Most of the time it works, but there are still corner cases throughout the python ML ecosystem where polars fails.

This and the decision not to natively plot with matplotlib keep me in pandas for most tasks. (Plus there's is still a relatively large demand for pandas training.)