In 1990, during a summer internship at GE, I wrote a program that scheduled electric motor production at like 7 factories around the country. You would download the next weeks worth of orders from their timeshare system, which seemed ancient then. Someone had done a time-motion study of each motor model and determined how much time each motor needed in different stations. They wanted each day's schedule to use roughly the same amount of time in each station.
This was after my first year of college and I really knew nothing about algorithms. I wrote a horrific program in BASIC on a PC that did what I know now is a greedy bin packing solution. It created a least squares metric and tried moving and swapping orders until nothing improved the metric.
I was shocked to hear that 20 years later they were still using that same program.