Designing Oversight Interfaces for AI AgentsHow to design for the conditions that make control meaningful?