The Canaries Are Still Singing, and Fewer Are Being Hired

The debate about AI and jobs has been stuck between two bad measures: headline unemployment, which shows nothing, and layoff announcements, which show whatever the announcing company wants them to. Two reports in August 2026 used better data and found the same thing. The effect is real, it is narrow, and it works through the front door of the labour market rather than the exit.

The Stanford update

On 12 August Erik Brynjolfsson, Bharat Chandar and Ruyu Chen of the Stanford Digital Economy Lab released a revised version of 'Canaries in the Coal Mine?', using ADP payroll records on millions of US workers through June 2026. The headline finding: employment among workers aged 22 to 25 in highly AI-exposed occupations is about 19% below where it would be had it kept pace with similarly aged workers in less-exposed occupations. The gap was 15% in the July 2025 data and has widened at every vintage since the paper first appeared. Experienced workers show no comparable gap.

In levels, employment of 22-to-25-year-olds in the two most exposed quintiles fell about 11% between November 2022 and June 2026, while the same age group in the three least exposed quintiles grew about 10%. The authors are explicit that there is no evidence of widespread, economy-wide job displacement, that their estimates are not causal, and that the mechanism is reduced hiring of young workers rather than increased separations. The divergence continued widening long after interest rates peaked and survives controls for rate exposure, education and remote work, though it is more pronounced in ADP data than in national survey benchmarks.

The global check

A week later, on 19 August, Goldman Sachs Research published a cross-country study by Dong and Briggs. Industries with greater AI-automation exposure have seen slower growth in job openings since the second half of 2022, most negatively in Germany, Australia and the United States. Call-centre employment sits 39% below trend in the US, 33% in Canada and 27% in Germany; software publishing, management consulting and advertising are also sharply below trend. The authors write that 'AI-related hiring headwinds are clearly visible in official and unofficial employment data, but impacts are limited to a narrow set of industries and workers', and that their entry-level analysis 'signals that junior workers may face greater headwinds to hiring due to AI adoption'.

Goldman's estimate of AI adoption, combining 11 surveys, is roughly 15% to 20% in major developed markets and 10% to 15% in emerging ones, according to CNBC's report on the note. The effects, in other words, are the early stage of a curve, not its end.

The counter-evidence

Not every dataset agrees. NPR reported a Ramp and Revelio Labs study of more than 21,000 US firms in which entry-level headcount grew 12% in the two years after adoption at the heaviest AI-spending companies. Reuters cites Challenger, Gray and Christmas counting about 113,000 US job cuts attributed to AI so far in 2026, a number that captures announcements rather than net employment. The honest reading is that firm-level and occupation-level evidence diverge, and that the Stanford and Goldman results are about exposure by task, not spending by company.

Who benefits, who is at risk

Beneficiaries: experienced workers in tacit-knowledge roles, whose employment is growing fastest in the Stanford data; employers who can hire fewer juniors without visible layoffs. At risk: graduates entering exposed occupations, and, in five years, the companies that find they have no mid-level staff because they stopped training any.