The problem is many times, you can't really reason through the issues your encountering. For example, if you are trying to talk to a server with an undocumented API flow, or you're trying to coax an in-house library to life, you have no option but take baby steps and test every single change to see if you made progress. In environments (like C++) where the compile times are high, and it's pretty easy to introduce almost impossible-to-trace bugs, I'm often tempted to do a bunch of changes, and then test them all at once, which will invariably result in something breaking and me having to spend an inordinate amount of time to figure out what went wrong.