Karpathy Kept Betting His Career on Fixing School. Here Is What Parents Should Do With That

The Part of the Karpathy Story That Gets Buried

Most coverage of Andrej Karpathy focuses on the career trajectory: co-founder of OpenAI, director of AI at Tesla, YouTube’s most-watched AI educator, now at Anthropic. That resume is genuinely impressive, and it makes for a clean professional narrative.

What gets less attention is what he did in between.

After leaving OpenAI in early 2024, Karpathy spent nearly two years building Eureka Labs, which he described as “a new kind of school that is AI native.” Not an app. Not a tutoring tool. A school. He has since paused that project to join Anthropic’s pre-training team, but his public statement on the way in was deliberate: “I remain deeply passionate about education and plan to resume my work on it in time.”

That is not the language of someone who dabbled. That is someone pointing at an unfinished problem.

What He Was Actually Pointing At

Karpathy’s stated concern was specific. He wrote that subject matter experts who are deeply passionate, patient, fluent in every language, and available to every student on demand are scarce. The current school model cannot solve that scarcity problem at scale.

This is not a novel observation. Sebastian Thrun, who ran the Stanford AI lab and helped build Google X, left to found Udacity for similar reasons. Andrew Ng has spent over a decade working on accessible AI education through Coursera and other projects. The pattern is consistent: people who understand where technology is heading keep looking at the K-12 model and arriving at the same conclusion.

The curriculum most kids are sitting through right now was not designed for the world these engineers are building. It was designed for a different one.

The Mistake Parents Are Most Likely to Make Here

The tempting response to a story like this is to wait. Wait for Karpathy’s next move. Wait for the school to update its approach. Wait for some cleaner version of AI-integrated education to arrive fully formed.

That response treats the signal as news about the future. It is actually news about the present.

The gap Karpathy kept pointing at, between what a standard curriculum delivers and what kids will need to navigate a world shaped by AI, is not a coming problem. It is a current one. The skills missing from most report cards are not advanced or exotic. They are things like:

• Framing a problem before trying to solve it. Most kids are handed problems with the frame already built. Real work does not come that way.
• Working with AI as a thinking partner, not a shortcut. There is a meaningful difference between a kid who pastes a question into a chatbot and accepts the output, and a kid who interrogates that output, redirects it, and decides what to trust. The second skill is learnable. It is not widely taught.
• Communicating with precision. Not presentation skills. Not public speaking for its own sake. The ability to say exactly what you mean, in writing and in conversation, in a way another person can actually use.
• Learning how to learn. This sounds abstract until you realize most kids are never explicitly taught it. They are assessed on content recall, not on how they build understanding when the content is unfamiliar.

Why the “AI Will Handle It” Answer Is Wrong

Some parents land on the opposite conclusion: if AI is going to be so capable, why does any of this matter. Let the tools do the heavy lifting.

This gets the relationship between the kid and the tool exactly backwards.

The kid who knows how to direct AI, question its outputs, identify where it is confidently wrong, and decide when not to use it at all, that kid is more capable because of the tool. The kid who is simply downstream of AI output, receiving and forwarding it, is less capable over time, not more.

The kid needs to be in charge of the tool. That requires judgment. Judgment is built through practice on real problems, not through passive consumption of content or even well-designed AI interactions that do the thinking for you.

This is what Karpathy’s vision for Eureka Labs kept returning to: AI as a teaching assistant that works alongside students, not as a replacement for the student doing the work.

What This Actually Means for Your Kid Right Now

Globeskool is an online school for kids aged 8 to 16 that covers core subjects alongside the skills that standard curricula consistently defer: critical thinking, problem framing, communication, creativity, and working with technology in ways that build judgment rather than replace it.

The free assessment takes about 5 minutes and shows you where your child actually stands, not on standardized test categories, but on the capabilities that will matter in the years ahead.

If engineers at the level of Karpathy keep circling back to the same broken system, parents probably should not wait for a note home from school about it.

Take the free assessment