Andrej Karpathy — OpenAI co-founder, former Tesla AI lead, and creator of the widely-followed neural network education series — has joined Anthropic. Karpathy announced the move on X and confirmed his start date as this week. He joins the company's pretraining team, which leads the large-scale training runs that give Claude its core knowledge and capabilities.
What he will be working on
According to Anthropic, Karpathy is launching a new research team focused on using Claude itself to accelerate pretraining research. The idea: apply the most capable version of Claude to automate and speed up the experimental loop — hypothesis generation, experiment design, analysis — that drives foundational model improvement. This positions Claude as both the product and a tool in its own development pipeline.
He will work under Nick Joseph, who leads Anthropic's pretraining organization. Pretraining is one of the most compute-intensive and strategically important phases of building a frontier model — it is where the model's baseline understanding of language, reasoning, and world knowledge is formed.
Why this matters
Karpathy is one of the most respected technical figures in AI. He co-founded OpenAI in 2015, led Autopilot AI at Tesla, and returned to OpenAI before departing in 2023 to work independently. He is perhaps best known outside research circles for his educational work: the Neural Networks: Zero to Hero series has been viewed tens of millions of times and is widely regarded as the clearest public introduction to how modern language models are built.
His move to Anthropic — rather than OpenAI, Google DeepMind, or independent research — is being read by the industry as a signal of Anthropic's growing momentum in attracting senior technical talent. It follows Anthropic surpassing OpenAI in annualized revenue for the first time in early 2026, and a string of benchmark-leading Claude releases.
Karpathy noted he plans to eventually resume his education work, but that the pretraining research opportunity at Anthropic was too significant to pass up.