Andrej Karpathy Joins Anthropic to Work on LLMs

Andrej Karpathy announced on Tuesday that he has joined Anthropic. He will focus on large language models and agentic systems and is returning to hands-on lab research after more than a year of independent projects.

Karpathy was a founding member of OpenAI in 2015. He left in 2017 for Tesla, rejoined OpenAI in 2023 for about a year, and departed again in February 2024. After that he started Eureka Labs and produced the educational video series Zero to Hero.

In a public post he wrote, “I’ve joined Anthropic. I think the next few years at the frontier of LLMs will be especially formative. I remain deeply passionate about education and plan to resume my work on it in time.”

Karpathy introduced the term vibe coding in February 2025. He defined it as a workflow where a language model writes code and the user accepts the model’s changes without reading every diff. He later described agentic engineering as a workflow in which people set specifications and monitor output while autonomous agents carry out tasks.

Anthropic was founded in 2021 by former OpenAI researchers and positions itself as safety-focused. Its Claude model family competes with OpenAI’s GPT series. Anthropic released Opus 4.7 in April with improved long-form reasoning and vision capabilities; OpenAI released GPT-5.5 days later.

Anthropic and Karpathy have not disclosed which team or projects he will join. Observers note his past work in neural network design, computer vision and synthetic data aligns with work on LLM training and agentic architectures. Karpathy indicated he plans to continue his education work alongside his role at Anthropic.

Developers have applied vibe coding and agentic approaches to rapid prototyping and to building tools for web3 development. Karpathy’s educational materials emphasized clear explanations of neural networks and language models and attracted a large developer following.