Great comment, I absolutely agree. I learned to code largely through problem sets, HIGHLY defined, clear issues with pre-cleaned (!!!) data. Once I'm out coding in the wild, I am the one creating the dataset, it's highly messy in its infancy, and most of my collaborators are in flux about what they want, how they want it, and which subgroup and sample they want to examine. The name of the game seems to be quick iteration and fast communication loops.