Interesting article: Are AI Agents Becoming Autonomous Coworkers?
This is fundamentally about a version of the principal-agent problem and managing the right balance between highly controlled (and slow, manual but low chance of damage) and autonomous (and lower friction, but greater potential for problems).