Dependent types and other proofs are not that hard to understand in principle; you can be working on them for a very long and that indeed makes it mostly impractical. However, we will see some interesting things in search space limiting between program synthesis, verification and LLMs.
I tend to prove only the parts that are vital and the things that come automatic (when the holes are automatically filled because they are already 'done').
So yes, I agree, if you want something done quickly and don't mind if it expires a few months later, you don't grab haskell, rust and definitely not a dependently typed language or proof assistent.
Just using some form of static typing offers easy guarantees without much effort. Trying to Monad-ify everything for a much smaller level of compiler guarantees involves much higher dev effort (as most devs do not understand Monads), and much higher effort involved in anyone reading the code.
It is also expensive to fix things without hardware: everything is iso/pci and needing recertification to changing anything. So the 100k you spend for going for perfection pays off. So far at least.
I do see (irl) that it improves the odds though.