AI and jobs: junior hiring squeezed, aggregate labor data still shows little

Fresh numbers come from jobsdata.ai. The site summarises a study by Chandar and Klein Teeselink that instruments AI adoption off job ads mentioning generative AI. The dataset is huge: 1.25 billion job postings and 154 million employment records across 41 countries.

Juniors are not being fired. The junior share of employment at adopters falls 1.9 percentage points by March 2026, about 3.3% of a 57.1% baseline. That decline comes mostly from senior employment rising 6.7%. Junior employment itself changes by minus 2.5% and is not statistically significant. Total employment is up 3.3%. The study calls it dilution, not displacement.

That distinction matters for how the story gets told. It also matters which occupations are affected.

Occupation mix barely moves, under half a point in 21 of 22 groups, according to the same summary. The exception is computer and mathematical occupations, the most exposed group. There the employment share grows 0.8 points while the junior share inside it falls 3.3. Among technology affiliates the pattern is sharper: junior share down 3.9 points, senior employment up 14.6%, total employment up 7.8%. The authors flag two caveats. This is a firm-level cross-border design, so it answers what happens inside adopting companies rather than to a national labour market. And junior here means seniority, not age 22 to 25.

A separate study finds pay and placement losses for exposed majors

Another study in the same collection identifies AI exposure by field of study. It uses Census PSEO and LEHD administrative records covering 6,665,500 bachelor's graduates, roughly 29% of all US bachelor's degrees conferred between 2016 and 2024. Graduates in the most AI-exposed decile of majors, largely computer science, information systems and software-adjacent fields, became 5 percentage points less likely to be employed in the quarter after graduation and earned about 13% less. Both shifts start immediately after ChatGPT.

For scale, the recession literature puts initial earnings losses from graduating into a downturn at 9-10%. This is worse, though concentrated in a few fields rather than economy-wide. Roughly half the decline is lower pay inside the same industries and half is graduates moving into worse-paying ones. The share entering Professional, Scientific and Technical services fell almost 6 points and Information over 3, while Accommodation and Food Services and Retail each gained more than 2. Counting that shift, top-decile earnings fell 15%, to levels last seen before 2016. The effect attenuates to about 5% after two years, and the sampled institutions skew large, public and research-heavy.

The counterweight arrives from the aleximas.substack.com now-cast, written with Jacob Schaal and reflecting research available through September 2026. It argues the impact of AI on the overall labour market has been consistently muted, with lagging indicators such as unemployment and layoffs hardly showing any effect. It concedes there is some evidence of impact on entry-level hiring, but calls that evidence mixed, noting Nordic data showing no reduction in entry-level hiring and surveys showing null or even positive effects. It also flags a confound: recent evidence on remote work complicates causal interpretations of exposure-based designs that lack pre-trend and work-from-home controls.

The Fed, and the data due this week

Policy attention is arriving ahead of the numbers. Traders Union reported on 28 September that Cook, speaking at Oakland Tech Week, described AI as a general-purpose technology with broad implications for monetary policy and financial stability. She said AI-driven investment can fuel short-term price pressures, which is why the central bank is watching its indirect effects on inflation closely. She added that adoption may cause both job creation and displacement. Her longer-term view is that AI could support stronger productivity and higher living standards, particularly for small businesses. The Fed is committed to responsible AI use in the financial sector, she said, and she is monitoring labour market effects.

The next test is scheduled. The Daily Upside notes that Friday brings the September jobs report from the Bureau of Labor Statistics, the last monthly employment summary before the Federal Reserve decides on rates at the end of October. Job openings figures land on Tuesday and ADP's private-sector estimate on Wednesday. Friday's report is expected to show the addition of 100,000 jobs and an unemployment rate of 4.2%, according to a Reuters poll of economists. The same piece recalls that Anthropic's CEO said last year the technology could eliminate half of all entry-level white-collar roles, and then lists studies pointing the other way: Stanford's Institute for Economic Policy Research found in July that unemployment is rising for exposed workers but no faster than for the least exposed occupations, while a Ramp and Revelio Labs report found companies making the largest AI investments grow employment by roughly 10% after adoption.

Surveys add another layer, and one that is easy to misread. The jobsdata.ai roundup covers a rubric by Jacobs and Imas rating eleven interventions, from retraining and wage insurance to UBI and universal basic capital, across welfare, agency, feasibility and durability. EITC tops feasibility at 79.8 but sits near the bottom on durability under full transformation at 31.9. Universal basic capital inverts that, first on agency at 76.3 and 93.5 on durability under transformation, second-last on feasibility at 33.1. The authors' argument is against picking one policy now, favouring sequencing against observable triggers. One caveat is stated plainly: the scores come from 51 AI agent personas built on survey data from 51 real economists, not from the economists themselves. The underlying survey numbers are separate: 85% of Americans back publicly funded retraining, 72% back UI and 54% back UBC.

None of this settles the question. The strongest claim the dossier supports is narrow: AI may already be affecting the hiring margin for junior white-collar roles most exposed to it, that attribution is contested, and aggregate disruption has yet to show up in the data. Forecasters cited in the now-cast expect the wave to arrive gradually, with the US labour force participation rate predicted to fall from 62% to 58% by 2050 and the recent graduate underemployment rate rising from 41% to 55% by 2035.