AI in healthcare is moving from answering questions to doing work,and that changes the economics. At 50 users, an agent can look inexpensive. At 50,000, every extra token, context window, model call, and tool invocation becomes a business decision.The industry is learning to optimize the meter. What's actually driving those costs, and where the industry is innovating? This session covers what drives token cost (context length, retrieval strategy, model choice, retries), how the industry is bringing it down (caching, distillation, routing, smaller task-specific models), and why the companies that solve for cost discipline, not just accuracy, are the ones that will actually get to scale.