Andrej Karpathy, OpenAI co-founder, former Tesla AI director, on what he believes kids should actually spend their time learning.
On the No Priors podcast, Andrej Karpathy, a founding member of OpenAI, Tesla’s former AI director, and now building an AI-native education company called Eureka Labs, was asked a simple question: if you have little kids today, what should they study for a useful future? He didn’t hedge. “There’s a correct answer in my mind,” he said, “and the correct answer is mostly like I would say like math, physics, CS kind of disciplines.” His reasoning wasn’t about which jobs will still exist in twenty years. It was about what those subjects do to how a brain learns to think.
The Reason Isn’t the Content. It’s the Shape.
Karpathy was careful to say this isn’t about memorizing physics formulas or mastering a specific language. Models already handle recall better than people do. His argument is different, working through math, physics, and computer science builds what he called a “thinking skill core.” Not what a kid retains, how a kid learns to reason through something that doesn’t give up its answer easily.
He put a number on it, 80%. “I do think 80% of it should be something like this,” he said, adding that everything else, the electives, the enrichment, the exploration, “you can tack on a bit later.” Not never. Later.
Why the Sequence Matters More Than the Subject List
The part of the conversation most worth sitting with wasn’t the subject list. It was his reasoning for why the sequence matters, kids’ early, high-attention years should go toward “simple manipulation-heavy tasks and workloads, not memory-heavy tasks and workloads.” In plain terms, don’t spend a child’s most capable learning years memorizing facts a search bar already knows. Spend them building the kind of reasoning capacity that’s hard to build later and easy to miss the window for.
His closing line is the one worth keeping, “we’re not efficient memorizers compared to our tools.” That’s not a knock on kids. It’s a fact about the moment we’re raising them in. Information got commoditized the day AI shipped at scale. What didn’t get commoditized is the capacity to reason through something new, and that capacity gets built early, or it gets built harder later.
The Real Question Underneath “What Kids Should Study for an AI Future”
Karpathy gave a confident, specific answer. But notice what he didn’t say, he never once mentioned which jobs will still exist. He wasn’t answering “what’s safe.” He was answering “what makes a person capable,” an answer that holds whether AGI arrives on schedule or not. That’s the real shift underneath his answer, stop optimizing a kid’s schedule around which credential survives, and start optimizing it around which brain is actually being built. The credential question has an expiration date. The capability question doesn’t, it’s the TALENT ingredient, in its earliest, most formative form.
If you’re a parent, the question worth sitting with isn’t which subjects are safest. It’s this: what is your child’s schedule actually building right now, and would you recognize it as capability if you saw it?
What Kids Should Study for an AI Future, Ingredient by Ingredient
What kids should study for an AI future goes beyond a subject list. It maps directly onto the same five capacities I’ve spent 27 years building in adults inside Fortune 50 companies, just starting decades earlier. IDEAS, the instinct to generate rather than retain, gets built the moment a child is asked to work out an answer instead of being handed one. SPEED, the willingness to move and adjust instead of freezing while waiting for certainty, gets built every time a child works through a hard problem without a shortcut. TALENT, the uniquely human reasoning Karpathy is really describing, is the most direct connection of all, and the earlier it develops, the more it compounds.
DISTINCTION starts the moment a child learns to think through something difficult in a way no one else in the room can easily copy. LEADERSHIP AT ALL LEVELS starts even earlier than most parents realize, the moment a child is trusted to work through a hard problem alone instead of waiting to be told the answer. None of these five capacities show up on a report card the way a grade in physics does. But this, not a specific subject list, is genuinely what kids should study for an AI future, and the gap between kids who build it early and kids who don’t will only widen as AI keeps advancing.
AI can replace what your people know. It cannot replace what makes them irreplaceable. That is the whole of what my work is about, and it is never too early to start building it. My next book, working title Future-Proof Your Children: Clarity on Education, Careers, and the Questions You Haven’t Even Thought to Ask Yet (Ages 3 to 30), is devoted to helping parents guide their kids through exactly this, from age three through thirty, in the age of AI.
If you want a clear read on where your own organization, or your own household, stands, the Kryptonite Scorecard will show you: realmikeevans.com/scorecard
Distinct or Extinct: Future-Proofing People and Organizations in the Age of AI – now available on Amazon.
Related Reading
- The Most Powerful Man in AI Just Said the Safest Job Isn’t in Tech, on the same TALENT question this piece raises, applied to adults already in the workforce.
- Two Brothers, Zero Employees, $401 Million: The Medvi Story Through All 5 Kryptonite Ingredients