ISSUE 004
Nobody gets laid off. The door just stops opening.
This is the most careful piece of research I've read on AI and jobs, and also the most uncomfortable, because the effect it finds is one that no layoff announcement would ever capture. It also turns out the researchers have spent a year publicly trying to disprove their own finding, which is worth watching on its own.
An idea worth thinking about
The effect shows up as a door that doesn't open, not a door you're shown
The Stanford finding I keep turning over isn't the size of the gap, it's the mechanism. The researchers report that the divergence for young workers in AI-exposed occupations operates primarily through reduced hiring rather than increased separations. Nobody is let go. Positions simply aren't opened.
That distinction matters enormously for whether anyone notices. A layoff produces a WARN notice, a news story, a number in a monthly report. A job that never gets posted produces nothing at all. The 23-year-old who sends out sixty applications and hears back from none of them has no evidence anything structural happened to her, and neither does the labour market statistician.
For a small employer this cuts an odd way. You are probably not the firm quietly not hiring a junior, because you never had a junior pipeline to stop. But you are competing for the same young people, and the research implies they are currently cheaper and more available in exactly the occupations you would want them for — software, customer service, clerical work. That is a real and slightly uncomfortable opportunity.
Read the August 2026 revision
A tool I disappeared into
Live employment trends by age and AI exposure, from actual payroll records
The Digital Economy Lab publishes the Canaries Dashboard with ADP Research, tracking employment across occupations and worker groups at different levels of AI exposure. The underlying data is payroll records from the largest payroll software provider in the United States — not a survey, not postings, but what firms actually paid people. That is a rarer thing than it sounds and it's why this work carries weight.
The dashboard lets you see the divergence directly: early-career workers aged 22 to 25 and the next group up, 26 to 30, against everyone else. From the July 2026 release, workers aged 22 to 25 accounted for 7.4 percent of employment in November 2022, which gives you a sense of how small the affected group is relative to the whole workforce — and therefore why the aggregate numbers stay calm while this happens underneath.
The limit worth stating: ADP's client base is firms that use ADP, which skews toward established employers with formal payroll. The very smallest businesses are underrepresented. It is a very good window, not a complete one.
Open the dashboard
Learn something
A researcher writing down what he isn't sure about
Bharat Chandar, one of the paper's three authors, wrote a long public assessment of the state of knowledge on AI and labour markets. It is free, it is written for non-economists, and its distinguishing feature is how much of it is devoted to uncertainty and to research that disagrees with his own.
He lays out, for instance, that several independent analyses using Current Population Survey data find at most small changes in hiring in AI-exposed jobs, and that Danish data through 2024 likewise shows small overall effects — concluding that overall hiring has not declined meaningfully due to AI. That is his own finding placed in a context that shrinks it.
The same team published a separate note in February 2026 testing whether interest rate changes explained their result better than AI exposure did. They found they did not, but they also reported that under a more stringent specification controlling for firm-level effects, the decline only becomes significant after 2024 rather than in late 2022. That is a meaningful weakening of their own timeline, published by them, voluntarily. If you want a model for how to hold a finding you believe in, this is it.
Read the assessment
What we’re seeing
No widespread displacement. But young workers in AI-exposed jobs are 19 percent behind their peers.
Brynjolfsson, Chandar and Chen revised their paper in August 2026, and the two headline findings pull in opposite directions. First: they find no evidence of widespread, economy-wide job displacement from AI. Second: employment of workers aged 22 to 25 in AI-exposed occupations now sits 19 percent below where it would be had it kept pace with their less-exposed peers. Experienced workers show no comparable gap.
The gap has widened steadily since they first documented it in August 2025, and it concentrates in specific work — software development, customer service, clerical roles. It shows up as reduced hiring rather than separations, which is the mechanism described above.
Both halves deserve equal weight and rarely get it. The version of this that circulates is "AI is taking jobs," which their first finding directly contradicts. The version that gets less attention is that a specific, small, early-career slice of the workforce is absorbing the whole effect while everyone else's numbers look fine. An economy can be healthy in aggregate and closed to the people trying to enter it.
I should be plain that this is contested. Independent work using survey rather than payroll data finds at most small changes, and Danish data through 2024 finds small overall effects. The researchers say so themselves, at length. What makes their result hard to wave away is the data source: administrative payroll records rather than people's recollections.
One more thing worth naming, given this entire entry rests on one lab's work. It does — slots two through five are all Stanford Digital Economy Lab, because nobody else has assembled comparable payroll evidence. A single group being the only source is not a reason to dismiss a finding, but it is a reason to hold it loosely until someone independent replicates it.
Brynjolfsson, Chandar and Chen, revised August 2026
One question
When did you last hire someone with no experience?
A year, roughly. Reply with it — I want to know whether the entry-level door is closing in small Missouri businesses too, or only in the data.
More in this pillar as it's written.
— Shana
Sources link to the original publisher, never to a summary.