However, triggering multiple cameras at once at the exact same time was a similar experiences with additional trigger lines and multiple discussions with the manufacturers application engineers.

100% this; synchronisation challenges are the bane of my existence. There are also a lot of cases when it would have been really helpful if data output from a device were timestamped (even with something incomplete like "number of milliseconds past the minute"), rather than having to timestamp it on receipt and make some iffy assumptions about latency.

Depending on the command the byte order also had to be swapped which was not documented at all and I had to find it by trial and error.

"Determining endianness is left as an exercise for the reader" type logic, honestly lol.

So I’m not sure if it’s the military grade that makes it such an experience. Industrial grade seems like a similar experience.

It's quite plausible (and disappointing to hear) that it's a broader issue with industrial grade hardware; my own experience is limited to products specifically aimed for military/defence applications.

I guess a lot of consumer devices also do shit like that under the hood but hide it in firmware and drivers.

Yep, but I think there's also a broader point about work that is niche or "cutting edge" versus industry standard. The example that comes to mind is the quality of code written in academia versus in industry, the former often being significantly worse. I think part of it is that code quality is far less emphasised compared to the higher level technical concept it's used to demonstrate. I honestly believe that a sizeable chunk of the professors and lecturers from my university would struggle to produce work that is "up to standard" in a regular software eng role, at least initially.