In a comment, Karsten Becker made the argument that a library footprint is more likely to be "correct" -- in quotes because "correctness" in this context is relative -- since it was looked at by more people. CircuitHub also claims this. I want to challenge the notion that this is, in fact, a good thing.
By creating our own footprints we end up getting exactly what we want, how we want it -- which is, as it happens, exactly different from what everyone else wanted in the past -- and, more importantly, it is another opportunity to get more familiar with the device. It's one thing to forget a bar on the pin's labelling, but quite another designing in the wrong component! Acknowledging each pin's existence forces us to think about its function, and discover well-hidden caveats and constraints buried somewhere in the datasheet.
We're also far from guaranteed that the device in the library is, in fact, "correct". Maybe only the creator looked at it? Maybe this creator forgot to correct a mistake after their board needed a re-spin? Maybe a lot of people looked at it and didn't notice a problem? Maybe what was good enough for everyone else (location precision, for example) isn't good enough for me? Maybe the mistake was never relevant to anyone, so it wasn't noticed? Maybe I'm going to use a different revision of the part? Maybe the footprint was created from an old version of the datasheet? Too many doubts considering what manufacturing mistakes cost me. I'd verify the hell out of any footprint no matter who looked at it -- even if it was me six months ago -- so the amount of time a component from a library could potentially save me is quite small.
Finally, accountability is kind of a hobby of mine, and I like owning my own mistakes. I would take ownership of any mistake on a board I design no matter the source, so I'd prefer the mistakes were mine, and not imported ones.
So this leaves us with the original argument -- creating fresh footprints should be easy; libraries don't cut it, at least not for me.