Some tech bro is going to go work at Anthropic, so what?

courtesy of Getty Images

Yesterday morning, Andrej Karpathy shared that he’s going to be joining the Anthropic pre-training team, and this news exploded across pockets on the internet, especially X and LinkedIn. And of course, everyone has an opinion about it. But guess what? I do too! While it may seem like it’s just some other tech bro joining another tech company, Karpathy’s news is significant for a few different reasons. Before we dive in, though, I should clarify that I am not usually a defender of the “tech bro” or care to give them any more airtime than they already receive. However, I believe this case is different, and after hearing me out, I think you’ll agree too.

Who is Andrej Karpathy?

I’ll save you the biographical deep dive because Wikipedia exists (and shoutout to the Wikipedia editors because they updated his profile almost IMMEDIATELY after he announced it. The devil works hard but Wikipedia works harder). But here are the things that you actually need to know:

• He received his PhD from Stanford and was supervised by none other than THE Fei-Fei Li (iconic pioneer who enabled rapid advances in computer vision). His research was on how deep learning models could be used in the intersection of NLP (Natural Language Processing) and computer vision.
• During his time at Stanford, he also created and was the primary instructor for CS231n: Convolutional Neural Networks for Visual Recognition. This is a foundational comp sci class at Stanford, and a significant number of Silicon Valley professionals today have taken this course (even if he wasn’t the instructor).
• Andrej Karpathy was one of the founding members of OpenAI and served as a research scientist for them from 2015 to 2017, and was instrumental in their early success.
• While Karpathy was at OpenAI, he was poached by Elon Musk (who served as a board member) to serve as Director of AI at Tesla to work on computer vision. Musk hoped that Karpathy would help bring Tesla’s vision for driverless cars to fruition, which still hasn’t come to pass. However, after a long sabbatical, Karpathy announced his departure from Tesla in 2022.
• In 2023, he returned to OpenAI, but a year later, in 2024, he left it again. At the same time, he was named in Time Magazine’s 100 Most Influential People in AI list.
• A few months later, he started his AI education company: Eureka Labs, which has produced courses on LLMs. While I can’t speak to Eureka’s success, Karpathy’s personal YouTube channel explaining the creation and deployment of LLMs has been wildly popular, with many of his videos gaining tens of millions of views.
• In 2025, just last year, he coined the term “vibe coding,” which has become ubiquitous in colloquial terminology to describe how the proliferation of AI tools has empowered anyone from hobbyists and amateurs to create apps, websites, and anything under the sun with just prompts.

Tweet Subtext: A Gen Z Analysis

[embed]courtesy of Andrej Karpathy’s X

Andrej Karpathy’s announcement came via X (shown above). And of course, as any good Gen Z, I started reading between the lines.

What’s key here is that he’s one of the biggest names in AI education. His choice to leave OpenAI a second time was because he believed in making AI more accessible — one can assume that, at the time, it was a more important mission to him than contributing to frontier research. However, he is now making a very active, specific, and intentional choice to go back.

I think that says a lot. Especially because of this part of his tweet: “…the next few years at the frontier of LLMs will be especially formative.” Clearly, he’s identified something that makes building and contributing to AI more critical than teaching it. As the news cycle every couple of months hints or outwardly prophesies at an AI bubble, Karpathy’s stance seems to allude to something else entirely.

The Broader Context?

Furthermore, it’s not just about Karpathy, though, is it? There has been a noticeable and steady exodus of top talent, independents, and those of other frontier companies to Anthropic. All of these people, including Karpathy, could have gone literally anywhere in the world. And yet, they’re choosing to go to — some might even argue that they’re flocking to — Anthropic.

This is key since Anthropic’s framing, while imperfect, is around frontier capabilities and safety. Even a few years ago, this was considered naive. Who could possibly succeed in injecting values and ethics into tech? Silicon Valley is a lawless land. Now this is not to say that Anthropic is some shiny, pure, infallible organization; it’s not, but it has set a standard for responsible, ethical decision-making around AI. Based on the events from a few months ago (which I discussed in my previous post: Anthropic vs. America: Round 1), it now has somewhat of a track record for making difficult choices to uphold these values where others won’t.

I think there are two takeaways here:

1. Frontier companies need to pay attention. If they want to attract and retain top talent, it’s critical to live, not just announce, responsible and accountable AI development. Anthropic may not have been the clear frontrunner in the AI space a few years ago, but they’ve slowly been putting together an almost ironclad case for why they are.
2. For those of us who have long believed that technology can be a force for good and despaired at the way it has been twisted and propagated as a tool of subjugation, exploitation, and harm instead, the tides are shifting. There is a broader conversation and culture shift prioritizing ethical development and deployment of technology, and especially AI. We are no longer naive, dreamers, or people who need to “grow up.” Our values are what distinguish us in what is otherwise a rather saturated space.

Final Thoughts

Andrej Karpathy announced that, after what was effectively a 2-year hiatus, he would be working for the top competitor of a company he helped found more than a decade ago. Think about it. We are living in truly interesting times. But if you take away anything, I think it’s that there’s so much more that can unfold in AI that can tip the scales either way. And it’s up to us as AI safety professionals, aspiring or established, to continue contributing in ways that grapple with the hard questions without taking shortcuts.