Traditionaly with PHP, C#, Ruby, Elixir, etc, your backend and frontend was tightly coupled, your code has complete access to backend resources, and the view/template would map the state to a HTML document.
Now, for the sake of interactivity, the view has effectively moved from the backend to the frontend with the introduction of SSR/SPA, JSON everywhere and code duplication so that they can all talk and understand each other. It's cool when it works, but it is easily at least 3x the amount of work.
There is of course the argument of "just write both the client and server in JS/TS and use a monorepo", that brings its own challenges. Limiting the backend stack to what browsers support is not great, especially when Node.js is single-threaded with cooperative scheduling. No, lambda functions don't solve this entirely, and they're freaking expensive for CPU time. Honsetly, other ecosystems have it so much simpler IMO, even if it isn't as flashy.
I say all this as a svelte developer, guilty.