Andrej Karpathy Joins Anthropic as OpenAI Co-Founding Member Defects

Andrej Karpathy, one of the original 11 co‑founders of OpenAI and the former director of AI at Tesla, announced Tuesday that he has joined Anthropic. The 39‑year‑old Slovak‑Canadian researcher posted a "personal update" on X that drew nearly 3 million views within an hour, Fortune reported. In the post,,³ Karpathy wrote:

Karpathy is one of the most influential figures in modern AI. He earned his PhD at Stanford under renowned scientist Fei‑Fei Li, then became one of OpenAI's earliest staffers when the lab was founded in 2015, Reuters notes. In 2017, Elon Musk poached him to lead Tesla's AI division, where he ran the computer vision team behind Autopilot and Full Self‑Driving until 2022.

He returned to OpenAI for a one‑year stint in 2023‑2024, leading work on midtraining and synthetic data generation, according to VentureBeat. He then founded Eureka Labs, an AI‑native education startup, and became one of the most‑watched AI educators online, with nearly 2 million followers on X.

In February 2025, Karpathy coined the term "vibe coding" describing a new kind of software development where builders describe what they want in plain language and let the model generate the code. The phrase became Collins Dictionary's Word of the Year and sparked the much‑debated "SaaSpocalypse" narrative, Fortune reported.

Karpathy joins Anthropic's pretraining team under Nick Joseph, the company's head of pretraining. He will build and lead a new group with a specific mission: use Claude itself to accelerate pretraining research, TechCrunch confirmed. Joseph welcomed the hire on X, VentureBeat reported:

Pretraining the massive initial training runs that give models like Claude their core knowledge is the most compute‑intensive and expensive phase of frontier AI development. Making it more efficient could fundamentally reshape the economics of building AI models.

Karpathy's new role is a natural extension of his most striking recent experiment. In March 2026, he built an autonomous AI coding agent called "autoresearch" that ran 700 experiments over two days on a small language model. The agent discovered 20 self‑found optimizations that, when applied to a larger model, reduced training time by 11%, ⁵ detailed in a deep dive on what became known as "the Karpathy Loop."

Shopify CEO Tobias Lutke validated the approach on internal company data. After running overnight, the agent performed 37 experiments and delivered a 19% performance gain. Karpathy described the technique as "the final boss battle" for all frontier AI labs.

The long‑term vision goes beyond single agents. Fortune reported Karpathy's plan for "asynchronously massively collaborative" agent swarms not emulating a single PhD student, but an entire research community.

Karpathy's move is the latest and highest‑profile defection in a growing exodus from OpenAI. Co‑founder John Schulman left for Anthropic in 2024. Former CTO Mira Murati departed to found Thinking Machines. Chief scientist Ilya Sutskever also exited. Of OpenAI's original 11 co‑founders, only a handful remain, The Next Web noted.

The hire comes during an extraordinary growth phase for Anthropic. The company's annual recurring revenue reportedly surged in early 2026, with independent analysts estimating it jumped from roughly $10 billion to $44 billion over a span of months. Anthropic has received investment offers valuing it at more than $900 billion, more than double its previous valuation, The Wall Street Journal reported. Both Anthropic and OpenAI are racing toward public offerings by late 2026.

The hire follows Anthropic's recent acquisition of Stainless, the SDK platform used by OpenAI, Google, and Cloudflare, for at least $300 million — and the addition of cybersecurity veteran Chris Rohlf to its frontier red team, TechCrunch reported.

Karpathy addressed the future of his education work directly in his announcement on X, writing,:³ "I remain deeply passionate about education and plan to resume my work on it in time." Eureka Labs, his AI‑native school, and LLM101n, the undergraduate course he designed for students to train their own AI models, are now paused.

His open‑source projects including autoresearch and the LLM Knowledge Base now face an uncertain future. While Anthropic has released the Model Context Protocol (MCP) as an open standard, its core models remain proprietary. Unlike his two‑year period as a free agent between Tesla and his OpenAI return, Karpathy's attention will now focus on internal Anthropic R&D.

Karpathy's move carries three concrete signals for anyone building with AI tools: