Sure, we can test the crap out of our code watching for npe's. We can lint like crazy. We can write large integration tests to better understand the larger picture better. But ultimately nothing is as pervasive as the compiler relentlessly asking you to be clear.
What's more, the benefits of a language like OCaml (where Tagged unions, GADTs, direct modeling of errors via an Either composition, etc. rule the day) on your code leads you to a really good place when the codebase is sufficiently large.
Giving up s-expressions and monads is indeed a sad thing. I'd miss it.
I'm contemplating similar chnages to my tool chain.
I don't mean to imply that core.tyoed is fictitious but I wouldn't consider it a substitute for real, pervasive static typing.
What sort of size is the breaking point in your opinion?
I've recently made the move from f# to Clojure and I'm significantly more productive but I'm worried that when I get to larger projects (10-100kloc) I will come unstuck.
If you reach out to me via email you'll probably find me. You can find that email by searching for my username on the C2 wiki. Posting it here would be troublesome.