Machines (algorithms) are becoming ever more capable of the robust detection of complex patterns, as well as automating the estimation of causal effects through randomization. Do these developments devalue theories? The author argues that theory (and theorists) will continue to be most useful when data are scarce, or when even the contours of what is the "right data" are unknown. Where data are sufficient for algorithms to make useful predictions or isolate causal effects even without any causal understanding, theorists may still be useful primarily to explain and justify to fellow humans (for instance in law courts), and because we derive satisfaction from explanations. Both, however, may be transient phenomena.