The presentation discusses whether reinforcement learning (RL) is reaching a tipping point for production use. While RL has achieved superhuman performance in research domains like games, its use in production is still rare due to challenges like high data requirements, online training limitations, and large state/action spaces. However, the talk notes recent progress in areas like distributed training, offline RL, and embeddings that reduce complexity. It identifies three patterns seen in successful production RL: parallelizable simulated tasks, low temporal problems like recommendations, and next-generation optimization. The presentation provides tips for simpler RL approaches and validation challenges when deploying RL models.