Built-in names are essentially reserved words, and there are dozens of them. The @ prefix ensures you don't step on user's variable names, and that you can add new built-ins without making breaking changes.

Why aren't they simply namespaced, like `core.some_function`?

Because there'd need to be a magical exception for the "core" namespace that makes it not just a file of Zig code somewhere like every other module.

Isn't there already a magical exception for "root" and "builtin"?

not really, those are two modules that are always available to you, but you still have to import them like any other Zig module `const builtin = @import("builtin");`

5 characters to type instead of 1.

My belief is that it is about builtin functions that are provided by the compiler versus part of the standard library. They are documented in the language reference [0] versus in the standard library documentation [1].

[0]: https://ziglang.org/documentation/master/

[1]: https://ziglang.org/documentation/master/std/

> I don't see why it needs to exist

The symbol namespaces builtins, as those are the only identifiers that aren't declared in the file (either directly or by using `@import`).

Although namespacing them keeps them out from under a programmer's feet, which is a significant benefit, it does seem like this would make it harder to find stuff. @cmpxchgStrong, @wasmMemorySize and @embedFile are completely unrelated, but since they're all builtins they're neighbours.

This is sort of an issue we already deal with in other languages, and imo it's not a huge deal in those. Personally I find @ more reasonable than __builtin_

one side benefit is that at lot of @ code is "dangerous shit" so it draws your eye during code review. You will want to code review the "dangerous shit" that GPT-5 gives you.