Definition
Plain language
A running estimate of how often different kinds of software and tasks show up in real AI-assistant usage.
As stated in the literature
A dataset from Anthropic estimating the frequency of software categories and occupational tasks in large-scale agent-traffic logs; used in CUA-Gym to weight which mock environments to build.
Why it matters: It lets builders focus their effort on the tasks people actually ask AI to do, instead of guessing what matters.
For example, if customer-service emails show up far more often than 3D modeling in real usage, the index would reflect that gap so the more common task gets prioritized.