Gödel tells us, paraphrased, that no system can be fully expressed in terms of itself. For example, the dictionary, which attempts to express the meaning of all English words, nevertheless uses English to do so. You have to know enough words to learn what the other words mean, and you can build up from there.
In other words, every system needs its axiomata. An axiom is essentially a fact about a system that is assumed known.
This is analogous to the design of a system. It seems somehow more elegant to design a system that works based on things that are already working than to write a new procedure to make something work: these are your axiomata. Overengineering comes in when you relentlessly try to base your system on axiomata instead of simply creating a new entry in your dictionary. When you find yourself trying to find the "most elegant" solution to your problem you might actually be trying to find the "least work" solution.
Overengineering, if you think about it, has the ultimate goal of having the entire thing just work if you prod at a particular pressure point in your towering mass of pre-existing code.
Well stop it. You can't make entire system without writing a bit of code. Heck you don't even have a system if it's just a collection of axiomata. You will have to write at least a bit of glue code. And don't try too hard to leave your system as a collection of axiomata for new systems. Make it work, first.