> I think a lot of it is "nobody has bothered building it yet" vs security.

All of this takes considerable time, money to build and after that you need to get people to buy into it anyway. Large billion dollar software companies have difficulty doing this. If you think it is so easy, go away and build a proof of concept.

BTW They have implementing sand-boxing in most desktop operating system. It is often a PITA. Phone like permissions model already exist in Windows, Linux and I suspect MacOS in various guises.

For development there are various solutions that already exist.

e.g.

https://code.visualstudio.com/docs/devcontainers/containers

So these things already exist and often people don't use them. The reason for that is that there is usually reduces usability by introducing annoyances.

> Eg Qubes runs everything in Xen isolates - which is a wildly complex, performance limiting way to do sandboxing on modern computers.

It exists though today. If I care about security enough, I am willing to sacrifice performance. That is a trade off that some people are willing to make.

> There are much better ways to implement sandboxing that don't limit performance or communication between applications. For example SeL4's OS level capability model. SeL4 still allows arbitrary IPC / shared memory between processes. Or Solaris / Illumos's Zones. But that route would unfortunately require rewriting / changing most modern software.

If you solution starts with "rewriting most modern software". Then it isn't really a solution.

BTW what you are suggesting is a trade off. You have to trade resources (time and money typically) to build the thing and then you will need to spend more resources to get people to buy into using your tech.