别再数 token，开始数交付成果
原标题：Stop counting tokens, start counting what ships
AI 摘要
MIT Technology Review AI 刊发 HPE 赞助内容，指出 AI 编程代理单任务消耗 token 可达普通对话的 10 至 50 倍，按 token 计费使高价值工作流成本高企，Uber 据称 4 月已耗尽 2026 年 AI 预算，GitHub Copilot 转向按量计费。文章主张改用「每个已合并 PR 的成本」衡量投入产出，并通过网关按任务复杂度路由、将常规工作交给自托管开源权重模型来降低成本。HPE 称其混合 AI 平台将 150 多个内部应用的常规负载路由到本地开源模型，token 成本下降 30% 至 50%。
正文节选
Sponsored Provided byHPE When a developer gives an AI agent a coding task, the agent does more than answer once. It plans, generates code, runs tests, and repeats the process to correct errors. A single task can use 10 to 50 times more tokens than a basic chat prompt. The more useful the agent becomes, the more calls it makes. With per-token pricing, the workflows creating the most value can also become the most expensive. The financial pressure is already visible. Uber reportedly exhausted its