Autonomy Level

Autonomy Level

Also called: 自主等级 · Levels of autonomy

A grading idea borrowed from the L0–L5 levels of driving automation, describing how far an agent runs without human intervention. No agreed standard exists yet.

L0 manualL1 assistedL2 partial autonomyL3+ supervised autonomy
Schematic (simplified)

The intuition behind the ladder

Roughly: manual operation → agent suggests, human executes → agent executes within bounded scope with human spot-checks → agent completes end-to-end, human only sets goals and provides a backstop. Each level up lowers how often the human intervenes and widens what the agent may decide on its own.

Why to use the term carefully

Unlike SAE J3016, the agent field has no authoritative, verifiable level definitions, and vendors' "L3 agent" claims are not consistent. It is fine as loose language for communicating autonomy scope, but not as an objective measure of capability or safety. Stating plainly *which actions skip approval and which require sign-off* is more informative.

Common misconceptions

  • There is no official "agent autonomy level" standard; vendor definitions are not directly comparable.
  • A higher level does not mean more reliable — only that humans intervene less.

Related terms

Sources

  1. SAE J3016 — Levels of Driving Automation (analogy source)
  2. Shavit et al. — Practices for Governing Agentic AI Systems, OpenAI (2023)

Compiled 2026-08-29 · This glossary is compiled from public papers, official specifications, and common industry definitions, and is updated as the field evolves. Corrections welcome.