If I'm writing a paper, it's great for collaboration if it can be written in the browser. Being able to preview my paper in the browser sounds nice too, but to publish it I need to be able to compile it into a PDF.
His response stuck with me: While scientific research is continuous and never complete, the whole point of publication is that it is a record in a fixed point in time. The results and discussion cannot be altered, or reused: it is a milestone, a historical record of a novel contribution to knowledge. Sure, you might reanalyse your data, rework an image with new parameters what have you, but the nature of sharing your work with your scientific community requires it to be static.
We are a long way from that now, but it is appealing.