One important contribution of the nouvelle AI revolution in the eighties was to deconstruct the traditional notion of reasoning in isolation. Brooks [16,15] fought against the division of cognition in layers (perception - recognition - planning - execution - actuation) and instead proposed the notions of reactivity and situatedness: the phenomenon we call intelligence stems from a tight coupling between sensing and actuation.
In the same spirit of doing without layered approaches, we reject the notion of parameterizing the functional parts of a robotic artifact. The different parts of a body -- torso, extremities, head -- are not interesting if we establish them manually. By giving the evolutionary code full access to the substrate, the search procedure does without conventional human biases, discovering its own ways to decompose the problem -- which are not necessarily those that human engineers would come up with.
Human cognition, as pointed out by Harvey  lacks the ability to design complex systems as a whole. Instead, we usually proceed by complexity reduction (the ``divide and conquer'' method). This is why the classic AI work took the layered approach that Brooks rejected so strongly. Perhaps the greatest strength of ALife methods such as artificial evolution is their ability to develop the organization and subparts together as a whole.