The most common worry these days is some version of “will AI take my job?” Even programmers aren’t exempt. As the AI that writes code keeps improving, there’s a steady drumbeat of predictions that developers will soon be unnecessary.
There’s an interesting rebuttal. Josh Comeau, who teaches web development, recently pointed out that handing different people the same AI tools produces wildly different results. A skilled developer can ride the tool to several times their usual output, while someone with less experience can reach a passable prototype and then hit a wall. There’s a whole style of “vibe coding” going around — building software by prompting the AI without ever reading a line of the code — and the people doing it often get stuck, flail for hours, and finally open the file to fix a single line in thirty seconds. The tool on screen is identical; what separates the outcomes is the skill of the person holding it. Lacing up Michael Jordan’s sneakers won’t let you dunk.
Comeau also names why we tend to overrate these tools. Because the AI talks to us like a person, we treat it less like a hammer or a calculator and more like a clever colleague. So when an expert wields it well and gets a great result, we hand the credit to the AI rather than the expert. In his telling, AI multiplies an expert’s skill; it doesn’t replace it.
It’s a fair observation. But the argument quietly blends two different questions. One is “is this kind of worker still needed?” The other is “how many of them are needed?” The fact that an expert becomes two or three times more productive speaks only to the first. What people are actually afraid of is the second.
And rising productivity, on its own, doesn’t even settle whether the number of jobs grows or shrinks. The automated teller machine (ATM) is a good example. When ATMs spread in the 1990s, everyone assumed bank tellers would vanish — but because each branch became cheaper to run, banks opened far more of them, and the number of tellers actually rose for a stretch. The economist James Bessen documented this: tellers per branch fell from roughly 21 to 13, yet total teller employment kept climbing. It wasn’t until mobile banking took hold in the 2010s that teller numbers finally began to fall. The same technology grew the job for a while and then cut it.
The premise that experts will always handle AI better isn’t a permanent law either. Right now there’s a wide band of judgment the AI can’t manage on its own, so the people who can fill that gap stand out. But as the AI narrows that gap, the skill premium narrows with it, and more people can produce similar results. “Skill is your moat” is less an eternal truth than a description of where we happen to be standing.
So “will AI eliminate jobs?” really splits in two: will the work disappear entirely, or will the number of seats doing it change? Comeau’s answer addresses the first; the second is still open. The very same gain in productivity can push it either way.