I tried on release day. Managed to break my entire Linux system. That was an annoying day.
And pip install some tensor flow nonsense has broken my projects many, many times. At this point I try really hard to ever avoid depending on anything in that ecosystem.
Is there something about Bazel's design choices that makes it so complicated? I felt like an idiot when I tried to build some old version of TensorFlow with some old non-default flag. I found even learning CMake easier.