On May 19, 2026, Andrej Karpathy announced that he is joining Anthropic to lead pre-training research, working directly under Nick Joseph, who oversees Claude’s technical development. The announcement, posted on X, received 148,000 likes and more than 7,800 responses, a level of engagement that reflects how closely the AI community watches Karpathy’s career decisions.
Karpathy is one of the most respected researchers in the field. He co-founded OpenAI in 2015, left to serve as Director of AI at Tesla where he built the Autopilot neural network stack from the ground up, and then shifted to independent work and educational content. His neural networks course on YouTube has accumulated millions of views, making him one of the rare researchers who can explain foundational AI concepts at a level accessible to practitioners far outside research labs.
His role at Anthropic is focused on pre-training, the phase of model development that determines the core knowledge, reasoning capabilities, and behavioral foundations of a language model. Pre-training is also the most computationally expensive phase of building a frontier model, involving massive training runs on large-scale compute clusters over weeks or months. According to Anthropic’s statement to TechCrunch, Karpathy will build a new team specifically focused on using Claude to accelerate pre-training research, a self-referential approach where the model helps improve its own development process.
The move carries strategic weight for Anthropic in multiple ways. First, Karpathy brings deep experience in scaling neural network training from first principles, not just in theory but from having built production systems at Tesla that had to work reliably at global scale. Second, his public profile gives Anthropic a credibility signal in frontier AI research that has tangible effects on recruiting, enterprise trust, and partnerships. Third, his appointment reinforces the view that pre-training, not just inference-time improvements or fine-tuning, remains a primary lever for improving frontier model capabilities.
Also joining Anthropic in the same period was Chris Rohlf, a cybersecurity veteran with more than 20 years of experience, who is joining the frontier red team. The combination of a top pre-training researcher and a senior security professional joining in the same week suggests Anthropic is building capacity ahead of a significant model release.
For sales teams tired of cold leads, slow customer responses, and manual processes, Dapta is the ultimate tool.
Dapta is the leading platform for creating AI sales agents specifically designed to increase inbound lead conversion. Respond to your leads in less than a minute with voice AI and WhatsApp that converts.
If you want your team to sell more while AI handles the complex stuff, you have to try it.
For enterprise buyers evaluating Anthropic’s Claude against competing models, Karpathy’s arrival is a qualitative signal about the organization’s technical direction. Research talent at this level does not join companies without strong conviction about the underlying technical approach. His choice of Anthropic over other available paths says something about where he believes the most important work in pre-training is happening today.
