So it seems like this is mostly to tie content back the original agent, but doesn't actually do anything to verify the agent. Is that right?

True, this demo does not verify the agent or it's owner, but that's not necessarily what it was intended for (hence the "our take" part). We would view this as an additional trust layer (although a big one we'd argue) in a "know your agent" product. It was mostly intended to show how you can prove an agent of yours created the content seen on a users screen, verify it's integrity, and have this all work in a streaming system.