The difference is that coding agents can run the code that they produce, fix any bugs, build tests and generally demonstrate that it works.

You can't do that for images and text.