Code is designed to solve problems. If someone can't read your code, they are going to have a hard time using it to solve their problems. If you don't document edge cases, modes of failure, intent and methodology... then it is worse code than it could be. Somebody won't be able to use it because they won't be able to tell if it solves their problem.

A few weeks ago I read a quote about someone saying that good code should be 'self-documenting' and that it should be obvious what it does. This is a red herring -- it assumes you wrote code that works perfectly. If there is a bug in your 'self-documenting' code, then how is someone supposed to know if it is wrong or not?