Mercor Finds 90% of US Work Hours Not Covered by AI Benchmarks

Most US Work Hours Involve Tasks Poorly Represented in AI Benchmarks

A Mercor analysis claims 90% of US working hours involve tasks not well covered by current AI benchmarks, highlighting a gap between AI evaluation and real-world labor.

Edited by The Intel Brief Desk · 17 Oct 2026, 02:00 · 1 min read

Key takeaways

A recent analysis by Mercor indicates that 90% of working hours in the US involve tasks that current AI benchmarks do not adequately cover.
This suggests a significant gap between the types of work AI systems are tested on and the actual activities performed by workers.
This finding highlights a potential misalignment between AI evaluation methods and real-world labor demands.

What we know

A recent analysis by Mercor indicates that 90% of working hours in the US involve tasks that current AI benchmarks do not adequately cover. This suggests a significant gap between the types of work AI systems are tested on and the actual activities performed by workers.

Why it matters

This finding highlights a potential misalignment between AI evaluation methods and real-world labor demands. If AI benchmarks focus on only a small subset of tasks, AI development may overlook many areas where automation or assistance could be valuable. Understanding this gap could influence how AI systems are designed and deployed across industries.

What is still unknown

No independent verification of the 90% figure has been published. Details about the specific tasks excluded from AI benchmarks, the languages involved, and any plans for expanding or updating these benchmarks remain unclear. There is also no information on when or if this analysis will be formally peer-reviewed or widely released.

Get the daily brief by email.

Most US Work Hours Involve Tasks Poorly Represented in AI Benchmarks

What we know

Why it matters

What is still unknown

Sources

Related