Agent autonomy levels
How much an AI agent may do without a person: suggest only, act with approval, act and report, or act alone within limits.
“Autonomous agent” in a sales deck can mean any of these. Autonomy is set per action, not per agent: the same agent can draft replies freely but need approval to send money. In G2’s 2026 AI Agents Report, 52% of buyers required human quality review before production, and overseeing agents, called bot-sitting, ate into the expected gains. More autonomy is cheaper only when the agent is reliable enough that checking it costs less than doing the work.
Related terms
- Human in the loop: A person approves, edits or rejects the agent’s action before it takes effect.
- Excessive agency: An AI agent with more tools, permissions or autonomy than its task needs, so a mistake or an injected instruction can do real damage.
- AI agent: A model that decides its next step itself, calls tools and repeats until the task is done.
- Pilot purgatory (PoC purgatory): When AI proofs of concept and pilots keep running or get dropped without reaching production, or go live but cost more in human oversight than they save.
