Since not all images benefit from Photo AI, it is useful to see results before saving. However, the software starts all over processing the image when you save the on-screen results. Why? This doubles the time per image.
Also, when processing an image on screen, any interaction with the original image as well as the processed image (preview updated) like zooming or repositioning triggers the processing to start all over again.
This is normal processing as what you see on the screen is the only part processed (it is faster than processing the whole image each and every time you adjust the settings), on save it will process the whole image.