📄 中文摘要
许多团队在初期使用提示词库时，会发现其效果显著，能够生成高质量、相关的输出。最初，提示词数量较少，使用场景有限，且由同一人编写和运行，反馈循环紧密，问题能及时解决。然而，当这些提示词开始驱动实际业务流程，例如生成邮件、总结工单或起草文档时，问题便逐渐浮现。随着提示词数量的增长和使用范围的扩大，维护和管理变得复杂。不同用户、不同场景的需求差异，以及底层AI模型更新带来的兼容性问题，都可能导致提示词在生产环境中表现不稳定，甚至完全失效。这种从实验阶段的成功到生产环境的失败，揭示了提示词库在规模化应用中面临的挑战，包括版本控制、性能监控、以及适应性调整等关键问题。
📄 English Summary
Why Prompt Libraries Always Break in Production
Many teams initially find prompt libraries highly effective, generating clean, relevant, and useful outputs. In the early stages, the system is small, use cases are limited, and the same person who writes the prompts also runs them, allowing for quick adjustments. This tight feedback loop ensures initial success. However, as these prompts begin to power real-world workflows—generating onboarding emails, summarizing support tickets, or drafting documents—issues start to emerge. The transition from experimental success to production-scale application reveals significant challenges. As the number of prompts grows and their application expands, maintenance and management become increasingly complex. Discrepancies in requirements across different users and scenarios, coupled with compatibility issues arising from underlying AI model updates, can lead to prompts performing inconsistently or failing entirely in a production environment. This highlights critical problems in scaling prompt libraries, including version control, performance monitoring, and adaptive adjustments, which are often overlooked until the system is fully integrated into operational processes.