I think the point of the line of questioning is to illustrate that "tools" like a code interpreter act as scratch space for models to do work in, because the reasoning/thinking process has limitations much like our own.