Andrej Karpathy, an OpenAI co-founder and former Tesla AI director, joined Anthropic this week as a member of the pre-training team under team lead Nick Joseph, with plans to spin up a dedicated unit using Claude to accelerate pre-training research, an Anthropic spokesperson confirmed to TechCrunch.
Karpathy announced the move on X on Tuesday, writing that he had joined Anthropic and that "the next few years at the frontier of LLMs will be especially formative." He added that he is "very excited to join the team here and get back to R&D."
The mandate
The role places Karpathy at the center of how Anthropic builds its next generation of Claude models. Pre-training is the most compute-intensive stage of frontier model development, and the new unit he is set to lead will be focused on using Claude itself to accelerate the work — designing experiments, evaluating runs, and triaging data quality. That direction matters because labs are increasingly betting on AI-assisted research as a multiplier on top of raw compute.
Career arc
Karpathy's CV reads as a chronology of three AI eras. He was a founding researcher at OpenAI in 2015, working on deep learning and computer vision. He left in 2017 for Tesla, where he led Full Self-Driving and Autopilot programs until 2022. He briefly returned to OpenAI before departing again in 2024 to start Eureka Labs, an AI education company. Outside the labs he is widely known for the "Neural Networks: Zero to Hero" YouTube series, a default onboarding path for many ML engineers.
What it signals
The hire lands during a stretch in which Anthropic's run-rate revenue has reportedly crossed $30 billion, and the company has expanded compute commitments through partnerships with Google and Broadcom. Pre-training talent is the binding constraint at the frontier: labs spending billions on TPUs and GB300 clusters still need researchers who can extract efficiency from attention, optimizers, and curriculum design. Adding a researcher with Karpathy's depth in scaling theory measurably shifts Anthropic's bench on a key vector against OpenAI and Google DeepMind.
For builders and enterprise
For teams building on Claude, the implications are twofold. First, Anthropic's pre-training research velocity stands to accelerate — particularly any Claude-assisted training tooling that gets productized downstream. Second, the move strengthens Anthropic's pitch to enterprise buyers consolidating on a single frontier vendor: the lab is retaining and recruiting the caliber of senior research talent that signals long-term capability investment, not just commercial scale-up.
Karpathy noted he remains "deeply passionate about education" and plans to resume that work "in time" — implying Eureka Labs is on pause rather than wound down.
