For the past couple of years, most predictions about AI and jobs pointed at managers, lawyers, and financial analysts as the ones most exposed. The logic behind those rankings was simple: list every task inside a job, then count how many of those tasks a computer could technically do. Professional roles scored high on that list.
Economist Gad Levanon decided to check that logic against real numbers instead of task lists. Using two years of government employment data, he found that management, engineering, and legal jobs kept growing, and in some cases sped up. Clerical work, records clerks, bookkeepers, customer service staff, shrank quickly over the same period.
He also ruled out an easy explanation. It is not that clerical jobs simply sit inside industries that are shrinking overall. Levanon compared each occupation's growth to the average growth across the industries where it is found, and the pattern held up: clerical work is shrinking inside the same sectors where management is expanding.
The explanation several experts point to is the difference between a task and a responsibility. AI can now handle a lot of the individual steps inside a clerical job, entering data, filing records, answering routine questions. It cannot take on the accountability that comes with a manager's or a lawyer's title, so companies keep the person even after software takes over part of the workload.
This finding also runs against a widely repeated prediction. Research firm Gartner has forecast that a fifth of companies will use AI to flatten their structure by the end of 2026, cutting more than half of current middle management jobs. So far, the employment numbers do not show that happening. Management keeps adding people, not losing them.
Other research lines up with the clerical story. A Stanford Digital Economy Lab study tracking payroll data found that employment for workers between 22 and 25 years old in the most AI exposed jobs, largely customer service and entry level technical roles, now sits about 19 percent below where it would be without AI. Separately, outplacement firm Challenger, Gray and Christmas counted close to 55,000 layoffs in 2025 that employers directly blamed on AI, concentrated in back office and support work rather than management.
Government projections point the same direction for the rest of this decade. Roles like data entry clerks, bank tellers, and postal clerks are among the fastest shrinking jobs in the country, with some expected to lose close to a third of their workforce by 2030.
The lesson for anyone running a business is not that AI has spared professional jobs. It has simply not automated the part of those jobs that involves owning a decision. If your team includes people whose entire job is a repeatable, rules based process, plan for that work to keep shrinking over the next few years no matter what the job title says.
And if you manage people yourself, the safest move is to spend less time on tasks a tool can already finish and more time on the judgment calls that carry your name on them. That is the part of the job nobody has figured out how to automate yet.