I don't think his main point is modifying the program using the interface. The important thing is the quick feedback loop. In his examples in Learnable Programing, you technically are editing in the same interface, but, really, you're just editing the code on the left and seeing the result on the right.

Additionally, I think that's actually the right way to do it. The result should be optimized for viewing, not for editing, and the editing environment should be optimized for editing, not for viewing. If the two can be merged into a tight feedback loop, then maybe we can get something of the best of both worlds.