An Open-Data Pipeline for Measuring Occupational AI Vulnerability

This paper replicates Manning and Aguirre’s (2026) occupation-level analysis of workers’ vulnerability to AI-induced displacement. It imports the same published AI-exposure scores, reconstructs the adaptive-capacity index, and replaces the original study’s proprietary Lightcast geographic input with public OEWS area employment and Census Gazetteer land-area data. The replication retains 388 occupations covering 147.4 million workers (95.6% of the OEWS employment universe), compared with 356 occupations covering 147.9 million workers (95.9%) in the original study. The employment-weighted correlation between AI exposure and adaptive capacity is 0.474, compared with the original estimate of 0.502. The replicated vulnerable group contains 6.0 million workers (4.1% of covered employment), compared with 6.1 million (4.2%) in the original, and remains concentrated in clerical and administrative occupations. The public geographic-density measure has a correlation of 0.417 with AI exposure, compared with 0.426 for the proprietary measure. The close agreement across these estimates confirms Manning and Aguirre’s principal findings and shows that they hold even when using only public data.