AI adoption in bureaucracies

Abstract This paper examines relationships between AI occupational exposure and workforce patterns in U.S. federal agencies from 2019–2024. Using administrative employment data, we document systematic associations between agencies’ concentrations of AI-exposed occupations and employment dynamics. Agencies with higher AI exposure exhibit declining routine employment shares, expanding expert roles, and wage compression effects. We develop a theoretical framework incorporating institutional constraints distinguishing public organisations: employment protections, standardised compensation systems, and political oversight. The model features strategic interactions between budget-maximising directors and electoral-sensitive overseers, predicting workforce evolution under institutional constraints. Our identification exploits fixed occupational exposure scores, so observed changes in agency-level exposure reflect workforce composition shifts rather than measurement artefacts. Patterns suggest agencies with greater AI-susceptible occupations experience reallocation rather than displacement, providing insights for understanding technological change in institutionally constrained environments and informing governance frameworks balancing modernisation with democratic accountability.