I ran across this piece from Nielsen Norman Group’s Caleb Sponheim and it’s worth a read:
A Concrete Definition of an AI Agent: https://www.nngroup.com/articles/definition-ai-agent/
There’s a lot of loose talk right now about “AI agents.” The term gets stretched to cover everything from simple scripts to systems that run on their own. We’ve seen this pattern before. When the words get fuzzy, the systems usually follow.
What I like here is the attempt to draw a clear line.
An AI agent is a system that perceives its environment, makes decisions, and acts autonomously to achieve goals over time.
That “over time” part matters.
Most of what we’re building with LLMs today is still reactive. You ask, it answers, and then it stops. Even when it feels sophisticated, it’s still waiting for the next prompt.
Agents are different. They keep going. They decide what to do next. They carry context forward and work toward an outcome, not just a response. There’s a bit of self-direction in the loop.
That shift changes how we design these systems. It’s less about crafting the perfect prompt and more about setting goals, defining boundaries, and making behavior visible.
It also means we need better language.
This is a good step in that direction. The more we can agree on what we mean by “agent,” the easier it will be to build and evaluate these systems in a useful way.
Read the full article here: https://www.nngroup.com/articles/definition-ai-agent/
