Most AI training is measured by who finished a course, not whether people can actually use AI in their role. That’s the gap between activity and fluency, and the metric most teams are defending is the one breaking the program.