HN had another article a couple weeks ago asserting you shouldn't add business processes to your company unless they had an expected value of increasing foo by at least 10%. The author argued this was a good way of avoiding creeping bureaucracy. It's striking how similar that is to this article's thesis re code complexity.