The word 'agent' now means nothing. It appears in banking apps, Slack bots, Notion plugins, and systems that have already deleted real users' email inboxes. Nielsen Norman Group cuts through the noise with a single working definition: an AI agent is a system that pursues a goal by iteratively taking actions, evaluating progress, and deciding its own next steps.
The definitional chaos has a real cost. Marketing, engineering, product, and leadership are each operating from incompatible mental models. Without shared language, practitioners cannot evaluate what they are building, assess its reliability, or decide when to trust it. NN/G's definition is aimed specifically at UX professionals, but the argument applies to anyone shipping or procuring these systems.
The full article is worth reading not just for the definition, but for the framework it builds around three properties that make agents actually useful: reliability, adaptability, and accuracy. If you are designing, building, or managing anything currently being called an agent, this is the vocabulary you need.
[READ ORIGINAL →]