In my experience, polish has an exponential cost in terms of time and complexity. Gameplay scripting can be taught to young children. Pioneering solutions to rendering and physics problems reserved for offline computation in a real-time field to provide that extra layer of polish only seen in AAA? Often exceptionally hard.

Engines like Unreal betray the complexity of what, at least when I was part of the industry, constituted "game programming". Most common use cases have already been accounted for, all of the complex work already done for you. This is similarly true for algorithm design in game dev with every indie parroting the same decades old industry tricks. The illusion lasts as long as you don't do anything too interesting or orthogonal to the design of the engine that's doing all of the heavy lifting.

Game programming isn't just performing a navmesh query, it's implementing recast. It isn't prototyping with horrendously inefficient post-process materials. It's adding, albeit fairly straight forward, custom shaders. Or not so straight forward.