Successful work per hour as a measure of AI productivity

A post suggests that successful work per hour may matter more than hours saved when measuring AI productivity. Citing a study with task success rising from 59% to 93% and completion time falling 38%, it calculates roughly 2.5 times as many successful outcomes per unit of time. The argument: there’s value in completing work that otherwise gets dropped, delayed, done wrong or done twice.