I'm generally not on board with the whole "programs are proofs" dogma. When I look into the details, all it says is that a successful type-check of a program gives a proof of the existence of a program of that type. In expressively powerful languages, this seems very uninteresting; so far as I know, you can easily construct a program of any type by just discarding the input and then creating an output of the desired type. (Handling the empty type is a little tricky, but I wonder if infinite loops or program failures would do.) So my understanding is that you only get the interesting type structure (where it isn't trivial to prove the existence of a type) by restricting the language. (Typed lambda calculus is not Turing-complete.)

So, while I see the curry-howard stuff as beautiful math, I don't think it justifies the 'programs as proofs' mantra that many have taken up! I don't see the appeal.