In a significant shift within the AI talent landscape, Andrej Karpathy, a renowned technical evangelist and the originator of the term "Vibe Coding," has announced his move to OpenAI's primary competitor, Anthropic.
Late on May 19, Karpathy posted a brief statement on X, which garnered nearly 3 million views within an hour. He wrote:
"Personal update: I have joined Anthropic. I believe the frontier development of large language models will be particularly important in the coming years. I'm excited to join the team and return to R&D work. I also remain passionate about education and plan to resume related efforts in due course."
Karpathy officially started this week, joining Anthropic's pre-training team, which is led by another former OpenAI employee, Nicholas Joseph. This team is responsible for all large-scale training runs for Claude. An Anthropic spokesperson confirmed to TechCrunch that Karpathy will form a new sub-team focused on using Claude itself to accelerate pre-training research.
The recruitment has been characterized by Axios as a "major victory for Anthropic in the talent war."
**Who is Karpathy: An Individual Capable of Shifting an Industry with a Word**
To grasp the significance of this move, one must understand Karpathy's unique profile.
Born in Slovakia in 1986, he immigrated to Canada at age 15. He took a course from deep learning pioneer Geoffrey Hinton during his undergraduate studies at the University of Toronto and later earned his PhD at Stanford under the guidance of Fei-Fei Li. During this time, he created the computer vision course CS231n, which grew from 150 students in 2015 to 750 in 2017. Its publicly available videos became a global entry point for engineers self-studying deep learning.
In 2015, he became a founding research scientist at OpenAI. In 2017, he was recruited by Elon Musk to
After leaving
In February 2025, he coined the term "Vibe Coding" on X—"fully intuitive, embracing exponential growth, forgetting the existence of code itself"—which was later selected as the Word of the Year by Collins Dictionary, sparking widespread discussion about the survival of SaaS companies and erasing billions in market value.
TIME included him in its 2024 list of "100 Most Influential People in AI." He has 2.5 million followers on X.
As X user @sa_vatsa commented:
"Andrej never felt like an AI influencer, more like a public interpreter for the field. He translates reality, he doesn't sell a position."
**What He Will Do: Using the Strongest Hammer to Improve the Hammer Itself**
Karpathy's specific role at Anthropic points to a clear direction: using Claude to accelerate pre-training research itself.
Pre-training is the foundational phase where large models acquire core knowledge and capabilities, and it is the most computationally intensive stage. According to Anthropic's confirmation to TechCrunch, Karpathy will form a team to explore methods for enabling Claude to autonomously discover better training architectures, data mixtures, and experimental directions—essentially pursuing the "AI improving AI" path.
This is not without precedent. In March of this year, Karpathy conducted an experiment: connecting an AI programming agent to a small language model and letting it run unsupervised for two days, autonomously testing and adjusting training code. After 700 experiments and 20 autonomously discovered optimizations, he stated that applying the same adjustments to a larger model reduced training time by 11%. He termed this method "autoresearch," and the project subsequently became a trending GitHub repository, with the approach dubbed the "Karpathy Loop."
His work at Anthropic will involve systematizing and scaling this very concept.
Nicholas Joseph wrote on X:
"He will build a team focused on using Claude to accelerate pre-training research itself. I can't think of anyone better suited for this—excited for what we'll build together!!"
TechCrunch noted that "Karpathy is one of the few researchers who can bridge the gap between LLM theory and the practice of large-scale training."
**Why Anthropic, and Not a Return to OpenAI?**
Karpathy's career trajectory follows a clear logic: he goes where the most significant current experiment is.
In 2017, he went to
This time, choosing Anthropic reflects his judgment: the pre-training revolution of "using AI to research AI" is happening here.
His decision not to return to OpenAI carries a notable structural signal:
* In May 2024, former OpenAI alignment lead Jan Leike joined Anthropic. * In August 2024, OpenAI co-founder John Schulman followed. * In May 2026, Karpathy joined.
Over three years, three core OpenAI figures have moved unidirectionally to Anthropic, with no comparable reverse cases.
Simultaneously, OpenAI's strategic focus has clearly shifted towards platformization—acquiring Chat.com, io Products, Windsurf, and others at an increasingly rapid pace and with larger sums. For a researcher wanting to "return to R&D," Anthropic's "winning on research quality" path is evidently more attractive.
According to Business Insider, Anthropic's valuation in secondary markets recently surpassed $1 trillion, exceeding OpenAI's. Anthropic is reportedly in talks for a new $30 billion funding round at a valuation of approximately $900 billion. Citing Polymarket data, traders have priced the probability of Anthropic having the best AI model by the end of June at 65%, compared to 4% for OpenAI.
**"Anthropic is Winning the AI Race"**
Reacting to this major personnel move, X user @Shashank commented:
"Anthropic is starting to look like Real Madrid in the mid-2000s, bringing in all the AI legends and CTOs as technical fellows. Anthropic is building an absolute super team."
X user @Balu0X stated:
"Wow, insane, did not see this coming at all. Anthropic is winning the AI race."
User @HegsethWarDesk added:
"Anthropic has great momentum, I think they will surpass OpenAI in the next three to four years."
**Why Anthropic Was Determined to Recruit Him**
Anthropic's recruitment motives can be viewed on several levels.
* **Technical Level:** Anthropic's compute budget cannot directly compete with Microsoft-backed OpenAI or Google with its TPUs. It must find ways to train better models with less compute. Karpathy possesses a rare combination of deep pre-training theory knowledge, large-scale engineering experience, and intuition for AI-assisted research. * **Talent Signal Level:** Every addition of a Karpathy-caliber individual lowers the psychological barrier for the next top talent to join. Talent attracts talent, creating a self-reinforcing cycle. * **IPO Narrative Level:** Karpathy is one of the most publicly recognizable technical figures in AI—with a million YouTube subscribers, a Word of the Year coinage, and a CLAUDE.md GitHub repository with 220k stars. His name on the employee roster is itself a statement that could be included in an IPO prospectus.
However, the deepest value, as some analysis suggests, might be unstated by Anthropic: Karpathy's ability to define paradigms. Any technical exploration he undertakes at Anthropic will be publicly disseminated through his tweets, blogs, and YouTube videos. When he names what's happening in his characteristic way, Anthropic naturally becomes the origin of that paradigm.
By hiring a top pre-training researcher, Anthropic has also gained the industry's most influential technical narrator.
Comments