How are people architecting "flows" for their agents? have you found that a framework is helpful for this or is it more or less a mixture of task classification and defined functions for you right now?
I've realized that basically my agent is an ensemble of functions that I have sorted into sequences. The problem with this approach is that in the long run I can see it evolving into like a dag which with reusable code modules may be okay but doesn't seem ideal
Fair, I'm trying to assess if I am missing out on some key abstractions that help ease development from the frameworks which may or may not be the case.
How do you organize functions into flows? Or are you just 1 step by allowing the model to pick the tool at any point?