Example: Most systems have uncritical components which are nice to have but nobody wants to code because "we-can-help-ourself-without-X-by-doing-this-manual-step-but-it-will-be-better-if-someone-does-X-somewhen-in-the-future-but-for-now-we-can-work-without-it"
E.g.: I have polished tons of report due to AI usage; financial reporting usually ends up somewhere with "large loops collecting data and copying it over to the report" - those things are usually "minimum-implemented" and nobody cares since "its good enough to do Y". Its mundane and frustrating to work on those code snippets.
Output streams in general are much more easy now: Lately I had to put together a component to display some numbers in a Windows App, this comes along with things like scaling the data/window correctly etc.; I would have had to put it in windows drawing calls. With a few prompts, I was able to get a perfect rendering component of 1500+ LOC which does exactly what i want and the result is perfect. (even better than in the reference app that we use to compare data output)
There are so many things where LLM are just a boon!