Actually, it's way more targeted to mathematicians. The level of symbolic computation it can do is way ahead of any symbolic toolbox you'll find. When I was in grad school, I used NumPy and my group mate used Mathematica. One day I was looking up identities in Gradshteyn and Ryzhik (a bible for identities), and he asked why I bother as Mathematica had it all. He challenged me to find any identity that Mathematica couldn't handle. I thought this would be easy as G&R has more stuff than any other identity handbook out there, but I lost the challenge.

I second that and there are quite a number of scientists which use MATLAB and FORTRAN side by side.

I would extend this statement: MATLAB feels like a REPL to FORTRAN while Mathematica like a REPL/library to/for LISP.