You can think of pytensor/theano==jax, a library that gives you autodiff and compiled code for array based computations.
Theano had an emphasis on the graph of the computation, which you could manipulate easily from Python, whereas other frameworks emphasize it less / hide it.
Theano influenced a bunch of these frameworks after which it was deemed “unnecessary” from a research standpoint.
When the original devs abandoned Theano, the pymc-devs forked into Aesara, then due to project-related conflicts it was again forked as PyTensor, still by the pymc-devs.