Sarah Guo

a lot of applied AI companies were playing katamari - get big enough that you know which capabilities are actually worth improving, can afford to train (and win) win your domain, and have immediate demand against which to amortize that investment