I'm not sure that's actually a great example. I had a project a while ago where I needed to extract certain information from Word and Excel files, and it was less work to just write my own parser (it's just XML in a ZIP file) that got exactly the information I needed than to figure out all the complexity of using a full-blown docx/xlsx parser. It ended up being 100 lines of Haskell, and half of that was imports.
https://gist.githubusercontent.com/duairc/db3e99a7808668e84e...
Edit: The docx part of it is only 10 lines of code.
The media wiki parser is a perfect example of what can go wrong with simple solutions.
You immediately went for two examples that are precisely not the kind of thing the author is talking about. Consider leftpad.
It's also not necessarily required that you understand all of what you're doing down to the logic gates, but if you don't (to crib your example) understand some of what docx parsing may entail, or can't actually write any sort function (nor understand how to compare them), you're stuck with black boxes at even the highest levels, and then you're very likely to be exposed to the article's defined fatigue as you continually jump to whatever looks shiniest, lacking the understanding with which to properly compare your options.
Writing my own sort routine? Pointless but pretty harmless.
Writing my own docx parser? Sure, that will be one mythical man century of work.
Sometimes we are using black boxes because we have created so much accidental complexity that we can't do better than "this works, we don't know why and we dare not change it". The browsers js must target are some of the biggest black boxes even though they are open(ish) source.