Table showing five levels of AI agent autonomy, from Level 1 (Draft only) where the agent produces output and a human reviews, to Level 5 (Autopilot) where the agent runs everything with human oversight. Each level defines the agent action, the human action, and what tasks it's best for. The framework helps product managers and AI practitioners decide how much control to give AI agents. Source: AI Agent Management Framework, Product with Attitude.