An example would be a REST API with a few endpoints where the database operations are handled directly in the route handlers uniquely for that specific task.

That’s a prime example for untestable code (not testable with unit tests/without IO). That might be fine for a tiny experiment, but I’d advise against it for projects of any size, even private ones. Always use a model like MVC, MVVM, three layers (data, business, user) …

I feel like we should have an in depth talk to better understand the problems you’re facing and the line of thinking that motivates your initial request. Unfortunately I currently do not have the time for that. The best I can do now, with the best of intentions, is to advise you to read literature about software development. The trouble is, that I’m not sure what to suggest, because I think there’s nothing that fits your premise. Maybe read about library development/reusable code so you better understand what not to make reusable by comparison? So maybe “Reusable Software: The Base Object-oriented Component Libraries” by Bertrand Myer or “Analysis Patterns: Reusable Object Models” by Martin Fowler. Though, both books are more on the old-fashioned side and I wouldn’t recommend them if you’re not an avid reader and (former) student of computer science.

Before I jump to “that’s a really bad idea” with my 20+ years of experience: why?

I mean, sure, don’t implement functionality you don’t need, but making code not reusable intentionally? Why?