I've been sticking to an older version of MPLAB X specifically to avoid using Harmony. For the most part we've been using PIC32MX chips and I've had a good experience with the deprecated PLIB functions, been able to do everything I need with those.
I think we had a project that was looking at an MZ part and noticed that the pre-Harmony versions of MPLAB did not support those, if I remember correctly.

I once did the initial bring-up of one board and the layout of another, each with the same PIC32MX on board, and passed them both to a colleague who actually chose to work with Harmony and apparently didn't have any issue. I tried installing a newer MPLAB and Harmony just to give it a try for the sake of getting up to date and getting more uniformity with what my colleagues are doing, but I took a short look at it and went back to the old MPLAB X.

Select option -> Generate Code -> Build -> Build fails.
Select different option -> Generate Code -> Build -> Build fails.
Try yet another option -> Generate Code -> Build -> Build fails.

Harmony is a giant framework where you select from hundreds of
options (drivers, TCP stack, etc), then generate your application's frame.
It favors a monster-loop state machine, at a point where the application
complexity (something needing PIC32 with 2MB flash and 512kb RAM)
makes this a loosing proposition.
RTOS options were pasted on later and appear to be 2nd class citizens.
Its a bit different than just pulling the drivers you need from
a vendor-provided library.
And it seems to generate a frame that doesn't build (for anything I've tried).

Is anyone using this successfully?
We were asked to quote a job using a newer PIC32MZ, though some of
the chip's attractive features have bugs making them potentially useless
for the customer application (like the crypto block).

Thanks!
Best Regards, Dave