Posted by 
For decades, writing good code and judging good code looked like the same skill. If you could build it well, you could usually tell when someone else hadn’t. The two abilities lived so close together that nobody bothered separating them. Seniority was mostly a proxy for having done enough of the doing.
That proxy is breaking, inside a lot of engineering teams right now.
AI has absorbed a meaningful share of the doing. Boilerplate, first drafts, routine implementation, the kind of work that used to take a capable engineer a few hours now takes a few minutes and a prompt. What hasn’t gone anywhere is the harder question sitting underneath all of it.
Is this actually right. Will it hold up under load. Is this the correct approach for this problem, or just a plausible-looking one. That is a judgment question, not an execution question, and AI is nowhere near as reliable at answering it as it is at producing the first draft.
So the two skills that used to travel together are coming apart. And the one that’s left, mostly untouched by the last two years of progress, is the one nobody built any ability to measure.
Most engineering organizations have never had to measure it, because until recently, judging the work and doing the work were the same person doing both, at the same time, almost invisibly.
That invisibility is over. If writing code is a smaller part of the job now, the honest question is what you were actually promoting people for all along. In most companies, “senior engineer” has meant someone who writes a lot of good code, fast. But it was just a convenient proxy that happened to correlate well enough with the real thing to go unquestioned for years. The real thing was always the judgment. Fast, high-volume production just used to be the only visible evidence of it.
Teams that figure this out first will stop measuring engineers by how much they ship and start looking for something harder to see. Can this person tell good work from work that merely looks good. Can they catch the plausible-but-wrong answer before it ships. Can they explain, specifically, why one approach beats another, rather than just producing one that works.
That is a rarer skill than fluent code generation. And it always was. It’s just been sitting underneath the thing everyone was actually measuring.
AI didn’t create a new kind of engineer. It removed the disguise the old kind of engineer was hiding behind.