BridgeWorks
We have been writing about AI and entry-level work for a while — what we rebuilt in our curriculum in April, and the broader question before that. Both were written largely on anecdote, because that was what existed. There is now real payroll data, and it deserves a careful reading.
What the research found
The Stanford Digital Economy Lab's paper "Canaries in the Coal Mine?" — by Erik Brynjolfsson, Bharat Chandar and Ruyu Chen — tracks employment across more than 730 occupations using ADP payroll records since late 2022. In an update published August 12, 2026:
- Employment for workers aged 22 to 25 in the most AI-exposed occupations now sits about 19 percent below where it would be had it kept pace with less-exposed peers — widened from 15 percent in the previous data vintage.
- Between November 2022 and June 2026, employment for that age group fell about 11 percent in the two most-exposed occupational groups, while rising about 10 percent for the same age group in the three least-exposed groups.
- Older workers in the same exposed occupations are largely unaffected so far.
Start with the part the headlines skip. The lab's own summary leads with "no widespread displacement." This is not a story about AI eliminating jobs across the economy. It is a narrow, sharp effect concentrated in young workers in specific occupations — software development, customer service, and similar roles.
The mechanism matters more than the number
Here is the finding we think is most important, and the one that changes what a program like ours should do.
The decline runs through reduced hiring, not increased separations. Nobody is being marched out. The junior role simply never gets posted. The summarizing, formatting, first-pass coding and basic research that used to be a new hire's first year is routed into a model, and a senior person plus a tool absorbs the work.
That distinction matters because the two situations look identical in an unemployment rate and require completely different responses. A layoff wave is a re-employment problem. A hiring freeze at the entry level is a pipeline problem — and pipeline problems are invisible until several years later, when there is no one with three years of experience because nobody got their first year.
Last week's August jobs report showed the same pattern in the aggregate: the information sector shed 23,000 jobs, close to triple its twelve-month average, in the sector with the highest AI adoption rate in the economy — roughly 39.7 percent, against a national average near 19.8 percent.
The distinction that decides it: automation or augmentation
The research draws a line that is genuinely useful for anyone choosing a training path.
Where AI automates a task outright — writing the boilerplate, handling the routine chat — entry-level hiring declines. Where AI augments a worker — helping them reason through a problem, checking their work, speeding up a step they still direct — employment holds steady or grows.
So the useful question is not "will AI touch this job?" It will touch nearly all of them. The question is "does the tool replace the beginner's task, or make the beginner better at it?" That is answerable job by job, and it is what we now ask when we evaluate a training track.
What this means for our participants
We are a workforce organization whose entire model is getting people onto a first rung. Research showing that first rung being removed in some fields is not comfortable reading, and pretending otherwise would be useless to anyone.
What we take from it:
- The trades and hands-on work look structurally better positioned. Not because they are immune to technology, but because their entry-level tasks are physical and site-specific. Construction added 22,000 jobs and health care 13,000 last month. Our construction program and support services track sit in that category.
- In exposed fields, the entry bar has moved up. Someone entering technical work now has to arrive already fluent with the tools, because the tasks that used to be the training ground are the ones being automated. That is precisely what we rebuilt the curriculum around in the spring, and this data says we should push it further.
- Apprenticeship models matter more, not less. When employers stop funding a beginner's first year through junior roles, the structures that still pay people to learn on the job become the main remaining path. We have made that case before, and this research strengthens it.
The honest bottom line
One study, one dataset, an effect visible over about three and a half years — and the researchers themselves are careful about causal claims, which we should be too, especially after getting a labor-market call wrong last month. Interest rates, sector-specific corrections, and post-pandemic normalization are all plausibly tangled up in the same numbers.
But the direction is consistent with what employers tell us, and the mechanism is specific enough to act on. If you are advising an eighteen-year-old this fall, the honest version is not "avoid technology." It is: choose work where the beginner still has something to do that the tool cannot do without them — and get very good with the tools regardless.
