Even the cleanest code will only ever tell you how something is done, and not why.

The best-factored simply-named controller source or whatever else will not, in isolation and without documentation, ever tell you why we are asking the computer to do the things it does.

The problem with this is that somebody (rightly or wrongly) will see the code one day, and infer what they think it is doing, and change it, and break things elsewhere in the world.

If you don't want that to happen, you are forced to only ever write code that matches the exact logic of what already exists--and because you, rockstar ninja cowboy that you are, don't have any documentation including business use cases or logic for a particular hack or inelegance, you must then look on and only make the mildest of changes to the sacred monolith.

Or, you know, you could document your projects and software like a real engineer, and others can later check that they can make changes with impunity--or even better, throw away chunks of unneeded code wholesale because their time is (according to the docs!) past.

EDIT:

This whole "The code is the documentation hurrrr" meme is really annoying--it can only express the hows and not the whys. There is more to software than simply the steps the computer takes to transform data.

If there wasn't, we'd still be using assembly and Perl every day.