In the words of Jeff Atwood, "Give me parameterized SQL, or give me death" - and HTML escaping is little different. My fellow "software engineers" are human. They will screw up. That's simply a matter of human nature and statistics. Even if we only employ godlings, we'll eventually hire a human intern.

"It happens to be safe for the moment because it's a constant" isn't a high enough bar. I've refactored plenty of string constants to new locations, and then to dynamic strings. If that's the bar you're using, every such refactoring is now a security critical change that requires a thorough security audit.

There are uses of innerHTML, but the post you're replying to acknowledges that - "almost never". If you're building the APIs that enable safe DOM manipulation, you have some use cases. If you're serving HTML fragments directly from the server, you have some use cases - although I'd hope you're using safe APIs serverside if it's dynamic, instead of hand-building HTML in code!