Supervision — Human Governance of Agent Execution
Patterns for staying in control without micromanaging — approval gates, execution pacing, capability gating, and behavioral observation. How humans govern what agents do at runtime, including the inversion where agents observe human patterns.
Entries
- Stepwise Task Execution — One-task-at-a-time execution with pause-and-approve checkpoints
- Progressive Tool Disclosure — Revealing MCP tools in layers to combat choice paralysis and hallucination (+15% accuracy vs flat exposure)
- Agent-Mediated Self-Reflection — Using agents to observe your behavioral patterns (conflict avoidance, intention-action gaps, unregistered learnings) from digital exhaust