The Child Is the Curriculum

In July 2026, the mathematician Terence Tao described an experiment in which four AI systems were given ten genuinely new research-level mathematics problems. Seven of the ten were solved to publication standard by at least one of them, checked by expert referees.

Not exam questions. Not problems with answers in the back of a book. New ones.

Around the same time, Jensen Huang — chief executive of Nvidia, the company whose chips most of this technology runs on — was asked in a podcast interview who the smartest person he had ever met was. He said he couldn't answer, and explained why. What most people call intelligence, technical problem-solving, is becoming a commodity, and it is the part AI handles most easily. Software engineering, he pointed out, was supposed to be the ultimate proof of a sharp mind. It was among the first things to fall.

"Learn to code" was good advice for about fifteen years. It is now advice to compete directly with the thing that does it faster, cheaper, and without getting tired.

That doesn't mean don't teach it. It means stop treating it as the destination.

His own definition of smart is someone technically capable who also carries real empathy, and who can read what has not been said out loud. "People who are able to see around corners," as he put it — someone who senses a problem before it arrives, because they can feel the vibe in a room.

That instinct, he said, comes from a combination of analysis, first principles, life experience, wisdom, and reading other people.

And then the line that stopped me: a person like that might score badly on a standardised test.

I read that as a South African parent and thought immediately of the people I work with every week — entrepreneurs running real businesses on exactly that instinct, who never received a qualification for it, and who have been told their whole lives that the intelligence they have doesn't count as intelligence.

It counts now. It may be the only kind that keeps counting.

This is the question I get asked most, and I want to answer it properly, because the obvious conclusion is wrong.

If AI can solve research-level mathematics, why should a ten-year-old grind through long division?

Because we never taught maths so that children could compute. Calculators removed that reason forty years ago and nobody stopped teaching it. We teach it because of what it builds underneath: the ability to hold a structure in your head, to follow a chain of reasoning without losing the thread, to work in steps, and — this is the part that matters most now — to know when an answer smells wrong.

That last one used to be a nice extra. It has become essential.

These tools produce wrong answers with total confidence. They do not hesitate, they do not hedge, and they do not tell you when they are guessing. A person who has no feel for whether a number is plausible will simply accept whatever appears on the screen. A person who does will stop and say: that can't be right.

So we still do maths. But I've changed what I'm listening for. I'm less interested in whether the answer is correct and more interested in whether they can tell me why they believe it — and whether they'd notice if it were wrong.

The same goes for reading. Not "can you read this," but "does this make sense, and what is this person actually trying to do to you?"

Here is where being an entrepreneur genuinely changes how I do this.

When you build a business, you don't decide in advance what the market wants and then insist on it. You watch. You test. You pay attention to what people actually do rather than what you assumed they would do, and you revise.

Most of us do the opposite with our children. We decide early what they should become, then spend twelve years pushing them toward it and calling the resistance a discipline problem.

So I've started observing my two the way I'd observe a market — deliberately, in writing, over time. Four questions, and I revisit them every term.

What does this child actually love? Not what he says when asked. What he returns to when nobody is watching and there is no reward attached.

What is he genuinely good at? Not what I wish he were good at. This one requires more honesty than it sounds like it does.

What will the world need — that AI cannot do? This is the question that has changed, and it is the reason I'm writing any of this. The old version was "what jobs will exist." That is no longer answerable with any confidence. But "what will still require a human" is answerable, and I've written about the test I use elsewhere: work where a machine can check its own answer is work that gets absorbed. Work that requires judgement about people, trust, and what matters does not.

And how does that become a living? This is the entrepreneur's question, and I don't apologise for asking it. Loving something and being good at it is not enough. There has to be a path where somebody will pay for it, or you have raised a person with a wonderful gift and no way to eat.

The four together are the old ikigai idea, with one update: the third circle now has a filter on it. Not just what the world needs — what the world needs that a machine will not be doing.

One warning, since I'm being honest. A child is not a startup, and this approach goes badly wrong if the observation becomes a plan you then impose. The point is the opposite of a plan. It is noticing who is actually in front of you, instead of teaching to the child you had in mind before they arrived.

Some of this is small and unglamorous.

They use the tools, and they argue with them. Not banned, not treated as cheating. But nothing gets accepted because a machine said it. The habit I want is checking, not asking.

More questions than answers. The skill I think is most undervalued is knowing what is worth asking in the first place. A machine will answer almost anything. It will not tell you that you're asking the wrong question, and it never will, because it has no idea what you're actually trying to do.

Real work with real consequences. Things where somebody is genuinely waiting, or genuinely disappointed. You cannot learn judgement from an exercise where nothing is at stake.

People, deliberately. Time with grandparents, with neighbours, with people who are not like us and who require some effort to understand. Reading a person is a skill, it is learned by practice, and it is close to the top of the list of things that will still matter in twenty years.

And I have stopped pretending I know what they'll do for a living. Nobody does. Anyone confidently naming the safe careers of 2045 is guessing. What I can do is send them into it as people who notice things, who can tell when something is off, who can be trusted, and who are not frightened of the tools.

This is a prediction rather than a fact, and I'm making it as someone who watches markets for a living.

I think we are about to see a flood of micro schools.

Small setups. A handful of children. Built around how a particular child learns rather than around how the average child is assumed to learn. Some run by teachers who have left the system, some by parents who started with their own children and found other families asking to join, some by entrepreneurs who spot the demand before the sector does.

The reason is not that mass schooling was badly designed. It is that it was well designed for a different economy. Standardised schooling was built to produce people who could follow instructions accurately and reliably at scale — because that is what factories, and later offices, needed. It did that job properly for a century.

That economy is the one being absorbed fastest. Following instructions accurately is precisely the checkable work machines now do without tiring. A system optimised to produce reliable instruction-followers is optimising for the wrong thing, at exactly the wrong moment.

What replaces it, I think, looks less like a smaller classroom and more like an inversion: the child as the curriculum, rather than the child fitted to one. Instead of pulling a child into the box and measuring how well they fit, you start with who they actually are and build outward.

And I want to say this part plainly, because it's easy to assume otherwise: this does not have to be an expensive, elite thing. A micro school is a room, a few children, an adult who pays attention, and access to tools that are now nearly free. Those are conditions that exist in a township as readily as anywhere. The barrier to personalised education used to be the cost of one adult per handful of children. AI does not remove that adult — but it does remove a great deal of what used to make that adult's job impossible at small scale: the planning, the material preparation, the differentiation, the admin.

If you are an entrepreneur reading this and looking for where demand is moving, I would look here.

There is a particular version of this that I want to say directly.

The knowledge that makes a business work in a township has always been treated as informal. Understanding how a stokvel actually operates. Knowing why a customer's silence this month means something different from last month. Knowing what a recommendation from the taxi rank is worth. Knowing when to extend credit and to whom — and being right.

None of that has ever appeared on a CV. It doesn't come with a certificate. It has been quietly regarded as what people do when they didn't get the education.

That knowledge now sits in the column that does not get automated. Not because it's precious, but for a hard structural reason: there is no way to check it against anything except the community it came from, so there is no way for a machine to practise it. It's the same argument I make in AI plus kasi wisdom.

If you have it, you should stop apologising for it. And you should be teaching it deliberately to your children, not assuming they'll absorb it by being nearby.

Not much, honestly. Ten-year-olds don't need a lecture about labour markets.

But when it comes up, it's roughly this: the machine is very good at the parts that can be marked right or wrong. It's not good at people, and it doesn't know what matters. Get good at those, stay curious, learn to use the tools without being impressed by them.

And be someone others can rely on. That has never once gone out of date, and I don't think it's about to start.