Nestor G Pestelos Jr · Writing

The Career Ceiling I Was Worried About Already Existed

Published August 14, 2026

TL;DR

AI-driven fears of a professional-athlete-shaped career ceiling turn out to predate AI by decades. Current data shows the ceiling is landing on junior engineers now, not seniors: cold comfort, since they're the ones who most need to hear it.

A pro athlete's career has a real ceiling, usually in their mid-thirties, not because they get worse at the game but because a body only holds up so long. Sean Goedecke argued software engineering might be entering that same shape, driven this time by AI eroding the skills unaided practice used to build. It worried me enough to check.

The worry isn't new

The worry isn't new. In 2009, John Fuex made structurally the same argument, with zero AI in it: technology shifts every decade or so, resetting the value of a veteran's experience without lowering the cost of employing them. Ten extra years mastering a dead stack isn't worth a premium salary; it's a junior's salary at senior overhead. Fuex's advice, thirteen years before ChatGPT existed, was the advice implicit in "plan for a ceiling": build a plan B before the gap between cost and relevance opens up.

That changes what this piece is about. What matters isn't whether AI is doing something new to my profession, but whether it's a second version of an old problem, or just a new coat of paint on an old worry. The way to tell the difference is to see who it's actually hitting.

Who it's actually landing on

So I went looking, not hypothetically. Stanford's Digital Economy Lab ran ADP payroll data across AI-exposed occupations, and the finding cuts against the worry that scared me. Employment for 22-to-25-year-old developers has fallen sharply; older cohorts have held or grown. The mechanism matters: declines cluster in work AI substitutes for outright, and employment holds or rises in work AI can only complement, especially for experienced people.

AI isn't sparing senior engineers out of respect for tenure. Senior work tends to be the kind AI can extend, not replace: the ambiguous ticket, the architecture call. Junior work tends to be the kind AI can just do outright: the well-specified ticket, the boilerplate endpoint.

If this ceiling is real, it's landing on the newest people in the field right now, not on people like me, the ones AI's boosters keep promising will be fine because AI supposedly just makes everyone more productive. If this were just the old economics repainted, it would still be landing on aging seniors; a different victim points at a different mechanism doing at least some of the work. So far, the job loss and the productivity story are landing on different age groups.

One split, not two

I have a live example of the same substitute-versus-complement split from somewhere Goedecke's piece didn't go. I know framework maintainers in the Rails community, people whose whole standing is deep, narrow, hard-won knowledge of one system, and none of them are worried about AI. If anything, their position looks reinforced by it: their work is exactly the kind AI complements instead of substitutes for.

Arora made a similar split about AI products: depth builds a moat, breadth commoditizes. The career version of that logic follows the same shape: a competitor can copy breadth but has to rebuild depth from scratch, and junior engineers are disproportionately the breadth case, by definition, without the years to build the narrow expertise a Rails maintainer has. "AI hits the young" and "AI hits generalists, not specialists" may not be two findings. They may be one, described from two angles.

Stanford calls its own six facts "early, descriptive indicators... rather than causal estimates." I couldn't find a study that measures expertise depth against age directly. What I have is a strong correlation and a personal anecdote that fits it, not a proof.

I also notice how convenient this conclusion is for someone in my position: senior, not junior, the ceiling landing on someone else's head. That doesn't make the data wrong, but it's exactly the kind of finding I should be more suspicious of, because I like it. That same suspicion sent me back to check the rest of the analogy this whole piece is built on.

What the analogy gets wrong

It has two limits worth naming honestly. First, the famous stat about athletes going broke, that most NFL and NBA players lose everything within a few years, traces to a 2009 Sports Illustrated piece and has been called fabricated elsewhere, with no basis given for the original numbers. The real, peer-reviewed number, from a 2015 study, is gentler: about one in six NFL players filed for bankruptcy within twelve years, and earnings barely moved that number.

Second, an athlete's ceiling is a body that stops working. A skill that atrophies from disuse may be more forgiving than that. My guess is it comes back faster than it was built the first time, though I don't have a study that proves it, and either way that reversibility is about the skill-fade worry I started with. It doesn't help someone who never got the job that would have let them build the skill in the first place.

What to actually do about it

Here I don't have a clean answer. What survives from the athlete comparison is the behavior, not the biology: earning more doesn't protect you from what comes after a career-ending event. The closest thing I found comes from a 2026 post on mid-career strategy: shift from execution to influence, stabilize income, protect a reputation. None of it was written with AI in mind, or tested against what AI actually erodes: the standing practice that keeps a skill current.

Construction eventually got cranes and dollies to offset the physical cost of the job. Nobody's built the equivalent for staying mentally engaged while an agent writes your code. I don't think that tool exists yet.

I started out worried about the wrong version of a real problem. The ceiling isn't new, and if AI is doing anything to it right now, it's moving earlier in a career than later. That correction is uncomfortable: the people who most need to hear "plan for a ceiling" are the twenty-three-year-olds nobody's worried about yet, reading pieces like this one written by people like me.

Sources