GlobeskoolWhat Karpathy Building an AI School Signals for Your Kid's Education
When one of the engineers most associated with the current AI era spends two years building an AI-native school, that is worth understanding. Here is what the signal actually means for parents.
Andrej Karpathy co-founded OpenAI. He built the vision system behind Tesla Autopilot. Then he left one of the most prominent positions in AI to found Eureka Labs, a company with one core idea: build a school designed for AI from the ground up. Not an app. Not a quiz platform. A school.
He has since joined Anthropic’s pretraining team, but he was deliberate about what he left behind. His parting statement on Eureka Labs described education as something he remains “deeply passionate about” and plans to return to. That is not a polite exit line from someone moving on. That is a diagnosis he has not stopped believing.
For parents trying to make real decisions about their kids right now, that pattern is worth sitting with.
What Most Edtech Got Wrong
The dominant response to the gap between school and the real world has been digitization. Take the existing curriculum, put it on a screen, add a progress bar, maybe a badge. The assumption underneath that approach is that the content was fine and the delivery was the problem.
Karpathy’s instinct, and it is worth taking seriously given what he has built, was different. His question was structural: what does a school look like if you do not start from the existing model at all?
That is a harder question and a more honest one. Because the content itself, the what-to-memorize, the passive absorption of information, is not what the next twenty years will reward. Content is abundant and getting cheaper by the month. The ability to evaluate an argument, frame a problem from scratch, and work alongside AI rather than simply consume its outputs is something no algorithm hands a kid automatically.
The Real Skill Gap
Here is the version of this that matters at the kitchen table level.
A kid who can retrieve facts on demand is not the same as a kid who can figure out which facts matter, question where they came from, and decide what to do with them. School, through no particular fault of individual teachers, is largely built around the first thing. The tests measure recall. The homework measures compliance with a format.
The skills that do not appear on most report cards:
- How to frame a problem before trying to solve it
- How to evaluate a source or an argument without being told what to conclude
- How to communicate an idea to someone who does not already agree
- How to work with a tool, including an AI tool, rather than be replaced by it
These are not soft extras. They are the actual capabilities that determine whether a kid can function well in the world being built right now.
Why This Signal Is Loud
Karpathy is not the only person with serious AI credentials who keeps gravitating back to education as the unsolved problem. The pattern is consistent enough to be informative.
What people who understand AI see when they look at current education: a system optimized for a world that is already changing fast, producing students who are comfortable consuming information but not necessarily equipped to think critically about it or create with it.
That is the gap. And it is not a gap that a slightly better quiz app closes.
What the Right Response Actually Looks Like
It is not about replacing school with AI. It is not about passive screen time dressed up as learning. The right model puts the kid in charge of the tool, not the other way around.
A student who uses an AI to generate an essay they did not think through is not learning. A student who uses an AI as a sparring partner to stress-test their own argument, check their reasoning, and revise based on what they actually find is building something real.
The question to ask about any learning environment: Does this build the capacity to think, or does it deliver content the kid can scroll past?
Content delivery is not the shortage. Structured practice in judgment, communication, and problem-solving is.
Where Globeskool Fits
Globeskool is built for kids 8 to 16. Core subjects plus the future skills that standard school has no consistent framework for: critical thinking, problem solving, communication, creativity, and working with technology as a thinker rather than a consumer.
The model is not passive. Real projects, real judgment calls, real feedback. The kid is the one doing the work. That is not a pedagogical preference. It is the whole point.
When engineers who build frontier AI keep identifying education as the most important unsolved problem, they are not being sentimental. They are reading the gap between what the world is becoming and what school is currently producing.
Globeskool is already working on that gap. The free assessment takes about 5 minutes and gives you a clear picture of where your kid actually stands on the skills that will matter in ten years, not just next semester.