What the reassurance skips
The standard reassurance about AI at work is that it takes the routine tasks off your desk so you can spend your time on the interesting ones.
Mostly true. It skips something.
The routine tasks were the training. Nobody learned to read a contract by having judgment about contracts. They learned it by summarizing forty of them badly and being corrected.
The numbers are now specific
Brynjolfsson, Chandar and Chen track payroll records for millions of US workers and find employment of 22-to-25-year-olds in AI-exposed occupations sitting 19% below where it would be had it kept pace with their less-exposed peers. A year earlier that gap was 15%.
Two things it is not. Not economy-wide job loss — experienced workers show no comparable gap. And not people being let go: the decline runs almost entirely through hiring that did not happen.
Now put the job ads next to it
PwC, reading more than a billion advertisements across 27 countries, finds AI-exposed entry-level roles are seven times more likely to ask for traditionally senior skills — judgment, leadership. Postings for those roles grew 35% since 2019 while other entry-level roles fell 10%. That is what employers are asking for, not what they hired.
Read together, that is a squeeze rather than a shortage. The tasks that built judgment are being absorbed, and the roles still open ask for judgment at the door.
The part that belongs to whoever runs a team
That routine work was never only output. It was a training program, paid in work instead of dollars, and nobody ever had to approve it. It arrived free with the org chart.
It does not anymore. And organizations have mostly not moved that line into a budget, a curriculum, or anyone's objectives. The subsidy was withdrawn, the requirement went up, and the plan is largely that somebody else will handle it.
So if your entry-level posting asks for judgment while your workflow removes the work that used to produce it, that is not a hiring standard. It is a bet that someone else trained them.
The version of this that is actually actionable is smaller than it sounds. It is not a graduate scheme. It is noticing, task by task, which handovers to AI were removing drudgery and which were removing a rung — and keeping a few of the second kind on a person's desk on purpose, with the correction still attached.
Nine hands-on sessions that build the skill by doing the work, not by collecting prompts: See the series →
Keep reading: Half of leaders already see their teams losing skills to AI · Codified vs Contextual
This is the thinking behind a Trusted AI Culture.
Sources: Brynjolfsson, Chandar & Chen, "Canaries in the Coal Mine? Six Facts about the Recent Employment Effects of Artificial Intelligence," Stanford Digital Economy Lab (August 2026 update), ADP payroll administrative records. PwC, "2026 Global AI Jobs Barometer," more than one billion job advertisements across 27 countries and territories.