Doesn't mean there isn't some meaningful signal in its results. Signal/noise ratio might be lower than with actual science, but there is so much data using the framework that the signal part is still useful.
(Although, that said, the one thing of MBTI that does comport with modern understanding of personality is that introversion-extroversion is one of the main axes, although unlike MBTI's theory, it's more of a gaussian normal distribution than a bimodal distribution).
Granted, this part of the theory doesn't really get explored much in its usage in popular psychology.