### Effects of LLM on the labor market
Study made by Stanford University shows six key findings on the early labor market effects of generative AI, using high-frequency payroll data from ADP. They show that since late 2022, coinciding with the widespread adoption of generative AI tools like ChatGPT, employment for early career workers (ages 22–25) in highly AI exposed occupations (e.g., software development, customer service) has declined significantly.
Employment for older workers in the same roles and workers in less exposed fields has remained stable or grown.
The declines are concentrated in roles where AI is used to automate (rather than augment) tasks, are robust to firm level shocks and remote work considerations, and are more evident in employment numbers than in compensation levels.
Here is a summary table:
| Statistic / Finding | Value / Description |
|-------------------------------------------------------------------------------------|---------------------------------------------------------------------------------------------------------|
| **Relative employment decline** for early-career workers (22–25) in high AI-exposure jobs | 13% (after controlling for firm-level shocks) |
| **Overall employment growth** for young workers (22–25) in high-exposure occupations | –6% (from late 2022 to July 2025) |
| **Employment growth** for older workers (35–49) in high-exposure occupations | +9% (same period) |
| **Adoption of generative AI** among U.S. workers (age 18+) by mid-2025 | 46% |
| **Sample size** (monthly payroll records in main analysis) | 3.5–5 million workers |
| **Key exposed occupations** | Software developers, customer service representatives, accountants, administrative assistants |
| **Key less-exposed occupations** | Nursing aides, maintenance workers, freight movers, home health aides |
| **Primary data source** | ADP payroll records (largest U.S. payroll provider) |
| **AI exposure metrics used** | Eloundou et al. (2024) GPT-4 β; Handa et al. (2025) Anthropic Economic Index (automation vs. augmentation) |
source:
https://digitaleconomy.stanford.edu/wp-content/uploads/2025/08/Canaries_BrynjolfssonChandarChen.pdf