A father looks at his son's working life, four employers in six years, income from two of them at once, no title that maps to anything, and reaches for the diagnostic vocabulary of 1985: unfocused, uncommitted, not building anything. The vocabulary feels neutral. It is not. It is a measurement instrument, and every instrument encodes assumptions about what it measures. This one assumes ladders exist, that rungs are load-bearing, and that time served converts reliably into safety. All three assumptions are now empirically checkable, and the checks are coming back negative.
The bottom rung is being sawed off
The clearest evidence concerns exactly the career stage the old advice treats as sacred: the entry level. Brynjolfsson, Chandar and Chen, using large-scale ADP payroll data, found that workers aged 22 to 25 in the occupations most exposed to AI have suffered a relative employment decline of roughly 16 percent, while experienced workers in the same occupations, and young workers in less-exposed fields, held stable or grew. The adjustment is happening through hiring, not wages, and it concentrates where AI automates work rather than augments it. The study's title calls these young workers canaries in the coal mine, which is precise: the ladder still stands, for now, for those already on it. It is the bottom rungs that are disappearing.
To an industrial-era instinct, a 24-year-old who cannot land the classic first job looks like a failure of hustle. The payroll data says something else: the door itself moved. Judging the applicant by the old door's location is measurement error, administered by the people who love them.
Churn is the system now, not the symptom
The second recalibration concerns stability. The World Economic Forum's employer survey across 55 economies projects that 22 percent of jobs will be disrupted by 2030, with 170 million roles created and 92 million displaced, and nearly 40 percent of the skills required at work changing in the same window. One can hold reasonable doubts about any forecast, but the direction is not in dispute among the people doing the hiring. In such a market, the worker who retrains sideways every few years is not flighty; they are tracking the terrain. The worker who sits still out of loyalty is accumulating a risk the old instruments score as virtue.
Meanwhile the tools themselves scramble the seniority signal. In a field experiment with 5,172 customer-support agents, Brynjolfsson, Li and Raymond found generative AI raised productivity about 15 percent on average, with gains up to roughly 34 percent for novices and least-experienced workers, and almost nothing for the most experienced: the machine, in effect, distributes the veterans' accumulated know-how to the newcomers. What forty years of tenure once uniquely conferred is now partially downloadable. Judgment remains scarce; time served, as a proxy for it, is depreciating.
A word from before the ladder
The strange comfort is that the deepest critique of career-as-ladder predates AI entirely. Ricardo Semler ran his companies on six-month budgets and treated any five-year plan as a sign of inattention; having built businesses without venture capital, he never planned beyond three years, on the grounds that three years is as far as anyone can see. And he liked to point at the etymology: career comes from the old word for a cart track. A career, literally, is a rut, one groove, pre-cut, followed to the end because the wheels fit it.
That framing converts the generational argument. The parent defending the forty-year single-track career is not defending prudence against chaos; they are defending one historically specific rut, built for vehicles that are no longer manufactured, against people navigating open ground. Open ground is genuinely harder, and more can go wrong. But the rut is not available in the old gauge, and steering by it produces advice that is confidently, lovingly wrong.
Retooling the instrument
What should a senior evaluator measure instead? The data suggests the questions. Not “is the job stable?” but “is the work augmentable?”, since the augmentation-automation line, not the industry label, is what predicted who kept their jobs. Not “how long have you been there?” but “what did you learn to do this year that you could not do last year?”, since skill turnover is running at decades-per-decade speed. Not “when do you get promoted?” but “who vouches for your judgment?”, since judgment is the one asset the studies consistently show machines redistributing least.
The irony is generous, in the end. The elders best equipped to judge AI-era careers are those who remember that their own rules were once new, adopted because conditions demanded them, not because they were eternal. The instinct worth transferring was never the ladder. It was the older one underneath: read the conditions, then build the rule.
Sources
- Brynjolfsson, E., Chandar, B. & Chen, R., Stanford Digital Economy Lab (2025). Canaries in the Coal Mine? Six Facts about the Recent Employment Effects of AI. Workers 22-25 in the most AI-exposed occupations show ~16% relative employment decline; experienced workers stable; losses concentrate where AI automates rather than augments. https://digitaleconomy.stanford.edu/publication/canaries-in-the-coal-mine-six-facts-about-the-recent-employment-effects-of-artificial-intelligence/
- World Economic Forum (2025). Future of Jobs Report 2025. Employer survey across 55 economies: 22% of jobs disrupted by 2030 (170M created, 92M displaced); nearly 40% of required skills change by 2030. https://www.weforum.org/publications/the-future-of-jobs-report-2025/
- Brynjolfsson, E., Li, D. & Raymond, L., Quarterly Journal of Economics (2025). Generative AI at Work. 5,172 support agents: ~15% average productivity gain, up to ~34% for novices, near zero for the most experienced; AI diffuses veterans' practices to newcomers. https://academic.oup.com/qje/article/140/2/889/7990658