Tensorflow is a proof that some people in Google don't understand software engineering: They don't know and can't decide even what interface they want to expose/hide.
It is a disaster.
It is a disaster.
This type of exchange is pretty much what happens all the time: https://github.com/williamFalcon/pytorch-lightning/issues/35
There is no reason why two things can’t be merged together for better UX and functionality, but two humans have egos, want to lead projects, and be in charge, so things remain separate.
Software Engineering isn't just about put stuff out there, it also involves careful planning and roadmap around abstractions.
As high profile as a project as TF, if they don't have a clear way to tell having many parallel implementations for essentially the same functionality is confusing and hurt developer experience and ultimately harms the project itself, then that only explains the disaster , as a result of mismanagement, but doesn't undo the damage.
At least in the PyTorch world, the team's done a phenomenal job at allowing an ecosystem of related projects to contribute to one another and co-exist!
The goal of TensorFlow is to put itself in as many ML deployments as possible, paving the way for auto ml, which is where the real money lies.