In my experience, the larger the software you write, the truer these become. At some point all obvious names will have collisions, and getting caching right is

In my experience, the larger the software you write, the truer these become. At some point all obvious names will have collisions, and getting caching right is crucial to do but difficult to achieve because it transcends the entire stack. You could group these two things into "getting the data model right" as the single hard thing, perhaps that rings more true to you :)