AGENTIC AUTONOMY LEVELS

Summary from the TLDR newsletter

Low autonomy limits risk and increases reversibility. Higher autonomy is better for explicit activities, as well as for fleets of parallel agents refactoring massive codebases. The edge of the frontier is the manager agent that delegates to helpers while continuously verifying their output, and only returning with the decisions that must be made by a human. This setup can be used to run hundreds or thousands of agents.