One example related to software is augmenting decision-making before trying to replace it. PayPal's first major fraud detection algorithm worked by automaticall

One example related to software is augmenting decision-making before trying to replace it. PayPal's first major fraud detection algorithm worked by automatically flagging/delaying a small number of cases and having humans review them. I've tried to do something similar in my jobs when working on cases that are trivial for humans to judge but very difficult to completely specify automatically. Of course you can generally improve these systems with machine learning over time, so that a human reviews 1 in 10,000 cases instead of 1 in 100, but having a dual human-computer system is generally way more efficient at the beginning.