When I started out programming in C#, I used reflection sometimes to circumvent the language’s design and restrictions. This resulted in brittle and hard to reason about code. Reflection should never be used to do this.

So I do think you’re right that reflection can (and will) be abused by beginners if present as a language feature.

With that said, I don’t know much about the internals of the C++ compiler, but having built a simple reflection system for C++, I think the important thing is just being able to serialize and deserialize POD structs to and from some representation (e.g. json). For more advanced data, such as images or specific binary (file) formats, it’s easier to write custom writing / reading, encoding / decoding logic.