The Matlab language has evolved in a questionable way, software licenses are problematic and harming portability, but the biggest deal after having seen it used at various places, is that Matlab code becomes unmaintainable very fast (this is even more true with Simulink, but I won't digress).
Matlab is 100% good for one thing: accessing already existing Matlab code; which happens when dealing with research papers because students have been fed with free Matlab licenses since a young age.
Matlab is also good for early prototyping, because of the toolbox and the interactivity (it's like the Excel of R&D).
But for that, today you get the same with other platforms/languages; Octave and Scilab are good, but have the same "engineering" issues ; Python for example (could be true about R/Julia/...) will get you "further" in no time.
My advice to anybody: keep the number of Matlab SLOC low, rewrite Matlab code to something else (with appropriate docs) whenever you have figured out your algorithm!