I had a product to help political campaigns canvas neighborhoods door to door. In the early days there was a "smart walk sheet" feature, which would magically pick the optimal doors to knock given how much time you had (e.g. 2 hours after work).
While I was building out this feature, I'd stay up late in the night manually selecting clusters of houses and setting them for the customer, so the next morning they'd be greeted with what looked like our backend systems auto-magically picking the perfect cluster of houses.
The bonus of doing it manually for a while is that I found a lot of edge cases I would have missed otherwise. The initial release of the finished feature was rock solid thanks to what I learned.
> The bonus of doing it manually for a while is that I found a lot of edge cases I would have missed otherwise. The initial release of the finished feature was rock solid thanks to what I learned.
This is one of the great benefits of doing things that don't scale. Reality has a surprising amount of detail, so to write code for a process you need to be an expert at that process. How do you become an expert at anything? You have to do it for yourself.
I feel like this is a fantastic example, as it’s one of those things a computer should be able to do perfect toy at scale, but for a quick and dirty solution on a smaller subset it’s something humans are really good at.
While I was building out this feature, I'd stay up late in the night manually selecting clusters of houses and setting them for the customer, so the next morning they'd be greeted with what looked like our backend systems auto-magically picking the perfect cluster of houses.
The bonus of doing it manually for a while is that I found a lot of edge cases I would have missed otherwise. The initial release of the finished feature was rock solid thanks to what I learned.