The word 'agent' now means nothing. It appears in banking apps, Slack integrations, Notion plugins, and hate mail campaigns. Nielsen Norman Group cuts through the noise with a single operational definition: an AI agent is a system that pursues a goal by iteratively taking actions, evaluating progress, and deciding its own next steps.
The ambiguity is not semantic, it is structural. Marketing, engineering, product, and leadership each operate on incompatible definitions. That fragmentation prevents practitioners from evaluating reliability, assessing trust, or designing coherent user experiences. NN/G's framing adds three requirements to the definition: the system must be reliable, adaptive, and accurate.
The full article is worth reading not for the conclusion but for the diagnostic framework it builds around that definition. If your team is shipping something called an agent, this gives you the vocabulary to ask whether it actually is one.
[READ ORIGINAL →]