Utilizing the text of US patents from 2003–2022, I categorize Artificial Intelligence (AI) inventions into an automation margin and an augmentation margin. I then estimate the exposure of occupations to these AI-shocks using O*NET data. Using a shift-share model design on the commuting-zone level, I estimate the causal effect of these shocks on employment levels and wages. I find that augmentation exposure increases employment levels 5–10 years after a patent is granted. I also find suggestive evidence that automation lowers wages. I decompose these effects by the skill level of exposed occupations and find that automation shocks arriving through low-skill occupations decrease local employment and wages while augmentation shocks arriving through them increase employment at the later horizons. Automation shocks arriving through middle-skill occupations raise local wages at the later horizons, and shocks of both types arriving through high-skill occupations are associated with increased employment levels.
Draft coming soon.
© 2026 Dean Y. Kaplan