I kinda made an argument about this in my undergrad thesis: Matlab's M language is cool until you actually want to implement something useful, then all that syntactic sugar becomes a nightmare, working code becomes less readable and optimized code requires obscure and poorly documented tricks