Modern light sources have dramatically increased available photon flux at beamlines, allowing greatly increased measurement speeds for many techniques and enabling high-throughput modes. The limiting factor in optimizing such processes is typically driven by variations in sample composition leading to differing requirements for measurement time to achieve optimal measurement statistics across all samples. When human-driven, such dynamic sample-by-sample scheduling operations are at best tedious and at worst unoptimized or mistake prone. Reinforcement learning methods offer a path to autonomously drive such high-throughput experiments, and the Bluesky suite allows for their ready integration. In this contribution we will discuss how reinforcement learning aids high-throughput data collection and practical considerations for implementing these methods on a beamline.