AI is already changing who gets hired in Britain, and the first people to feel it are those at the bottom of the career ladder. A study by Dr Bouke Klein Teeselink of King’s College London found that firms whose work is most exposed to AI cut their junior staff by 5.8 per cent on average.
more
Klein Teeselink drew on millions of job postings and LinkedIn profiles dated 2021 to 2025. He found that ChatGPT’s public release in November 2022 marked the turning point.
What the UK data shows
Across the exposed firms, total employment fell by 4.5 per cent on average. Almost all of that fall came from junior positions. These firms were also 16.3 percentage points less likely to post a new vacancy.
The damage was uneven by job type. Software engineers and data analysts saw the steepest drops in listings, while sales roles rose slightly. Klein Teeselink read that as a sign that jobs built on direct contact with people are holding up better.
Pay told a similar story. High-paying firms cut employment by 9.6 per cent, while low-paying firms saw almost no change. Average pay inside the exposed firms rose by more than £1,300, because the loss of junior jobs shifted the mix towards senior staff.
Why the first rung matters
Klein Teeselink warned that junior roles act as “training grounds” where workers learn by doing harder tasks in sequence. Without them, he said, firms may struggle to grow senior staff from within. He added that policy-makers would need to watch the effect closely.
Is AI replacing jobs or changing them?
Suraj Srinivasan, a professor at Harvard Business School, draws a line between two effects. Automation removes the need for a person on a task. Augmentation makes a person better at it.
His research on new job adverts found that postings for automation roles have fallen since November 2022. Postings for augmentation roles have risen by about as much, or more. He cautions that the line moves, because what counts as augmentation today may become automation tomorrow as AI improves.
Goldman Sachs Research found the same split in US data. Its economists estimate that AI cut monthly payroll growth by roughly 16,000 jobs over the past year and raised unemployment by 0.1 percentage point. In jobs where AI works alongside people, payroll growth rose by about 9,000 a month.
The bank’s team points to Jevons paradox to explain the second figure. If AI makes a worker more productive, the cost of each unit of output falls, and demand for that output can rise enough to add jobs. The same analysis found that the weaker hiring falls mostly on younger, less experienced workers.
Telephone operators, insurance claims clerks and bill collectors face the highest risk of substitution on the bank’s measure. Education workers, judges and construction managers have the most to gain from augmentation.
How much of the UK workforce is exposed?
The UK government’s Department for Science, Innovation and Technology cites International Monetary Fund estimates that around 70 per cent of UK workers are in occupations with tasks AI could perform or enhance. The figure for the US and other advanced economies is around 60 per cent. The gap reflects the weight of services in the UK economy.
Exposure is not the same as job loss. The department says such indices have not been extensively validated against real outcomes, so they are indicative rather than precise. It also states that the available evidence does not yet answer many of the questions that matter most for policy.
The department notes that the length of tasks AI agents can complete on their own has roughly doubled every seven months in coding, cybersecurity and research. It is less sure that the same pace will carry over to other fields.
What job seekers are running into
AI is also changing the hiring process itself. Srinivasan says candidates can now use it to tailor CVs and cover letters to a job description, and HR departments report a many-fold rise in applications.
His concern is that weaker candidates can look strong on paper, which buries the best ones. Some employers now ask for video clips, essays or code samples, which adds cost for both sides.
The result is a picture with no single answer, and Srinivasan says the line between automation and augmentation keeps moving.
