Cursor's Head of Security: Never trust the agent writing your code

A podcast episode featuring Travis McPeak discussing non-deterministic AI agent behavior, least-privilege access models, and designing secure-by-default workflows for agent-driven development.