AI

There is only one way to learn AI

And textbooks ain't it

Kieran Ball 3 min read

I’ve been thinking this for a while. Structured educational content is dead when it comes to AI.

I don’t mean that for every subject. If you’re learning history, a textbook is still a beautiful thing. The Roman Empire isn’t going anywhere. Maths, philosophy, classical literature, the basics of biology, all of these deserve stable books and a curriculum someone wrote ten years ago is still mostly right.

AI is just different.

The tools keep shifting under my feet. What was the “right way” in February is the wrong way in April. A YouTube tutorial filmed six months ago is showing you a button that no longer exists, in an interface that’s been redesigned, using a model that’s been deprecated. Any 30-lesson course on this stuff would be half embarrassing before the last video had even uploaded.

And AI itself is a brilliant teacher. You open Claude or ChatGPT. You try to make it do something. It does the wrong thing. You ask why. It explains. You try again. And again. By the end of an evening you’ve understood more than any course would have got you to, because every step has been you, doing it, in the version of the tool that exists today…

That’s the loop. The tool teaches you, by responding to what you do.

I’ve got two kids under 10 and I think about this a lot when I look at them. Nobody really knows what the job market will look like in 15 years. The careers I might nudge them toward today might not even exist by then, and a load of new ones will.

The one thing I’m fairly sure about is that the people who thrive will be the ones who can pick up a new thing, try it, fail at it, try it again, and not get embarrassed about any of that.

Trial and error is the actual skill.

It’s not subject-specific. It works for any tool, any field, anything new that turns up tomorrow. It stops you being a perfectionist. It gives you permission to fail and keep going.

That’s also why AI happens to be such a good teacher right now. Every prompt is an experiment. Every wrong output is data. The whole format trains the muscle.

A lot of people still feel guilty about learning this way. They want modules. They want a certificate. With AI, those things won’t give you what you think they will. The subject has just stopped slowing down long enough for any of that to work.

If you’re struggling to get started, my advice is simple. Choose a project or idea you’re passionate about, and commit to building it.

Join 4,500+ no-code builders

I send a regular email with all my latest experiments, learnings, tips and hacks.