End-to-end learning means you can learn the model like a black box. That is, you only need to take care of the input and the output and ignore the processing/abstraction in the middle.
This concept is illustrated in the picture below. Take computer vision pipeline as an example, you don’t need to take care of the mid-level features and classifiers if you use end-to-end learning.
The picture above comes from Sergey Levine.