How Well Does Agent Development Reflect Real-World Work?

A research study analyzing 43 AI agent benchmarks and 72,342 tasks reveals significant misalignment between current agent development efforts and real-world human work patterns across 1,016 U.S. occupations. The study finds that agent development is overly programming-centric compared to where human labor and economic value are actually concentrated in the economy.