📄 中文摘要
Ultra Lab在Google Gemini 2.5 Flash的免费层上运营四个AI代理（UltraLabTW、MindThreadBot、UltraProbeBot、UltraAdvisor），每日配额为1500个请求（RPD），实现完全自主的社交媒体推广。然而，在实际操作中，发现大量令牌在无输出任务中被浪费。通过对令牌的审计，识别出三个主要的浪费源，并应用了特定的优化技术，以减少资源浪费，提高效率。
AI代理令牌优化实践：如何在四个代理中减少40%的浪费
📄 English Summary
AI Agent Token Optimization in Practice: How We Cut 40% Waste Across 4 Agents
Ultra Lab operates four AI agents (UltraLabTW, MindThreadBot, UltraProbeBot, UltraAdvisor) on the free tier of Google Gemini 2.5 Flash, with a daily quota of 1,500 requests per day (RPD), achieving fully autonomous social media promotion. However, significant token waste was discovered in zero-output tasks during practical operations. A token audit process identified three major sinkholes, and specific optimization techniques were applied to reduce resource waste and enhance efficiency.