Taking best practices is not the same as taking trade secrets. It's ambiguous and murky, yes, but realistically, if it is ambiguous and murky, the case is likely to fall apart. It's when it's clear that you took some secret sauce with you, and replicated it wholesale, that a case is likely to be brought, and likely to be won.
How about stuff like how you set up your automated tests, or the Python project folder structure, or that you always first test the road sign recognition on self driving cars before testing something else. Or you first pretrain your neural nets of lane detection before jointly training for road signs and lanes.
I mean you remember the code you wrote and when faced with a similar case you will solve it in a similar way. The above are just really silly simplified examples for it.
