The thing about larger-scale architecture is that you can be correct in any specific sense that it’s more than you need, but when you actually try to make the thing across a development team, you end up there because the code reflects the organization, and having it broken up like that lets you more easily rewrite your previous decisions.

At the small scale this occurs when you notice that the way in which you have to approach a feature is linguistically different - it needs conversion to a substantially different data structure, or an interface that compiles imperative commands from a definition. The whole idea of the database having a general purpose structure and its own query language emerges from that - it lets you defer the question of exactly how you want to use the data. The more configuration you add, the more of those layers you need. When you start supporting enterprise-grade flexibility it gets out of control and you end up with a configuration language that resembles a general purpose programming environment, but worse.

Casey Muratori talks about this kind of thing in some depth.

In the end, the point of the code is to help you “arrive in the future” in some sense - it’s instrumental, the point of automating it is to improve the quality of your result by some metric(e.g. fewer errors). For a lot of computations, that means you should just use a spreadsheet - it aids the data entry task, it automates enough of the detail that you can get things done, but it also gets out of the way instead of turning into a professionalized project.