The "theory" of a program is supposed to be majority embedded in its identifiers, tests, and type definitions. The same line of reasoning in this article could be used to argue that you should just name all your variables random 1 or 2 letter combinations since the theory is supposed to be all in your head anyway.
Indeed, it's quickly obvious where an LLM is lacking context because the type of a variable is not well-specified (or specified at all), the schema of a JSON blob is not specified, or there is some other secret constraint that maybe someone had in their head X years ago.