// GLOSSARY -- AGENTIC AI
What are Agent Autonomy Levels?
The degree of independent financial decision-making an agent has — from fully supervised (human approves every transaction) to fully autonomous (within pre-set boundaries).
WHY IT MATTERS
Not every agent needs the same freedom. Testing agents need full autonomy. Production agents managing $1M need oversight.
Levels: L1 (human approves all), L2 (auto within tight limits, HITL above), L3 (auto within broad limits, circuit breakers), L4 (fully autonomous within policy).
Right level depends on: amount at risk, track record, decision complexity, organizational risk tolerance. Most start L2, increase gradually.
HOW POLICYLAYER USES THIS
PolicyLayer adapts controls to autonomy level — fully autonomous agents get stricter boundaries, supervised ones get permissive limits with HITL.
FREQUENTLY ASKED QUESTIONS
How to increase autonomy over time?
Based on observed behavior. If an agent consistently operates well within limits with zero violations, gradually widen the boundaries. PolicyLayer's analytics provide the data for this.
Can different tasks have different levels?
Yes — PolicyLayer supports session-based autonomy where the same agent has different permission levels for different tasks.
What triggers a downgrade?
Policy violations, anomalous spending, security incidents, or operator decision. PolicyLayer can automatically downgrade autonomy when violations are detected.