Problem
The agent asks you to perform an action — restart a service, test the TUI. You comply, not questioning it. Neither you nor the agent realizes that the agent could have done it itself. Both sides share a blind spot about the agent's actual capabilities.
Pattern
When the agent asks you to do something, tell it "you do it" — even if you're not sure it can. The agent will often discover a way to accomplish the task on its own, revealing capabilities neither side was aware of.
This works because agents have broad tool access and problem-solving ability, but default to handing off tasks that feel like they belong to the user. A simple push back breaks that assumption.
Example
The agent asks you to install an APK on your phone. Instead of grabbing the phone, you tell it to do it. The agent discovers it can use wireless ADB to install the APK itself — something neither of you had considered.
