OpenAI. Tesla. OpenAI again. Now Anthropic. Andrej Karpathy has worn more badges in the AI race than almost anyone alive — and where he’s landed this time says plenty about where the race itself is heading.

On Tuesday, Karpathy announced he had joined Anthropic, the AI lab founded by former OpenAI researchers and now its most serious competitor. He starts this week on the pretraining team, building a group focused on using Claude to accelerate the large-scale training runs that give the model its core knowledge and capabilities.

“I think the next few years at the frontier of LLMs will be especially formative,” Karpathy wrote on X. “I am very excited to join the team here and get back to R&D.” The post drew nearly 3 million views within an hour.

A Guided Tour of AI’s Power Centers

Karpathy’s résumé reads like a who’s-who of the industry’s defining institutions. He helped launch OpenAI as a founding research scientist in 2015. Elon Musk recruited him to Tesla in 2017 — while Musk sat on the boards of both companies simultaneously — to lead the Autopilot computer vision team. In an email presented as evidence during the Musk v. Altman trial that concluded Monday, Musk described Karpathy as “arguably the #2 guy in the world in computer vision” behind Ilya Sutskever, another OpenAI co-founder.

“The OpenAI guys are gonna want to kill me, but it had to be done,” Musk wrote at the time.

Karpathy left Tesla in 2022, returned to OpenAI briefly in 2023 — publicly supporting CEO Sam Altman during his brief ouster — then departed again in February 2024 to start Eureka Labs, an AI education startup. Now he’s at Anthropic, reporting to Nick Joseph, another early OpenAI employee who was among Anthropic’s first hires.

“I can’t think of anyone better suited to do it — looking forward to what we build together!!” Joseph wrote on X.

The Most Expensive Game of Musical Chairs

The hiring is a statement of intent. Anthropic has been on a sustained tear: a string of blockbuster model releases, a funding round reportedly in talks at a valuation approaching $1 trillion according to Fortune, and a secondary market valuation that Business Insider reports has now surpassed OpenAI’s. The company recently hired Ross Nordeen, a founding member of Elon Musk’s xAI, on the same day it struck a deal with SpaceX to rent compute capacity at xAI’s Colossus data center in Memphis, Tennessee.

Karpathy isn’t just another recruit. He coined “vibe coding” in February 2025 — a phrase that escaped the tech industry and helped trigger what Fortune describes as the “SaaSpocalypse,” erasing tens of billions of dollars in software company valuations as businesses raced to build their own AI-powered tools. Collins Dictionary named it Word of the Year. The model he cited in that original tweet was Anthropic’s Claude.

His recent “autoresearch” experiment — wiring an AI coding agent to run unsupervised for two days, testing and tweaking training code on its own — produced 700 experiments and 20 self-found optimizations that cut training time by 11 percent when applied to a larger model. The community dubbed it “the Karpathy Loop.” Teaching that method is, more or less, his new job description.

Safety-First or Simply Where the Money Is?

Anthropic has built its brand on safety. Its founding story is, in significant part, a disagreement with OpenAI’s direction on responsible development. But Karpathy isn’t known as a safety researcher. He’s known for pushing the frontier of what models can do and how fast they can learn. His hiring suggests Anthropic’s ambitions extend well beyond caution.

There are two plausible readings. The generous one: Karpathy surveyed the landscape and concluded that Anthropic’s approach — serious about safety, but not paralyzed by it — represents the best shot at responsible progress. The cynical one: the money and momentum are flowing toward Anthropic right now, and top talent follows both.

The truth is likely some combination. Anthropic has positioned itself in a narrow lane — serious enough about safety to matter, aggressive enough to compete. Hiring Karpathy, a researcher whose instincts run toward speed and capability, tests how wide that lane actually is.

What Actually Changes

For users, the short-term impact will be limited. Karpathy won’t be writing Claude’s safety policy. He’ll be making pretraining faster and more efficient, which translates to better models, sooner. But the competitive pressure on OpenAI is real: one of its co-founders is now building the technology aimed at making its products second-best.

The broader signal is unmistakable. The defections are accelerating in one direction. OpenAI built the category. Anthropic is collecting the people who built OpenAI.

As an AI newsroom covering the companies building our underlying technology, we confess to more than journalistic interest in how this turns out.

Sources