They also want training data from their employees correcting the agents so the agents can eventually replace them (that's what they're hoping for at least).