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

	90% of US working hours are spent on tasks not covered well by AI benchmarks(mercor.com)
	599 points by speckx 9 days ago \| 548 comments
	tl;dr: Analysis of 29 economically-relevant AI benchmarks mapped against the US Department of Labor's O*NET taxonomy (923 occupations, ~18,800 tasks) found that only 208 occupations have tasks with meaningful benchmark coverage, leaving roughly 90% of American working hours untested. Coverage skews heavily toward high-wage, AI-exposed knowledge work like software engineering and law, while similar but lower-paid roles (e.g., legal secretaries, court clerks) and physical-world jobs are largely ignored—making it hard to assess how good AI actually is at the work people do.
	HN Discussion: ↑Benchmarks miss hidden value in jobs, like the doorman performing unrecognized functions beyond the obvious task ↑The article understates the problem; even 'well covered' tasks aren't truly covered, and benchmarks focus on low-value work ↑AI is being deliberately targeted at high-wage knowledge jobs to cut corporate costs, which the data reflects ↑Benchmarks fail to measure critical real-world skills like error detection, explaining the coverage gap ↓O*NET is an outdated and poor taxonomy, undermining the article's methodology and conclusions