That’s exactly what I mean. If you have two independent variables (say “age” and “income”), then apply PCA, we will get two variables (we’ll call them Z1 and Z2) that are both combinations of age and income. If we were to fit a linear model by regressing Y on Z1 and Z2, we will be able to understand how a one-unit change in Z1 affects Y, but we won’t easily be able to understand the effect of age on Y.

If you don’t care about how age affects Y but just want to build the best model, then you may not care.

However, if you do care about understanding how age affects Y, then PCA is an inappropriate method for you to use because you won’t be able to understand the effect of age on Y.