I argue that is more important because in practice you usually don’t have enough samples (or can’t resample) to use the normal distribution.