An AI agent can combine a model with tools and a process for pursuing a task. The scope of its authority matters as much as its reasoning: reading information, drafting a change and making that change affect the world in different ways.

A useful design starts with explicit boundaries. What may the system access? Which actions require approval? Who can interrupt it, and how can a mistake be reversed? These are practical questions for the people deploying and using a system.

The NIST AI Risk Management Framework is a reference for considering risks throughout an AI system’s lifecycle. Our editorial lens is similarly concrete: describe the task, the affected people, the available evidence and the responsibilities. Greater capability does not remove the need to make those choices.

Sources & further reading

Links point to original reporting or primary research.

  1. NIST · AI Risk Management Framework
  2. Yao et al. · ReAct (2022)
See an error? Read our editorial and corrections policy.