Users fault personal agents more for holding back than for overstepping
Users of nine personal agents fault them more for asking again and for staying silent than for acting on their own. The benchmarks the field cites mostly find the opposite failure, one test at a time.