“Where there’s muck, there’s brass”
In general, if you can do the more difficult aspects, then you’ll be more in demand (both because people need them done, and because it’ll serve as a badge that yes you know what you’re doing in a general sense).
In games specifically, it’s more likely that you’d find yourself doing optimization, or the low levels of porting a game from one system to another, as opposed to writing an engine from scratch. If it were me, I would probably get involved in some existing project and try to write a new low-level system (e.g. a Gaussian splatting renderer for Unity or something) as opposed to just reinventing a new engine of my own. But absolutely, if you’re skilled with the deeper systems aspects as opposed to just using the existing tools, it’ll help you generally speaking.
Quite a while ago I decided, for what reason I don’t know, to take Dwarf Fortress’s exported fully-populated world, translate it through a layer of LLM to obfuscate all the races to new fantasy races and create outlandish points of interest, and put the whole thing in a renderer that could dynamically move around and zoom to create a nicely rendered watercolor-style map for a massive explorable TTRPG type world.
I gave up the idea when it was like 1% done but if you’d like to have any of the nonsense I created in case there’s something useful you can turn it into, you’re welcome to it. It’s in Python.