Today's signal

Andrej Karpathy, OpenAI co-founder, former Tesla AI director, and the researcher who taught millions of people how LLMs work for free on YouTube, announced on May 19 that he is joining Anthropic. He starts this week on the pre-training team, which handles the large-scale training runs that give Claude its core capabilities. He will also lead a new team focused on using Claude itself to accelerate pre-training research.

Why it matters

Pre-training is where the fundamental capability ceiling of any AI model is set. Every downstream application inherits whatever limits the base model has. Anthropic's bet is specific: that AI-assisted research can compress the timeline for improving the model, creating a compounding advantage that pure compute spending cannot easily replicate. Karpathy is one of a small number of researchers in the world who can credibly lead that effort. His joining is also not an isolated event. Mike Krieger (Instagram), Peter Bailis (Workday), Bryan McCann (You.com), Niki Parmar (Adept AI), and Henry Shi (Super.com) have all taken Member of Technical Staff positions at Anthropic in the past year, passing up titles and comfortable trajectories to do foundational work.

The take

Of all the places Karpathy could have gone, he chose Anthropic. He did not go back to OpenAI, the lab he helped found. That choice is not a footnote. Anthropic is winning the argument that safety and capability are the same bet, not opposing ones.

The number

6 — the number of senior technical leaders who left positions at major companies to take research roles at Anthropic since mid-2025: Mike Krieger (Instagram), Peter Bailis (Workday), Bryan McCann (You.com), Niki Parmar (Adept AI), Henry Shi (Super.com), and now Karpathy. The AI race is framed around compute and funding. The talent data is pointing somewhere else.

Read the full breakdown → https://analyticsdrift.com/andrej-karpathy-joins-anthropic