Not to be an old programming fart (but, hey!), but this comes up about every 5-10 years. Someone has created a system for automatic program generation that is going to replace programmers (4th generation languages, anyone? How about "The Last One"?), and it turns out to have only limited usefulness.
Of course, code generation programs exist. They've existed almost as long we've been programming computers. The most common are assemblers and compilers, which take in text specifications and generate running code (or sometimes bytecode to be interpreted). And if you stop and think about the difficulties that most of us who code have with making source code that we write produce running code that meets our needs, you can immediately see the issues with replacing or bolting on top of that system a 'source code generation' system. It can work very well as long as you don't exceed what it can actually do and only if the code generation system itself is well-written and reliable. (This is why developers feel a sense of betrayal and anger with compiler bugs more than any other kind of tool bug.)
So, yeah, like strong AI, self-coding systems are always 5 to 10 years out and have been for half a century.