Exploring runtime execution boundaries for AI actions, explicit autonomy models, and ways to make judgment paths traceable in real-world systems.