In the past few weeks I’ve been talking a lot about how cool it is to make things with AI — but maybe what’s even cooler is how it changes the way we learn.
Have you seen those “leaked MrBeast files” about his YouTube retention tricks? For example: in the first three seconds you have to show the most visually striking moment in your video. Maybe you’re lifting a house with a crane, maybe you’re releasing a thousand balloons into the sky. You create a hook that makes the viewer stay: “How on earth did the author end up lifting a house with a crane?!”
That’s basically the whole attention economy: hook them in the first three seconds. The new match-three mobile game shows you a desperate mom thrown out of her house by her evil husband. A Michael Bay trailer starts with an explosion. The first ten pages of a novel drop you into the middle of the action — you don’t yet understand what’s going on, but you feel it must be important. An elegant old gentleman buys every peach from a street vendor in a Netflix teaser and you just have to know why.
Before they give you the fundamentals, they first show you why you’ll need them.
Learning in school was the opposite. In algebra class I was nauseated by dozens of identical exercises: multiply this, expand that, apply the rule we just explained. You had to grind through an entire lesson of primitive tasks before getting to a single one where you actually had to think. And that’s the sad part — algebra really does have tons of real-life applications, but you only glimpse them much later.
Foreign language classes were no better. You don’t learn to write like Dickens or Nabokov. Instead you’re forced to write endless sentences about “the London is the capital of great Britain” or “the weather is calm.” (Ask any former student: these phrases are burned into our memory not because they were useful, but because we wrote them a hundred times.)
That was learning before AI.
Now I get to start with the coolest stuff. I tell the AI: “Let’s build a website where I can press a button and see a random cat photo from the internet.” A minute later I have it. Then I say: “Okay, add a switch so I can toggle between cats and dogs.” It works. At the start, working with AI feels like pure magic: setting tasks is easy, execution is instant, and the result is tangible. I just lifted a house with a crane, because I could.
Over time the creativity shrinks, because I want more specific things — and “specific things” aren’t always easy for an AI. I have to figure out how to phrase requests. I have to ask one AI to write prompts for another (recently, ChatGPT wrote me a prompt for Bolt that was over 15,000 characters). I have to feed it whole folders of code and say: “Explain to me what each of these files does.”
Yes, that’s less exciting than pressing the magic button. But now I already have a working foundation, I know where to go, and most importantly — I didn’t waste energy on the routine. The endless trivial exercises are done for me by the AI; I only spend my effort on solving new problems, each time new ones. Because once I’ve solved a problem once, I can automate it. Maybe not on the second try — but certainly long before it gets boring.
In the end, I learn the details only when they matter. I don’t know what to compare it to… Maybe to being a catch-and-shoot player in basketball, playing in a system so good that you never even need to run. And then, if the coach suddenly wants you to become a 3-and-D guy, you can shift focus instantly.
So my work turns into learning, and learning instantly folds back into work. It really is a brave new world.