一名开发者利用Spotify的Portal产品和AiKA Modes技术，通过将AI编程任务路由至成本更低的模型，成功将Claude Code的token使用量削减90%1。该方案将重复性I/O工作和代码生成任务从高成本模型转移到Gemini 2.5 Flash等轻量级模型，有效缓解了AI编程成本快速上升的压力1。
该系统采用两个专用模式实现任务分发1。bulk-reader模式负责文件读取与分析工作，而code-writer模式则用于生成测试、配置和类型存根等可预测输出1。通过Claude Code插件的自动路由机制和PreToolUse钩子功能，当文件超过配置阈值（默认350行）时自动阻止直接读取，确保任务被正确分配至相应模型1。
这一优化方案应对了日益加剧的成本挑战1。根据行业数据，到2028年AI编程成本预计将超过开发者平均薪资，其中四分之一的工程领导者已每月每开发者花费200至500美元，部分企业支出甚至超过2000美元1。该开发者的实测结果显示，通过bulk-read和code-write两个脚本分别处理委托任务，平均token节省达到90%1。
A developer has shared how Spotify's Portal product, using AiKA Modes, successfully reduced Claude Code token consumption by 90% by routing AI programming tasks to cheaper models like Gemini 2.5 Flash 1. The approach addresses mounting concerns about AI coding expenses, as some engineering leaders already spend between $200 and $500 monthly per developer on AI programming tools, while others exceed $2,000 1.
The system employs two specialized modes to distribute workload efficiently 1. The bulk-reader mode handles file reading and analysis tasks, while the code-writer mode generates predictable outputs such as tests, configuration files, and type stubs 1. Automation is achieved through a Claude Code plugin that automatically routes tasks based on file complexity, with a PreToolUse hook preventing direct reads on files exceeding a configurable threshold—set to 350 lines by default 1. According to the developer's testing, this configuration delivered an average token savings of 90% 1.
By 2028, AI programming costs are projected to surpass the average developer's salary, underscoring the urgency of such cost optimization strategies 1.
评论
还没有评论，欢迎留下第一条。