Same reason we (SW engineers) think X field is easy and can be self-taught: the fundamentals are easy to learn, but figuring out where and how to deal with edge cases when they pop up is something that can only be learned with experience.

I see the same with a certain subset of finance people who learn Python, and start thinking coding is easy. Yeah, fair you can whip some simple Jupyter notebook using Pandas to analyze time-series. Now build out a distributed, fault-tolerant ETL (CRUD++) system that follows all business rules, is maintainable/readable, and can scale to atleast 100 "servers."

Perhaps not the most apt comparison -- but the fundamentals in every field are easy to learn; but working at the edge is something you have no experience in until you do.