What insight is being gained by going with the more complex model? You can work out a lot using classical EM and predict a bunch of cool optical/diffraction phenomena without a lot of math, and the math you do is crystal clear. QFT or even just QM+time-dependent perturbations are a pain in the ass in comparison.

If you have engineering mindset (how to do something), then, sure, you are better off with the most useful approximation. If you have scientific mindset (how things are), then practical usefulness of the model of reality is secondary.