I'm not sure exactly how to phrase this, but to give an analogy, I love card games (ex: Legends of Runeterra, Race for the Galaxy, Hearthstone, etc). There are a fixed set of rules and a fixed set of cards. I can "grind" games and get better by noticing patterns, picking up new strategies or tactics by playing against a diverse set of players. The outcome of an interaction is usually idempotent (i.e. card 1 interaction with card 2).
In real life, things are quite complicated. Asking a certain question in a certain way to person 1 and person 2 may give wildly different responses, and may even depend on your mood, their mood, your tone, time of day, etc. It's super messy and unpredictable.
I also feel a similar way about system design. The nearly infinite possibility of inputs, outputs, TPS, throughput, scenarios make it difficult to reapply the same set of rules to different scenarios. This is just talking about one component, when we bring in N components, the interaction gets very complicated and the "rules" change" case by case. I'm sure it gets better with practice, but I also feel I have a limited opportunity to learn or practice these on the job.
Has anyone found a way to structure these learnings in terms of a repeated "grind", because oftentimes I feel overwhelmed and don't know where to start. This is a complicated question, so answers regarding either a) soft skills or b) system design separately I will treat as valid answers.