Some developers are lazy. Some probably don't know that that is the right way to do it. There is a lot of legacy code from when checking user agents was more acceptable. It is much more difficult for server code to know the capabilities of the client (although in practice this isn't usually much of an issue).

Also, sometimes the feature exists so the feature check is positive, but there is a bug in one browser that breaks your functionality, so you put in a user agent check. Then the bug gets fixed, but the user agent check isn't removed for years. I've seen that happen many times.