I found modelling them as humans made it easier to imagine how to structure them, but there's nothing saying that's it's a better way to model your agent tree. Would be curious for somebody to use this to run some experiments with different names/traits for the fractal agent tree.
