AI's Effect Shows Up First at the Bottom of the Career Ladder

When people ask which jobs AI will replace, they often picture a whole occupation disappearing at once. Looking at the data published through 2026, the reality seems more complicated, and the effect appears to be distributed unevenly.

The clearest declines seem to fall not on occupations as a whole, but on their entry rung — the junior, first-year version of the work that newcomers usually start with.

The same job title, a different picture by career stage

Stanford University's 2026 AI Index (published April 2026), for example, reported that employment for software developers aged 22–25 fell by roughly 20% from 2024, while employment for developers aged 30 and over continued to grow. Those figures suggest that the situation can differ within the same job title depending on where someone sits on the ladder. Related work led by Erik Brynjolfsson also found a relatively larger decline among younger workers in the occupations most exposed to AI.

The way roles shrink also has a pattern. In many cases it is not announced as "we replaced this with AI"; instead, positions are quietly left unfilled after someone leaves. Because it looks from the outside like a hiring slowdown, the change can be hard to see.

A similar direction shows up globally. The World Economic Forum estimates that around 92 million roles may be displaced by 2030 while around 170 million are created. The net figure is positive, but the displaced roles are said to skew toward clerical and routine work, and the created ones toward specialised roles requiring training — so it may be worth remembering that the two do not map neatly onto each other.

Why this matters for reading a score

Attaching a single score to an occupation tends to average out this within-occupation difference. For that reason, we think our scores are best read not as "this occupation disappears" but as "the routine, entry-level parts of this occupation are relatively more exposed." Even where a score is high, it is closer to a signal that the repetitive parts of the work — often what newcomers are handed first — are more exposed, rather than a verdict that the profession ends.

That is also why we treat a score as a tool for comparison, not a judgement about any individual. A senior translator handling contractual or specialist nuance and a first-year translator handling high volume can sit in quite different positions under the same job title.

Rather than checking whether your occupation is "on the list," it can be more useful to separate which parts of your own work are routine, and which depend on judgement people still hold.

The Stanford researchers also raise a longer-term concern: if a generation struggles to take its first step in an AI-exposed field, the experience that protects senior workers may not accumulate, and the same fields shedding juniors today could face talent shortages later.

If there is one thing worth taking away, it may be this: rather than checking whether your occupation is "on the list," it can be more useful to separate out which parts of your own work are the routine, entry-level parts — and which parts depend on the judgement, accountability, and human relationships that AI still finds hard to take on. Our scores are meant to help locate that line.