A few months ago I wrote that everyone is a five-star developer now. That the gap between juniors and seniors is shrinking because the tools got so good. I buried the most important part at the very end of that post and called it "the catch."

Turns out the catch was the whole story.

Not everyone is a five-star developer. Nobody became one because of AI. What actually happened is something else, and I think the honest version is more interesting than the optimistic one.

AI is leverage, not skill

If you've ever looked at trading, you know what leverage does. Trade with 20x leverage and every move gets multiplied by twenty. A good trade doesn't just win, it wins big. A bad trade doesn't just lose. It liquidates you. Fast. 📉

Nobody would say leverage makes you a better trader. It makes you a bigger trader. The direction still comes from you. The tool just adds zeros.

That's what AI coding tools actually are. Not a skill upgrade. A position size upgrade.

If you're a good developer, you're now moving at 20x. You ship in an afternoon what used to take a week, and it's your quality, multiplied. But if your instincts are bad, you're not producing slightly worse code than before. You're producing bad decisions at industrial scale. Ten thousand lines of plausible-looking architecture that nobody understood, reviewed, or questioned. Merged by Friday.

The junior with AI doesn't converge toward the senior. Both got leverage. The distance between them didn't shrink. It got multiplied too.

Your ceiling is what you can judge

Here's the mechanism, and it took me embarrassingly long to see it clearly.

AI produces code at the level you can evaluate, not at the level you can write. Those used to be roughly the same thing, which is why the confusion is so easy. Writing was the bottleneck, so writing ability was a decent proxy for everything else.

That proxy is dead. The model will happily produce code far above your writing ability. But it will also happily produce confident, well-formatted, plausible garbage, and it presents both with exactly the same tone. The only thing standing between those two outcomes is your judgment.

A senior looks at the output and sees the missing transaction boundary, the retry loop that will hammer a dying service, the abstraction that will collapse the moment requirements change. A junior sees green tests and clean formatting and hits merge. The tool didn't fail them. It did exactly what leverage does: it executed their judgment, at size.

Your skill ceiling didn't go away. It just moved. It's a review ceiling now.

What I actually got wrong

The old post said the floor has been raised. I still believe that, mostly. Someone who understands their domain deeply can now build things that were out of reach before, and that part is genuinely great.

What I got wrong is the shape of the change. I described AI as an equalizer. It's an amplifier attached to whatever you already are. Equalizers compress the range. Amplifiers stretch it.

And a stretched range has an ugly bottom end. The mediocre developer of 2023 wrote mediocre code slowly, and the slowness was a safety feature. There was only so much damage one person could type. That limit is gone. The same person now ships mediocre decisions at a pace no code review process was designed for, and it all looks professional, which makes it strictly harder to catch than the honest mess of the pre-AI era.

The uncomfortable career advice

If AI is leverage, the question "how do I get better with AI tools" is the wrong question. Prompting is easy. Any tutorial covers it in an afternoon.

The right question is the old one, the one that never stopped being true: how do I get better at judging software? Reading code critically. Understanding failure modes. Knowing what correct looks like in your domain, not just what working looks like.

Because that's the multiplicand now. Every point of real skill you gain gets multiplied by twenty. And every gap in your judgment does too.

Traders figured this out a long time ago, usually the expensive way: leverage doesn't change whether you're right. It changes how much being wrong costs.

Welcome to 20x. I hope you're right. 📈