We're doing air gapped development (and even single-machine deployments) with web tech by just also including a server component with the client or on the air gapped network.
There is one instance where we still use machine control / HMI non-web, and that's when there is no TCP/IP stack available. Doesn't happen that often anymore.
I'm not talking about faking a web app, just use a normal webbrowser and point it to the internal server. It's what we do in industrial automation as well (of course with watchdogs so it doesn't end up displaying a dead page if windows is used and the HTTP channel shits itself).
We did this with meta-refresh and progressive loading (https://blog.codinghorror.com/the-lost-art-of-progressive-ht...) earlier on as well. Benefit is that you get that client-server model without having to do all that much special for the client side, works on any client that has a working browser (even ancient IE5/Netscape4 rendering engines), even embedded systems and weird stuff like NetFront and Opera Mini Java (which runs nearly everywhere).
A lot of actions are essentially reading for status and anomalies, and every now and when adjust or control things (not real-time actuator control like a robotic arm or something like that, that requires hard realtime loops for us), anything bigger than that you'd do on a full desktop style system anyway and you can just use a modern web app in the browser for that. Those then also work on tablets (wallmounts! yay!), PDAs (well mobile phones without modems today) and other devices.
There is one instance where we still use machine control / HMI non-web, and that's when there is no TCP/IP stack available. Doesn't happen that often anymore.