How much verification before an agent's task counts as done?
A user running agents that publish on their behalf found the real failure mode is not a wrong answer but a task marked done too early — one agent reported a successful upload with a real ID, but to the wrong place. The thread asks how much checking a task needs before it counts as finished. The practical answer is evidence, not confidence: require the agent to show the actual file, page or state change before the task leaves the queue.