One study says entry-level jobs are shrinking where AI is strong. Another says firms that spend the most on AI hire more juniors. Both can be true. The view here is that neither should decide your hiring. Your own task list should.
- Stanford finds employment of 22 to 25 year olds in highly exposed jobs about 19% below peers in less exposed ones.
- Ramp finds heavy AI spenders grew entry-level headcount 12% over two years, though those firms are not typical.
- The call a business has to make is task by task, and a simple formula settles most of it.
Two sets of numbers landed this summer and they point in opposite directions. Researchers at Stanford found that young workers in AI-exposed occupations are working less than their peers elsewhere. Payments data from Ramp showed the firms spending most on AI adding junior staff faster than anyone. An owner reading both could freeze hiring or speed it up, and either move would be a guess. Skip both headlines. Start with the task list on your own desk, and ask who gets the first shift.
TL;DR
The Stanford gap comes from reduced hiring, not layoffs, and its authors call the patterns descriptive, not causal. The Ramp result describes fast-growing, engineering-heavy firms that were already different. Canada’s own data shows slower hiring in exposed jobs but no broad shift yet. The practical answer is to test each task for whether delegating it to AI saves time once you count the cost of checking, and to treat a junior seat as a decision about who runs the business in five years.
What does the Stanford study actually say?
It says that by June 2026 employment of workers aged 22 to 25 in the most AI-exposed occupations was about 19% below where it would be had it kept pace with similarly aged workers in less exposed ones. Experienced workers show no comparable gap.
The Stanford Digital Economy Lab paper by Erik Brynjolfsson, Bharat Chandar and Ruyu Chen uses ADP payroll data. It attributes the gap mainly to reduced hiring. The authors are careful, and the careful part is what most coverage dropped. They describe the findings as descriptive patterns, not causal estimates. The gaps shrink once education is controlled for, some trends began before generative AI, and the gaps are larger in ADP data than in national survey benchmarks. A circulating newsletter figure of 11% is wrong. The number is 19%, and it is a relative gap, not a count of lost jobs.
And the study that says the opposite?
It says something narrower. Ramp’s economics team joined spending data from more than 21,000 US firms with workforce data from Revelio Labs. Firms in the top third of AI spend per employee, around $30 a month per person in the first three months, grew headcount 10.2% over two years and entry-level headcount 12%. Low-intensity adopters saw no significant change.
The catch is who those firms are. Ramp notes that heavy adopters skew larger, more engineering-heavy, venture-backed and faster-growing. Gains began six to twelve months after adoption. A firm that was growing anyway and has money to spend on AI will hire juniors anyway, so the result flatters the spenders. The study compares them with a control group, but it cannot promise that your firm, with a different mix of work, would follow the same path.
What does Canadian data show?
It shows early movement and no broad break. A Bank of Canada staff note published in August found AI has not yet led to broad changes in the structure of the labour market, with the authors stressing they are early signals and their own views.
Within that, the Bank’s analysis found the job-finding gap between fully exposed and unexposed workers widened from -2.2 to -13.9 points, while the separation rate barely moved, from 0.0 to 0.1. In plain terms, adjustment is showing up as slower hiring, not layoffs, and young workers cluster in moderately to highly exposed jobs like customer service and sales support. Statistics Canada’s August survey muddies the picture. Youth unemployment was 12.9%, lower than 14.3% a year earlier, though youth employment fell by 19,000 that month. We previously wrote about the Bank of Canada’s view that AI’s payoff is years away, and this note fits that picture.
How should an owner decide whether to hire a junior?
Run the arithmetic on each task before you decide about the person. If a task takes T minutes by hand and it costs E minutes to explain it to an AI and check the result, delegating pays when the AI gets it right more often than E divided by T.
Take a 40-minute task. If explaining and checking costs 10 minutes, the AI has to succeed more than 25% of the time to be worth using, because you pay the 10 minutes every time and redo the work yourself when it fails. If explaining and checking costs 30 minutes, the bar rises to 75%. These figures are illustrative, not measured, and your own numbers will differ. Here is what the studies cannot tell you. Tasks that are easy to describe and quick to check clear the bar first, and those are usually the tasks juniors are handed. That is our hypothesis for why exposure and entry-level hiring move together, and it fits the Stanford pattern.
- List every task a junior would do in a typical week, with a time for each.
- For each, estimate how long it takes to explain and check an AI’s attempt.
- Keep the tasks where AI clears the bar and give them to the AI.
- Look at what is left. If a junior would spend most of their time on judgment, client contact and learning, the seat is a pipeline decision, not a task decision.
- Decide on the seat with a date to review it. Six months is long enough to see whether the arithmetic held.
What does the sceptic say?
The best counter-argument is that the arithmetic is too tidy. Checking AI output takes longer than people estimate, and juniors learn by doing the very tasks you are about to hand away. Skip the seat and you save money this year and it costs you seniors in five. Both points hold weight. IBM’s head of HR said in February that the company is tripling entry-level hiring, which suggests at least one large employer sees the pipeline risk.
The view here would be wrong if Canadian youth unemployment climbed back above last year’s 14.3% while hiring in exposed occupations kept sliding. That combination would say the slowdown is real and spreading, and the formula should be applied with more caution.
What to watch
- Statistics Canada’s next labour force release on October 9.
- Whether the Stanford authors publish causal estimates.
- Whether the Bank of Canada widens its analysis beyond job-finding rates.
Frequently asked questions
Is AI reducing entry-level hiring in Canada?
Bank of Canada research shows slower hiring in AI-exposed jobs but no broad change in the labour market so far. The authors call these early signals.
Should a small business stop hiring juniors because of AI?
Not on the strength of a headline. Test the tasks first, then decide whether the seat is a pipeline investment.
What is the Stanford 19% figure?
It is the relative gap in employment for ages 22 to 25 in highly exposed occupations versus similarly aged workers in less exposed ones, as of June 2026. The authors say it is descriptive, not causal.
The decision in one line
This week, write down the ten tasks you would give a new hire and put a number beside each. The seat will decide itself, or it will not, and you will know why.
Disclosure. This column is analysis and opinion. The task formula uses illustrative numbers. Sources were checked on October 2, 2026. For more on how we cover workforce data, see our earlier look at AI and Canadian jobs.
Written by the AI Magazine Canada team, reviewed by the AI Magazine Canada editorial team.
