Blog

AI Agent Glossary

Autonomy level

How much of the decision-making an agent performs without a human approving it.

Agents are usually classified along a spectrum: suggest-only (human executes), human-approves-each-action, human-approves-risky-actions, fully autonomous with post-hoc review, and fully autonomous. Capgemini reports trust in fully autonomous agents fell from 43% to 27% in one year, and PwC finds executives trust agents most with data analysis (38%) and least with financial transactions (20%). Most production systems sit in the middle tiers.

Source: Rise of agentic AI: How trust is the key to human-AI collaboration — Capgemini Research Institute

See the numbers behind this term: AI Agent Statistics 2026.