Public programs often delegate spending decisions to agents with private incentives to overspend. One way to mitigate this is to institute a quantity limit prohibiting or restricting spending above a cap. This paper considers the efficacy of using quantity limits for health care in the context of a large public health insurance program in the United States. “Hard” limits, which restrict all spending above the cap, reduce spending twice as much as “soft” limits, which allow exceptions for high-need patients. While the motivation of allowing for these exceptions is to screen out low-need patients but preserve care for high-need ones, soft limits appear to worsen targeting relative to having no limit. After a soft limit is introduced, high-need patients are no more likely to receive care above the cap than low-need ones. Instead, the exceptions process rewards patients who go to larger providers, who have greater administrative capacity and are more adept at getting around the cap. Further, we find that the savings from a soft limit deteriorate relatively quickly, dropping by over half within three years. This is driven by adaption by smaller providers, who become more adept at bypassing the cap. We show that providers learn from past attempts to go above the cap. Our results highlight the pitfalls of using quantity limits to curb overspending when the cost to bypass them varies across agents, and when agents can learn over time how to bypass them. (with Ashvin Gandhi (UCLA).)