90% of US working hours are spent on tasks not covered well by AI benchmarks

Current AI benchmarks cover only about 10% of actual American work tasks, with 208 of 923 occupations having meaningful benchmark coverage and only 18,797 of tasks represented across 924 benchmarks. The coverage is skewed toward high-wage occupations and AI-exposed roles like lawyers, while leaving middle-income knowledge work and smaller job categories largely unmeasured, making it difficult to assess how well AI models perform on tasks Americans actually do.