De-bloat may be an interesting AI opportunity because the existing software serves as a detailed spec (we have 40 years of evidence that humans won't suddenly start writing debloated code)