🤚 The Open-Palm Defection

Andrej Karpathy — co-founder of OpenAI, former Director of AI at Tesla, and one of the most recognizable names in deep learning — announced on Monday that he has joined Anthropic. He will work on pre-training under team lead Nick Joseph, specifically building a new team focused on using Claude to accelerate pre-training research.

In his own words: “I think the next few years at the frontier of LLMs will be especially formative. I am very excited to join the team here and get back to R&D.”

For those keeping score at home:

• Karpathy co-founded OpenAI in 2015
• Left to become Tesla’s AI Director, where he led Autopilot’s neural network architecture
• Returned to OpenAI briefly, then departed again in 2024
• Founded Eureka Labs, a startup applying AI to education
• Now walks through Anthropic’s doors carrying a résumé that reads like the entire history of modern AI compressed into one career

The announcement hit #1 on Hacker News with over 1,200 points and 500 comments, which in developer circles is roughly equivalent to the white smoke emerging from the Vatican.

👐 The Two-Handed Talent Heist

Let us be clear about what just happened: Anthropic hired one of the people who built OpenAI and assigned him to the team responsible for making Claude smarter at the most fundamental level. Pre-training is where models learn everything they know before fine-tuning polishes the edges. It is the most expensive, most compute-intensive, and most strategically critical phase of building frontier AI.

This is not a lateral move. This is not a “pursuing other opportunities” LinkedIn post. This is the AI equivalent of your co-founder leaving to coach the opposing team in the Super Bowl.

Karpathy’s decision also says something about the current state of the AI landscape. OpenAI — once the unquestioned destination for top-tier AI researchers — has watched a remarkable parade of talent walk out the door over the past two years. Ilya Sutskever left to start Safe Superintelligence. John Schulman went to Anthropic. Alec Radford stepped away. And now the man who helped write OpenAI’s founding story has chosen to write the next chapter somewhere else entirely.

Meanwhile, Anthropic continues to collect ex-OpenAI talent the way some people collect rare wines — deliberately, expensively, and with the quiet confidence of someone building a very specific kind of cellar.

🌿 The Gentle Awakening

There is something almost poetic about Karpathy’s trajectory. He helped create the organization that kickstarted the modern AI race, then left to teach machines to drive, then left to teach humans about machines, and now he has arrived at the company whose entire thesis is that these machines need to be built carefully.

His note about remaining “deeply passionate about education” and planning to resume that work is the kind of sentence that sounds like a soft landing but is actually a philosophical position statement. The man who built one of the most popular deep learning courses on the internet — who has arguably done more to democratize AI knowledge than any university department — is now working at a lab that believes the stakes of getting this right are existential.

Perhaps when you have spent a decade building the most powerful technology in human history, you eventually want to work somewhere that takes the “powerful” part as seriously as you do.

👑 The Gold-Leaf Recruitment War

The broader story here is one of talent gravity. The AI industry runs on roughly 200-300 people worldwide who truly understand how to train frontier models. Every major lab — OpenAI, Anthropic, Google DeepMind, xAI, Meta FAIR — is fighting over this microscopic talent pool with compensation packages that would make a Wall Street managing director weep into their bonus letter.

Anthropic has now assembled what may be the single most formidable pre-training team in the industry: Nick Joseph leading, Karpathy building a new sub-team, and a roster of researchers who collectively represent more institutional knowledge about large language models than most countries’ entire AI programs.

For OpenAI, this is less a wound and more a pattern. When your co-founders keep leaving for your competitors, the problem is not recruitment — it is retention. And retention, in Silicon Valley’s current AI gold rush, requires more than stock options. It requires a story people want to be part of.

Karpathy just told the world which story he chose.

“When your co-founder joins your competitor’s pre-training team, you don’t send a congratulations email. You send a sommelier to pair the appropriate wine with your existential crisis.” — The Slap of Wisdom Talent Acquisition Bureau, currently updating its LinkedIn profile to ‘Open to Poaching’