No Evidence of Aggregate AI Job Loss

Summary

The article argues that nearly four years into commercial large-language-model deployment, there is no statistical evidence of net aggregate job destruction in the United States or the United Kingdom. It points to record employment totals, unemployment rates of 4.9% in the UK and 4.1% in the US, and high prime-age participation in the US as evidence that the predicted economy-wide collapse has not appeared. The author distinguishes aggregate employment from occupational turnover: Stanford Digital Economy Lab research using ADP payroll data covering more than 25 million US private-sector workers reportedly found no widespread displacement, while an NBER paper found modest overall effects because productivity gains can increase labor demand elsewhere. Research cited from the Journal of Financial Economics is said to find that firms investing heavily in AI expand employment by 2% to 4% over two to three years. The article nevertheless describes localized harm. Workers aged 22 to 25 in highly exposed occupations allegedly trail less-exposed peers by about 19%, while routine gig-platform translation and first-draft writing work has reportedly seen project volumes and hourly rates fall by 15% to 25%. The main adjustment appears to be slower entry-level hiring rather than mass layoffs of experienced employees. The author attributes broader 2026 hiring weakness primarily to high interest rates and energy costs, citing oil above $105 per barrel, US 10-year Treasury yields of 5.23%, and UK gilt yields of 5.42%. The conclusion is that AI may automate tasks and reshape career entry without yet producing measurable aggregate employment loss, while the article’s broader economic claims remain an interpretation of the cited data and studies.