On data as infrastructure, and why the semantic web's academic instincts were pointed the wrong way — given at dConstruct.
The talk made a case that the networking protocols underneath the internet are best understood the way we understand roads: it isn't the road itself that transforms a place, it's the traffic the road makes possible. Increasingly, the traffic on the network isn't web pages — it's data, generated by how people actually use services, both online and through data-enabled devices in the physical world.
I'm in favour of a web of data by any means necessary — not academic purity, just data marked up, describable, and out in the world.
The provocation in the room was aimed at the Linked Data community specifically: a concern that it was being driven more by academic research than by real business need, with an expectation that everything be perfectly formed before it counted. The talk's alternative was scrappier and, I think, more honest about how the web actually grows.