The AI Jobs Transition Framework: Mapping AI’s Near-Term Impact on Jobs
Citation. Richmond, A. M. (2026). The AI Jobs Transition Framework: Mapping AI’s Near-Term Impact on Jobs. OpenAI Economic Research.
Summary. Richmond argues that AI exposure alone is too blunt to identify near-term job disruption because technical capability interacts with human necessity, demand elasticity, and observed AI usage. The report classifies more than 900 occupations covering nearly all U.S. employment into high automation risk, reorganizing, AI-growth, and less-immediate-change archetypes, then compares those archetypes with observed ChatGPT usage. Its useful contribution is not a job-loss forecast but a measurement frame for separating capability, adoption, human constraints, and demand-side responses.
Key claims
- AI exposure is a necessary but insufficient signal for labor-market pressure: the report combines technical exposure, human necessity, and demand elasticity, then validates the framework against observed ChatGPT usage data.
- In the Drive version, the framework covers 921 occupations representing 147.9 million jobs and classifies 18% of jobs as relatively higher short-term automation risk, 24% as jobs likely to reorganize with possible declining employment as task composition shifts, 12% as jobs that could grow with AI because lower effective cost may expand demand, and 46% as likely to see less near-term change.
- ChatGPT is used about 3x more in jobs the framework labels as most susceptible to automation than in jobs facing less immediate pressure, but realized use remains far below theoretical exposure across job categories.
- Human necessity includes physical, relational, and regulatory/accountability constraints that can keep workers central even when many tasks are technically AI-exposed.
- Demand elasticity matters because productivity gains can reduce labor demand, expand output enough to offset labor savings, or reorganize task composition without cleanly eliminating the human role.
- The report treats its categories as a map of where near-term labor-market pressure may emerge first, not as forecasts of job loss or stable long-run occupation boundaries.
Evidence & limitations
- The analysis uses ONET occupation information, prior task-level AI exposure estimates, 2024 employment coverage, GPT-5.4-assisted elasticity estimates, and consumer ChatGPT usage traces mapped to ONET tasks; it does not observe or infer individual users’ occupations.
- The demand-elasticity component is a rough occupation-level estimate of response to a standardized 10% productivity shock over a two-to-three-year horizon, not a causal labor-demand estimate.
- The report is OpenAI Economic Research rather than peer-reviewed scholarship, and its source organization has direct strategic interest in shaping labor-market interpretation of AI.
- The Drive PDF appears to be an earlier 2026-04-16 version; the public OpenAI CDN PDF at the recorded URL is a later 2026-04-30 version with a different binary hash, page count, and some text/figure differences, so the frontmatter file hash records the ingested Drive version rather than the public current PDF.