Specialists flourish in such “kind” learning environments, where patterns recur and feedback is quick and accurate. By contrast, generalists flourish in “wicked” learning environments, where patterns are harder to discern and feedback is delayed and/or inaccurate.
This is sort of like the switch in thinking in AI in the '70s (?) where it became clear that teaching computers to get better and better and playing chess or deducing propositions was never going to produce intelligence, and that we were going to have to teach them to see and move around on their own instead.