A lot of it depends on the work you do, and what other kinds of tools you use. If you’re working with simpler code and use something like flake8 or have a reasonably fast test cycle, you can be pretty productive either way. If you work with harder-to-type data structures (e.g. nested JSON or XML), you’ll see less wins from typing then validation (this is why Django apps tend to have fewer issues this way because the database validation reduces the amount of data motion before something validates it).
That’s not to say that typing isn’t useful - I use it daily, especially since VSC has made it a lot faster than mypy - but I do think the Python community is wide enough that you can find people legitimately saying it is or isn’t a big deal for them as a function of where they work. You also have a certain amount of PTSD from languages like Java which make typing far more labor intensive and through some combination of limited expressiveness, dynamic logic, and culture end up making the benefits less than anticipated.
I can absolutely see that there are many cases where the code is not run many times or the dataset is dirty and that types actually get in the way. As a example, I cannot stand using stuff like VSCode with lsp because it is unbearably slow for both linting and code completion. I feel pycharm is faster (actual typing is slow, but code completion makes me rarely type more than 3 characters before completing which makes it feel faster). I’m mentioning this because I would never dream of typing out a data structure, which is probably because my work is so different from an analysis job. For me code for data structures should be either generated or be simple enough to type out in a few minutes and there should be as few as possible of them. In sharp contrast to data engineering type of work where data just is dirty because it is so much of it from everywhere. Anyway, in both cases I dont want to type a full variable name without some completion, ever, much less 100 times which it seems some people are completely fine with. I’m showing my inexperience with python here, but I would love if there existed something like the F# type providers so I do not have to type so damn much:D
One of the pycon type talks was about type checking json. TypedDict or pydantic can both be used to deal with that fairly well now. For a complex json sure the type might be long but write it once and used the typeddict name after that. It feels similar to needing to write protobut def/thrift def. Some of the json I even work with is given a spec with protobuf and loaded that way which gives you type support. protobuf can generate python type stubs with mypy-protobuf. One practical annoyance is many libraries that use protobuf generated code don't currently include type stubs with them. I currently just make the stubs myself, but that is something most people would probably get stuck on and just needs more libraries to update the build scripts to include them in wheels.
XML could have a similar approach although I'm not sure if anyone has done it already.
Yes, I use pydantic for that and it’s quite helpful - one of my favorite libraries in recent years. I do think your last point is key: this seems to be a community maturity point because a fair fraction of people will bounce off of a bad first experience and say it’s not worth the cost.
That’s not to say that typing isn’t useful - I use it daily, especially since VSC has made it a lot faster than mypy - but I do think the Python community is wide enough that you can find people legitimately saying it is or isn’t a big deal for them as a function of where they work. You also have a certain amount of PTSD from languages like Java which make typing far more labor intensive and through some combination of limited expressiveness, dynamic logic, and culture end up making the benefits less than anticipated.