The most evergreen example of this is and probably always will be the rollout of VisualEditor, their WYSIWG editor. It was completely unwanted, was proven to be not ready for general use (due to all the problems that plague "combining a WYSIWG editor with a code view"), yet forcibly deployed on all Wikipedia languages without much room for warning or feedback. After that, the WMF promised communities the option to opt-out if they didn't want it... before violating it on the German Wikipedia because they voted to not roll it out. To my knowledge, VisualEditor hasn't improved over the years (beyond "rewrite Parsoid into PHP"), but editors have just gritted their teeth. That's the extreme example, but the WMFs relationship with building features for Wikipedia is... pretty much in a vacuum from what users want. A lot of their projects seem to be conceived by WMF managers first, without consultation to their existing community of volunteers.
Which is I guess fine for moonshot projects like Wikifunctions (since they don't involve the volunteers before release) but results in a lot of friction when a managers pet project is forcibly rolled out on all language Wikipedias. Very few of their big extension deployments have gone smoothly with the community. I distinctly remember the Minerva skin (mobile Wikipedia) needing tons of changes and improvements before it was usable as well, after it was already rolled out and caused a ton of negative fedback.