把 Jev 塞进 chief-of-staff agent 做 worker 路由决策
作者用 Jev 给 agent 做「下一个 worker 是谁」的类型化决策，返回带置信度的选项，单次决策约 0.000042 美元，并列出 Browser Use 等实测案例。
i think Jev is the part most AI agents are missing every agent makes the same small calls all day: which tool, which worker, retry or stop, safe or not. most of them pay a frontier model for each one Jev by typesafeai does only that job. you give it the state and the allowed answers, it returns a typed decision with a confidence score. it can't write, and it can't invent an option you didn't give it I put it inside a chief-of-staff agent and gave it one question: which worker goes next? research 0.88 write 0.09 review 0.03 0.88 clears the 0.85 bar, so the job goes to research. anything below goes to a human the price is the whole point. Jev 1.13 costs $0.042 per 1M input tokens and output is free. at ~1k tokens per decision: 1 decision = $0.000042 10,000 decisions = $0.42 builders already proved it outside my setup: > Browser Use: flights found in 7s for $0.0039 > 1,018 research papers classified for $0.08 > 500 emails triaged for 3.5¢ > a Claude session cut from 1M to 86K tokens in ~1s the split: the LLM thinks Jev decides code executes the full Jev setup and the router are in the guide below
这条是引用着另一条写的：「长文：如何搭好 Jev 驱动的 agent 技术栈」