I also find myself in an environment where a lot of people wouldn’t even consider using Julia because of its relative obscurity. However, I nevertheless find that the modern concept of “microservices” is the primary thing that allows me to use Julia for work. Since nobody is likely to use the code I produce as a proper API, it doesn’t really affect me that it might be difficult to do that in the languages others are using (primarily Python and Go). What I really need is for the serialized IO of Julia to be good. Thanks to packages such as ProtoBuf, FlatBuffers (which I can’t get anyone to use), JSON (which everyone insists on using), HTTP and LibPQ the fact that I use Julia doesn’t much affect my colleauges who would likely find it extremely annoying otherwise. The biggest problem I have with using Julia at work is the lack of an equivalent to JDBC or SQLAlchemy (for which I use JDBC). I have toyed with the idea of writing a halfway-decent database package interface for Julia at least, but those efforts have fizzled out because it doesn’t solve the core problem (i.e. having a nice Julia-native alternative to JDBC).
Most of my actual work involves mixed integer linear programming and some graph theory, and it just so happens that the alternatives for doing what I do in other languages are abysmal at the moment. There is another group that does some of the linear programming in python and it’s absolutely awful. So, if someone asked me not to use Julia, they’d get a pretty lengthy lecture on why moving to something else would be a terrible idea.