Although manual query optimization can be challenging and tedious, attempts to automate this process with machine learning have suffered – including from long training times, poor tail performance, opaqueness, and integration challenges. To address these issues, Marcus et al. propose a different approach wherein they use reinforcement learning to improve an existing optimizer instead of learning an optimizer from scratch. Bao leverages tree CNNs and Thompson sampling to correct the PostgreSQL optimizer by providing a subset of recommended execution strategies to enable on a query-by-query basis. Although Bao slightly increases optimization time, it can be disabled for short-running queries.