Ah yes, the language that's so famously bad at performance that its only noteworthy usage in industry is... a trading firm that needs to be able to handle data in microseconds to make money. If only they had chosen better.
* The full compiles can be slow, but incremental compilation with ghci and cabal is never so awful in practice. And I've worked on some big projects.
* I don't think the talent pool is especially shallow. I've never had trouble hiring Haskellers or finding a job as a Haskeller. If you broaden your scope to the Haskell-interested (and have teaching and learning core to your culture), it's even easier. I built some projects entirely with teams of new Haskellers. Went great.
So, first you have to hire quite senior Haskell engineers, and then at one point switch them to tracking Haskell to new hires.
There aren't many companies who are willing to do that, and for a good reason.
Only prerequisite is a good attitude. If you start bean counting, you've already lost. If you think this level of software engineering cost optimization is necessary, you're just making your workplace worse.
If those interns and juniors came to you knot knowing Haskell, then this is a very bad way of teaching people. You end up with bad, non-idiomatic code, poorly understood concepts etc.
> If you start bean counting
It's not bean counting.
Turns out you can learn the ropes & values in a couple months tops. Then you can learn with anyone and grow together.
All the things you fear are barely consequential in practice in Haskell. All those things - who cares? What's the worst that can happen? The language puts such a high floor on the code it's okay. The main mistakes these newbies can make are stuff about database & distributed system design - orthogonal to Haskell and something they should be mentored on anyways.
It’s a lazy language so estimating real world performance is hard. It has a slow compiler and a very shallow pool of talents. You don’t really need to be a complainer to think that removing it from your organisation code base is a step in the right direction if you are not an Haskell shop. I would do the same with Ocaml despite liking Ocaml.