💭 maybe llm agents should have some growing metric causing them to get bored/change approach, and this could get them out of inefficient rabbit holes or encourage them to automate things