Having an industrial focus for most of my career, I personally find MATLAB to be very useful. It is only on very rare occasions that I find Maple (or Mathematica) to be that helpful (the last time I recall using symbolic code for more than convenience was in confirming my calculations for Haar measures for a non-unimodular Lie group some years ago).

One of the nicest things about using something like MATLAB (or Octave) is that using a numerical environment forces you to be careful in a way that symbolic ones don't, while allowing you to gloss over details that would be very time consuming in other languages. Often times using Boolean variables or clever use of primes allows a considerable amount of "symbolic" computation that is much easier to translate into practical code. It's not Grobner bases or whatever but it's usually pretty effective.

I have made it a standard practice for most of my company's code to be prototyped in MATLAB, whether it's mathematically oriented or not. This is also very helpful for validating outputs against C or Java code as well as checking that graphs using Swing, OpenGL, etc. are displaying properly.