[R] Agentic AI and Occupational Displacement: A Multi-Regional Task Exposure Analysis (236 occupations, 5 US metros)

A new research paper introduces a novel framework for measuring the occupational displacement risk posed by agentic AI—systems that can complete entire workflows end-to-end, not just isolated tasks. The study extends the established Acemoglu-Restrepo economic model by adding a 'workflow-coverage' term that accounts for AI's ability to chain tool calls, maintain state, and self-correct. This approach challenges previous models (like Frey-Osborne or Eloundou et al.'s GPT exposure) that assumed tasks were independent and that occupations would survive as coordination shells. The analysis was applied to 236 occupations across five major US tech hubs: San Francisco Bay Area, Seattle, Austin, Boston, and New York City.

Key results reveal a significant shift in which jobs are most vulnerable. Cognitive, high-credential roles previously considered safe, such as judges, credit analysts, and regulatory affairs officers, rank higher in displacement risk than software engineers. This is because agentic AI excels at automating structured, multi-step workflows common in these fields. The study forecasts that 93% of information-work occupations in the SF Bay Area will cross a 'moderate-displacement' threshold by 2030, but no single occupation reaches a 'high-risk' threshold, suggesting widespread moderate exposure rather than the complete elimination of roles. The framework also identifies a 2-3 year adoption lag between metros, predicting Seattle's job market in 2027 will resemble NYC's in 2029.

The research is rigorously validated, showing strong correlation (Spearman rho = 0.84) with existing AI exposure indices. The authors transparently report a null result when testing against 2023-24 employment data but make a falsifiable prediction for 2025 data. Limitations include reliance on keyword-based scoring for 'coordination overhead' and a wide sensitivity range for adoption rates. An international extension of the study is currently in development.