For example, learning all of the HTTP status codes before you know anything about HTTP could seem like a waste of time.
However, when you are writing a server that makes api calls to a third-party service, when you receive a 400, if you know the codes by rote you won't have to research / reason why you are receiving that code.
I think "learning by doing" usually ends with more self-directed, problem solving learning. This type of learning can show your weaknesses and provide you with holes in your knowledge that you can use rote memorizing to fill.
So often your System 1 wins and your System 2 is never aware, that relegating System 1 to "impulse" is just hiding the problem from yourself. Training System 1 at least gives you a fighting chance to have accurate-enough first impressions/instincts.
Of course there are more-successful and less-successful ways of training System 1, but that's true of anything.
Then, when I actually learn the more complicated stuff, I don't ever think about the basic stuff (and I get that out of the way very early in the learning process). I think it frees up my brain to focus on the complexities of any domain a bit sooner than I would otherwise.
I'm not sure this method generalizes to other people, but it seems to work for me across various domains. I think the use of rote memorization is a bit underappreciated (given how useless it seems in high school or university), but I've continually found it useful in my professional life.
[Edit] It's also useful to identify what the "common action" are, and in general that's only from "learning by doing" and realizing what's important/what's not - then saving the important stuff to the "hot path" of processing.