Mapping the Structural Divide: Institutional Resilience, Post-College Market Position, and Artificial Intelligence Exposure Across U.S. Higher Education

We present a transparent, reproducible framework for mapping U.S. four-year colleges and universities along two composite dimensions — Institutional Resilience and Post-College Market Position — using publicly available federal data. The framework covers 1,609 currently operating institutions and includes, as an exploratory extension, an institutional-level measure of AI-related labor market exposure derived from O*NET occupational task analysis weighted for entry-level career pathways. We document three principal findings. First, American higher education exhibits pronounced tier-level stratification: R1 research universities and well-endowed liberal arts colleges occupy a structurally distinct tier, while baccalaureate and small master's institutions concentrate at the opposite end, with 40.5% of institutions simultaneously shrinking in enrollment and scoring below the median on institutional resilience. Second, when our theoretically-derived AI task exposure scores are compared with observed real-world AI adoption rates from the Anthropic Economic Index, the field-level correlation is approximately zero (ρ≈−0.09), suggesting that what AI can theoretically automate and what AI is currently being used for are largely independent — a pattern consistent with two distinct timelines of potential disruption. Third, the degree fields our model identifies as most exposed to entry-level AI task displacement are currently the highest-earning fields for new graduates, meaning that if disruption materializes at scale, it would affect career pathways with the highest current economic returns. Sensitivity analysis across 13 alternative specifications confirms that institutional positions at the extremes are robust to methodological choices, while approximately 31% of institutions are robust to all analytical specifications, while 69% near quadrant boundaries are sensitive to at least one alternative assumption. The framework is intended as a strategic classification heuristic rather than a predictive model; we provide the complete dataset and replication materials for public use.