Building a large SaaS AI product with Python: The tale of three ecosystems

We have spent eight years building an enterprise cloud-based AI software product, Rossum (name unrelated to Guido, we promise). Rossum now automates handling business transactions worth trillions of dollars - fully in Python (on the backend). Let's recount some war stories: What surprised us, what did we regret, and what were some important right choices we made?

The biggest war story is about the three major components of Rossum, and three major Python ecosystems - plain old-school Python, Django, and the emerging Pydantic / FastAPI school. Which one to choose? Why ever choose more than one? But what happens then - how much are you locked in?

We will wrap up with an appeal to Python maintainers - is the Zen of Python still true? "There should be one - and preferably only one - obvious way to do it?" Let's consider along which lines the Python ecosystem is fracturing, and how to possibly mend it.