enter nanochat, his bold antidote to the industry’s noise. It’s essentially an entire AI education compressed into a single folder of code.
For roughly the cost of a nice dinner, you can run one script that builds a small but functional chatbot in just four hours. It constructs its own vocabulary, trains on billions of words, and chats through your web browser. The result? It’s rough around the edges—Karpathy calls it a “kindergartener” that “hallucinates a ton”—but it’s yours.
“What makes nanochat unique,” he says, “is that it’s fully yours—tweakable, hackable, and trained by you from start to finish.”
It’s not trying to dethrone the giants like OpenAI or Google. It’s here to lift the curtain. Nanochat turns the mysterious art of AI into a hands-on masterclass. You learn by building.
And here’s the kicker: Karpathy revealed that AI assistants were practically useless while creating nanochat. Their canned suggestions couldn’t handle the nuance. Ironically, the very project built to expose AI’s limitations couldn’t be built by AI. In the end, human intuition—the spark of creativity machines still lack—carried the day.
Training the Next Generation
This blend of critique and creation forms the backbone of Karpathy’s new venture, Eureka Labs. His dream? To build a “Starfleet Academy” for AI—a place where anyone can learn the craft and understand the systems shaping our future.
He’s not warning about killer robots. His fear is subtler and more immediate: that we’ll slowly hand over our thinking to shallow, overconfident “slop” bots and forget how to reason ourselves.
Nanochat, then, isn’t just code—it’s a manifesto. Karpathy’s betting that the next wave of innovation won’t come from larger models or slicker marketing. It’ll come from people—people who truly get the tech that’s reshaping their world.
In an era overflowing with hype, fear, and glossy promises, Karpathy stands out as something rare: a teacher. He looked at the chaos of his own field, called it out with one ruthless word, and then offered the world a challenge.
Stop consuming the slop. Go build your own.