Autonomy concerns what a system is permitted and able to do without another decision from a person. A tool that drafts a suggestion and an agent that executes a sequence have different operational authority, even when they use similar models.

Describe autonomy through concrete boundaries: available tools, accessible data, time limits, approvals and interruption. Responsibility remains a design and governance question. Strong task performance alone does not determine which permissions a system should receive.

Sources & further reading

Links point to original reporting or primary research.

  1. Yao et al. · ReAct (2022)
  2. NIST · AI Risk Management Framework