The Agentic Frontier: Artificial Intelligence in Commercial Real Estate

Abstract

We review the AI literature in commercial real estate through a framework linking four market frictions (opacity, search costs, illiquidity, and heterogeneity) to AI methods and testable market outcomes. The literature shows strong evidence that AI improves prediction, but almost none that better prediction reduces frictions or changes market outcomes. Adoption data from the Anthropic Economic Index, merged with occupational task profiles for 16 CRE occupations (324 tasks), show that AI usage divides along task type rather than task difficulty: information-processing tasks show high engagement regardless of importance, while tasks requiring physical presence or professional judgment show none. AI engages the highest-importance tasks in each occupation, not the most routine. A friction-by-method classification of 66 papers shows that half the cells are empty. The high-importance tasks at zero AI penetration define the "Agentic Frontier."

Keywords: Commercial Real Estate, Market Frictions, Machine Learning, Large Language Models, Agentic Systems, Automated Valuation Models, Anthropic Economic Index

JEL Classification: C45, C55, G17, L85, O33, R30, R33