Augmenting or Automating Labor? The Effect of AI Development on New Work, Employment, and Wages
Citation. Marguerit, D. (2025). Augmenting or Automating Labor? The effect of AI development on new work, employment, and wages. SSRN Electronic Journal. https://doi.org/10.2139/ssrn.5169611
Summary. Marguerit studies U.S. labor-market outcomes from 2015 to 2022 using novel Stack Overflow-based measures that separate AI exposure into automation AI, which substitutes for occupational task performance, and augmentation AI, which enhances worker output. The paper links these exposures to new job-title emergence, employment, and wages, using instrumental-variable estimators based on AI development in countries with limited economic ties to the United States. The central contribution is not simply that AI exposure matters, but that automation and augmentation exposure point in different labor-market directions and vary by occupational skill group.
Key claims
- Automation AI and augmentation AI should be measured separately because they can have countervailing effects on new work, employment, and wages.
- Augmentation AI exposure is positively associated with the emergence of new work; in the 2SLS results, a one standard deviation increase in augmentation AI exposure increases the cumulative share of new work by 1.5 percentage points in the fully controlled specification.
- Augmentation AI exposure also predicts higher employment in the aggregate 2SLS employment model, with a one standard deviation increase associated with a 3.1% increase in employment size in the most refined specification.
- Automation AI exposure has a negative wage effect in the aggregate 2SLS wage model; a one standard deviation increase in automation AI exposure reduces hourly wages by 7.7% in the most refined specification.
- Heterogeneity by occupational skill group matters: automation AI reduces new work, employment, and wages for low-skilled occupations, while augmentation AI fosters new work and raises wages for high-skilled occupations without clearly increasing high-skilled employment.
- Middle-skilled occupations sit between those patterns: automation AI reduces new work and wages, while augmentation AI increases new work and wages but does not significantly increase employment.
- The findings imply that AI development may contribute to wage inequality by harming lower-skilled occupations through automation exposure while raising wages in higher-skilled occupations through augmentation exposure.
Evidence & limitations
- Design: empirical labor-economics working paper using U.S. occupation-industry labor-market data from 2015–2022, Stack Overflow-derived AI exposure indices, new job-title measures for new work, and instrumental-variable estimates using AI development in countries with limited economic ties to the United States.
- Evidence strength: separates automation and augmentation exposure, tests new work, employment, and wages rather than only exposure, reports heterogeneous effects by occupational skill group, and includes robustness checks for IV country selection, transition-matrix construction, tag descriptions, and employment weighting.
- Limitations: working paper evidence rather than peer-reviewed publication; exposure measures infer AI development from developer-question data rather than observed firm deployment; estimates are occupational/industry-level and may miss within-firm redesign, job quality, or worker-experience mechanisms; the study covers 2015–2022 and may not capture post-ChatGPT GenAI diffusion.
- Provenance: the Drive PDF exactly matches the public arXiv v1 PDF for arXiv:2503.19159; Crossref also identifies the same title and author as SSRN DOI 10.2139/ssrn.5169611.