[treesknees]

That's been the experience I've had as well. One company I worked for around 2015 had 5 or 6 mainframes (which they kept upgraded to the latest hardware on an IBM lease program every few years), and it was well understood that the number of distributed computing servers required to keep up with the volume and resiliency of the mainframe would be too great to even consider.

Another company I interviewed with did just what you said, they built REST APIs and other "modern" interfaces to the mainframes so that 99% of app developers in the company don't have to learn anything about the mainframe.