Andrej Karpathy joins Anthropic to build frontier LLMs

One of AI’s most famous researchers and educators just jumped back into the frontier race.

On Tuesday, Andrej Karpathy announced on X that he has joined Anthropic. The former OpenAI co-founder and Tesla AI director said the next few years at the frontier of LLMs will be especially formative, adding that he is thrilled to return to hands-on R&D.

Karpathy also noted that he remains deeply passionate about education and plans to resume that work down the road, putting his recent startup Eureka Labs on the back burner for now.

Using Claude to build Claude

Karpathy is not joining Anthropic in an advisory role. He has already started working this week inside Anthropic’s pre-training team under team lead Nick Joseph.

Pre-training is the most compute-heavy and expensive phase of frontier model development, responsible for baking core intelligence into Claude.

An Anthropic spokesperson confirmed to TechCrunch that Karpathy will build a brand-new team focused on using Claude to accelerate pre-training research. In short, Anthropic wants to leverage its own models to streamline and speed up the exploration of next-generation training pipelines.

Karpathy is one of the rare researchers who bridges deep learning theory and massive-scale engineering. After co-founding OpenAI, he led Tesla’s Autopilot and FSD programs until 2022, rejoined OpenAI for a year, and then left in 2024 to launch Eureka Labs and produce widely acclaimed educational videos like Neural Networks: Zero to Hero.

At the same time, Anthropic quietly strengthened its safety bench. Cybersecurity veteran Chris Rohlf announced he is joining Anthropic’s frontier red team to stress-test advanced models against critical threats, coming from Meta and Yahoo’s legendary Paranoids security team.

Why it matters

For developers and students, the immediate downside is that Karpathy’s legendary YouTube tutorials and Eureka Labs projects will likely slow down. He is easily one of the best AI teachers in the world, and taking him off full-time education leaves a noticeable void in public AI learning.

For the frontier race, Anthropic creating an AI-assisted pre-training team under Karpathy is a massive strategic tell. Competing with OpenAI and Google by simply throwing hundreds of thousands of GPUs at brute-force training is becoming economically unsustainable. Anthropic is betting that using Claude to automate research and optimize pre-training architecture is how a leaner lab stays ahead.

We think landing Karpathy is also a huge cultural win for Anthropic. He was an OpenAI insider who saw both Sam Altman’s playbook and Elon Musk’s engineering engine up close. For him to choose Anthropic for what he calls the most formative years of LLM R&D suggests that Dario Amodei’s lab offers an R&D environment that top talent genuinely believes in.