Well that's the problem. Part of why Python is nice for ML stuff is because it was adopted for several other purposes too. I've never used Julia but have used niche things like Erlang, and there's always more friction that way.