I work performing a lot of GIS (geospatial information systems) work for emergency services. Sometimes, I produce maps. Sometimes, someone wants to know where they ought to locate a five million dollar building for maximum results.
My computing interests lie in automating tasks previously performed by humans, ETL, and analysis. I was doing web stuff back in the 1990s and went through so very many CRUD apps, but this is where my interests lie. What this means in my current job is that I find time-consuming tasks done by my predecessor and say, "Let's make this faster." Or I can look for subtle errors.
Just as an example, because I am routing emergency services to houses and workplaces, I have a network of streets to deal with. Imagine them as a graph, where most of the data is attached to the edge rather than the node. Now, people have drawn these streets in using tools like ArcMap, but there can be problems. How many different kinds of mistakes might exist in connecting roads together? Try to enumerate the cases.
I have found ten thus far. Each case is something that, if a human eyeballs it up close, they could say "oh yes," but it is not readily apparent. And so I have written code for a number of these cases that takes my human domain expertise of what is wrong with something and translates it into something even a rock (silicon) can understand. And the end result of finding these little mistakes is very real: help gets to someone just a few seconds faster.
I interface with organizations that are not other IT organizations: property records, postal records, phone records, and so on. Sometimes this means wrangling proposed plats from developers. It's all real stuff, and it's all done in service of helping people who have immediate needs.
It's very different from being in a room full of people who have little stickers on their Macs making Agile noises about whatever random shifts have occurred in the "client??" (and doesn't that word encompass a multitude of sins)'s latest wishlist.