It seems like he's not arguing against query-based compilers per se, but rather he's arguing against the thinking that often comes with it. Namely, that you can amortize an expensive compilation process across incremental changes (which often doesn't work in practice because of the avalanche effect).