What I didn't see answered in the article was how content is discovered.

The only way we are able to productively use git is because there is a convention to have some state in a non content-addressable location (.git/refs, .git/HEAD, etc...).

Saying that IPFS could replace the web means either: 1) Introducing shared mutable state; or 2) full knowledge of everything on the network.

I'm guessing that the existing web is what provides that layer right now. Is there any work going on for novel IPFS-based content discovery mechanisms?

Another thought: Given the content-addressable, immutable nature of this graph, how does one discover that a new version of something is available without a central authority? How could we discover the tip of a blockchain with IPFS alone?