Computers are good at solving recurrent problems. Much better than humans are. And still, we keep them dumb with a set of simplest heuristics when it’s about monitoring complex infrastructures, leaving the largest part of the job – issue recognition and analysis – to ourselves. This might work with a server or two, but definitely won’t in a larger setup, even if we convince ourselves it would. We need new approaches to monitoring our systems that combine the best of software engineering and mathematics. In this talk, I will explain the vision and the targets towards it.