Definition
An agent is software that pursues a goal by selecting actions, observing their effects, and adjusting its plan — with a degree of independence from human direction.
Three words that matter
Four tests to apply Monday morning
For any “AI initiative” presented to you, ask:
Does it have a goal that persists across calls?
If no, you have a chatbot. If yes, you have something with a trajectory.
Can it choose to retry, branch, or escalate?
Decision points are where risk concentrates. Map every one.
Does it write to systems of record?
Read-only agents are an order of magnitude easier to govern. Know which yours are.
Can a human reconstruct what it did and why?
If your audit story is “check the logs,” you do not have an audit story.
The taxonomy is the cheap part. The hard part — trust architecture — is next.
SECTION 2 →