Benchmarking the Future of Work: Mapping AI Progress to Occupational Tasks with Kris Gulati 

• Under review at Management Science

Abstract: "Technological progress can become visible in technical evaluations before its consequences appear in firms, occupations, or productivity statistics, creating a challenge for measuring economically relevant technological change. This challenge is particularly salient for artificial intelligence, where capabilities can advance rapidly and unevenly across domains. Benchmarks - the scoreboards that track AI progress - provide an underused signal of emerging technological capabilities, but technical performance alone does not reveal how capable AI systems are relative to humans or how relevant those capabilities are to real-world work. We introduce the Benchmark-to-Work Index (BWI), a dynamic measure of occupational AI exposure that makes two contributions. First, BWI maps individual, time-stamped AI and robotics benchmarks directly to O*NET human abilities. Second, BWI measures how capable AI systems are relative to humans and how relevant those capabilities are to the workplace, distinguishing technical progress from economically meaningful capability. Applying this framework to 894 occupations from 2020 to 2026, we find that exposure is highest in knowledge-intensive occupations. The dynamic structure of BWI also reveals that exposure has evolved unevenly, with the largest increases between 2020 and 2026 concentrated in mathematical, statistical, financial, and scientific occupations. Earlier BWI vintages are positively associated with subsequently realized AI use in the Anthropic Economic Index (Spearman's ρ = 0.61), suggesting that benchmark-based exposure can provide a leading indicator of where AI use subsequently emerges. Incorporating human-relative capability and workplace transferability modestly improves this predictive relationship compared with benchmark coverage alone. However, substantial differences remain between demonstrated capability and realized use, highlighting occupations where technological possibilities may be ahead of organizational practice. Overall, BWI provides an economically interpretable framework for identifying where AI capabilities relevant to work are emerging before they become visible in realized adoption."

The Missing Colleague: Generative AI and the Informal Organization of Scientific Research with Kris Gulati

• Submitting to Management Science by November