COMPEL Glossary / autonomy-spectrum
Autonomy Spectrum
The autonomy spectrum is a classification framework that describes how independently an AI agent can operate, ranging from Level 0 (no autonomy -- fully deterministic instructions) through Level 5 (full autonomy -- self-directed goal setting).
What this means in practice
The COMPEL framework defines six levels: assisted autonomy (human approves every action), partial autonomy (predefined actions within constraints), conditional autonomy (multi-step execution with human escalation triggers), high autonomy (independent with periodic review), and full autonomy (self-directed, reserved for research environments). Each level requires progressively more rigorous governance controls. The autonomy level, combined with the risk tier, determines the governance intensity applied to each deployed agent -- from basic documentation for low-autonomy assistants to real-time behavioral monitoring and automated kill switches for highly autonomous systems.
Why it matters
The level of autonomy granted to an AI agent determines the governance intensity required — from basic documentation for simple assistants to real-time behavioral monitoring for highly autonomous systems. Organizations that deploy agentic AI without a clear framework for classifying and governing different autonomy levels risk either over-governing simple tools (slowing delivery) or under-governing consequential agents (creating unmanaged risk).
How COMPEL uses it
The autonomy spectrum defines six levels from assisted autonomy through full autonomy, designed during the Model stage within the Governance pillar. Each level requires progressively more rigorous controls. During Produce, agents are classified on the spectrum and appropriate governance is applied. The Evaluate stage verifies that governance intensity matches the autonomy level, and the autonomy classification combined with risk tier determines the governance requirements for each deployed agent.
Related articles in the Body of Knowledge
Related Terms
Other glossary terms mentioned in this entry's definition and context.