ANDREJ KARPATHY to CLAUDE FC — HERE WE GO ✅

GPT United's founding midfielder walks out of Eureka Labs and into Anthropic's pre-training lab. 148,000 likes by end of day. The story everyone saw coming, and nobody predicted.

On the morning of Tuesday, May 19, ANDREJ KARPATHY posted nine sentences to X. He'd joined Anthropic. He thought the next few years at the frontier of large language models would be "especially formative." He was glad to be back doing research. By end of day, the post had 148,000 likes.

That number tells you how closely this particular career gets watched. Karpathy is not just a known name in the AI industry — he is the industry's most-cited explainer, its most-watched educator, and one of its eleven founding scientists. When he picks a club, the whole league notices.

Position: Pre-training researcher, Claude FC (Anthropic)

Karpathy's career arc reads like the perfect transfer dossier. He completed his Stanford Ph.D. in 2016 — neural networks, computer vision, NLP — and left straight for OpenAI as one of eleven original co-founders. Eighteen months of foundational model research, then Tesla pulled him sideways.1

At Tesla Autopilot, he served over five years as senior director of AI, overseeing the computer vision team that taught cars to read roads. He returned to OpenAI in early 2023 for a second stint, left again roughly a year later, and spent the stretch after that building Eureka Labs — an AI-native education startup built around the idea that any subject can be taught if the AI tutor is good enough. His Neural Networks: Zero to Hero video series accumulated millions of views. He coined "vibe coding" in February 2025 — a phrase that became the informal slogan for an entire generation of developer workflow.1

He could have gone back to OpenAI a third time. By most accounts he would have been welcomed. He chose Anthropic.

GPT United's recent seasons have not been without turbulence. Chief scientist Ilya Sutskever departed to found Safe Superintelligence Inc. Former CTO Mira Murati left to launch Thinking Machines Lab — currently raising a $2B seed round that would rank as the largest in history.2 Karpathy's second departure adds another founding-era figure to the alumni roll.

His reasoning, as stated publicly, is quiet and precise: the next few years at the frontier are especially formative, and he wants to be in the room where the work happens. Eureka Labs — his education project — is "paused for now." Education can wait. Pre-training at a lab approaching a trillion-dollar valuation cannot.

His new manager, Nick Joseph, who himself moved from OpenAI to Anthropic nine months earlier, put it plainly on X: "He'll be building a team focused on using Claude to accelerate pretraining research itself. I can't think of anyone better suited to do it."

The assignment is narrow and, if it works, structural. Karpathy is not being asked to train a single model. He is being asked to build a team inside Anthropic that uses Claude to make future Claudes faster and better — a research-on-research function that bets the next capability leap comes from AI-accelerated science rather than raw compute scaling alone.1

The week Karpathy joined also saw Anthropic close a $65 billion Series H at a $965 billion valuation, acquire Stainless for $300 million, and drop Claude Opus 4.8 — with new dynamic workflows, agentic coding upgrades, and improved honesty metrics. The timing is probably not coincidental. The club is building for a title run, and it just hired the person best able to explain how the engine works.

Pseudonymous commentator @signulll (198K followers) summarized the hire in sporting terms: "This is like a leading franchise recruiting someone who's simultaneously the best player, the league's best broadcaster, and its most watched developmental coach all in one."

In the 2011 winter transfer window, Pep Guardiola's Barcelona signed Cesc Fàbregas back from Arsenal — a player who grew up in La Masia, left for the English Premier League, and returned a decade later to run the engine room of the side he was always supposed to play for. The move wasn't about buying talent from a competitor. It was about recovering an identity.

Karpathy started AI's equivalent of La Masia — OpenAI's founding cohort. He left, shaped an industry, and is now back at a frontier lab. Except this time the lab is not the one that originally built him. That's the interesting part. He passed on the obvious homecoming and chose somewhere new.

The bet Anthropic is placing: that a researcher who has spent the last decade explaining how language models work, at Tesla scale and in lecture halls, is the right person to point those same models at their own training stack. AI improving AI via the person who taught the most humans to understand AI.

Karpathy's Eureka Labs work is paused, not cancelled. At some point the education thread picks back up. In the meantime, Claude FC has its most prominent signing of the season — maybe the decade — and a research function that didn't exist three weeks ago.

The standings across the league shifted on May 19. File that date.

#AILeague