I’d like to discuss an optional approval mode for ChatGPT agents: a middle ground between requiring a separate confirmation and granting standing permission for actions. When I explicitly ask an agent to perform an action, confirming it again can feel repetitive.…