Some say that the ideal size of an individual function in a codebase is related to the amount of information you can hold in working memory. Maybe the ideal size for a library is the amount you can fit in an LLM context window?