An argument about how we learn

When AI can finish the work faster than you can learn to do it, how does anyone still get good?

AI deleted the grind that used to turn juniors into seniors. That's a gift to the engineers who already climbed — and a quiet crisis for the ones who haven't.

Output vs. understanding, over a career
Output with AI Understanding

Speed you can borrow. Judgment you have to build.

The thesis

The real cost of AI coding is young engineers.

For decades, engineers got good the same way: by doing the boring work. Reading the stack trace at 2am. Writing the boilerplate by hand. Getting stuck, staying stuck, and slowly building the intuition no tutorial teaches. The grind was the curriculum.

AI just deleted the grind. And for anyone who already climbed that ladder, that's genuinely great. A senior with judgment plus AI is a force multiplier — they know what to ask, they can smell when the output is wrong, they steer. For them the shortcut is pure upside.

But juniors don't have that judgment yet. They were supposed to build it — on exactly the tasks AI now absorbs. When the work gets done without the struggle, the code ships but the understanding never forms. You get output that works and an engineer who can't tell you why.

That's the hidden cost. It doesn't show up this quarter. It shows up in five years, when a generation can produce but can't diagnose, can prompt but can't reason — and there's no senior bench deep enough to catch what they miss, because we quietly stopped growing seniors.

We're not paying the cost of AI coding today. We're charging it to the engineers who haven't been trained yet.
A point of view

So how should you learn now?

01

Ship real work, not tutorials

The struggle was always the curriculum. Choose problems hard enough that you're forced to understand them — and resist the one-tap answer while you're still learning the shape of it.

02

Understand systems, not snippets

AI hands you the what for free. Your entire edge is the why — how the pieces fit, where they break, and what happens under load.

03

Use AI as a tutor, not a crutch

Ask it to explain, not just to answer. Make it walk you through the reasoning until you could have written the code yourself. Make it teach you out of a job.

04

Build judgment, not just output

Anyone can generate code now. The scarce, durable skill is knowing when it's subtly wrong — and having the taste to reject it.

In fairness

The strongest objection to all of this.

The counterargument

"Every tool got this same panic."

Compilers, IDEs, Stack Overflow, high-level languages — each one was going to rot a generation of programmers. Each time, engineers adapted by learning at a higher level of abstraction, and the field got stronger, not weaker. AI is just the next rung.

The reply

Those removed tedium. This removes the struggle.

A compiler spared you punch cards; it didn't think for you. What made judgment was never the tedium — it was the fight to understand. AI is the first tool that offers to skip that fight entirely. Skip it too early and the abstraction has nothing underneath it.

Keep reading

One essay when there's something worth saying.

No schedule, no filler. Just sharp writing on how engineers stay sharp. Unsubscribe anytime.

// no spam. one reader-first list.