Learning by Doing Is Getting Rare — and Expensive

The fastest way to learn anything has always been the same: do the thing, fail at it, figure out why, try again. AI lets you skip the failing part. That feels like a gift. It acts like a debt.

What failure was actually teaching

When you tried to deploy your first backend and it broke in six ways, you did not learn six facts. You learned a map. You learned which pieces touch which other pieces, which errors come from which causes, what the shape of "something is wrong" feels like. That map is the thing senior engineers have that juniors do not. It cannot be read. It has to be earned.

When an AI fixes your deploy in nine seconds, you get a working deploy and a smaller map. Both are real. One of them is invisible until you need it.

Tutorials were never the bottleneck

Before AI, people who watched fifty tutorials and built nothing were already failing at learning. Tutorials give you confidence, not skill. Doing gives you skill. The only reason "learning by doing" is now a premium skill is that AI made the "doing" easier to fake.

If you cannot tell whether you understood something without AI help, you did not understand it. That used to be obvious when you had to actually build the thing. Now it takes deliberate effort to notice.

The weekly practice that keeps real skills sharp

• Once a week, build a small thing end-to-end with zero AI assistance
• Keep it small enough to finish in one sitting. A tiny CLI, a one-page site, a utility function with tests
• When you get stuck, struggle for 15 minutes before you look anything up. The struggle is the point
• After finishing, write a three-sentence "what I now understand differently" note
• Four weeks of this compounds more than four weeks of AI-assisted features

Why "use AI for everything else" is fine

This is not an anti-AI post. I use AI every day. I rely on it for the 80% of my work that is boilerplate, translation, first drafts, lookups. That is fine. But I protect a slice of my week where I do the thing alone, because that is the slice where real skills keep compounding.

The goal is not to use less AI. The goal is to not lose the muscles that make AI useful in the first place. A senior engineer with AI is unstoppable. A senior-looking engineer who never built one of anything is a liability.

The career implication

Over the next five years, "I learned by doing" will quietly become a paid qualification. Companies will care less about which bootcamp you took and more about what you have shipped without help. That is already starting. Protect the habit now, not after the market forces it.

If you cannot tell whether you understood something without AI help, you did not understand it.