The discussion opens with major AI developments, including Gemini 3.7 Flash, the rising importance of speed and pricing, the surprising total cost of cheaper models, and continued leadership changes at OpenAI. The central idea is the AI Deputization Audit, a framework for deciding which tasks AI should handle independently, which should be completed collaboratively, and which should remain human responsibilities. The episode argues that access to personal and workplace context is becoming a key competitive advantage. Features that allow agents to observe activity or learn from demonstrations may make them far more useful, but they also create serious privacy concerns. To assess a task, users should consider how often it occurs, whether it can be taught clearly, whether the result can be checked, the consequences of mistakes, and how much human judgment or personal connection it requires. The framework leads to three categories: deputize, duet, and defend. Suggested experiments include inbox triage, Slack monitoring, morning briefings, automated dispatch, and chief-of-staff-style agents, with simpler tools potentially better suited to newcomers.