> We will have to get used to judging software based on the output not on the input.

This has been the case for far longer than AI has been a consideration. We all know that counting lines of code isn't a mark of quality, and yet we venerate those that put in all the hours to get something done.

I would say you need to go a step farther than that. The output needs to be impactful. Increased MAU (or GH stars, downloads, etc.), lowered support load, lower ops costs, whatever it might be.