Query optimizers are an important part of modern database management systems, responsible for transforming the user’s declarative query into an executable plan. Because the space of executable plans is so large, traditional query optimizers narrow the search space using complex hand-crafted heuristics, which must be manually tuned on a per-application basis for maximum performance. Both the maintenance of these heuristics and their manual tuning require significant human effort. In this talk, I will present two new approaches to query optimization that employ machine learning to entirely remove — or at least significantly lessen — the need for such heuristics. These learned approaches outperform traditional heuristics with a small amount of training overhead, while also automatically adapting to changes in the user’s workload or even hardware. Finally, this talk will highlight the opportunities for applying similar learning techniques to systems more broadly.