Agent日均零成本：任务分层路由实战

昨天，一整天的 agent 任务——阅读、总结、分拣堆积如山的资料——API token 费用为 R$0。不是 promo 赠送额度，不是免费层的把戏。仅仅是一个架构决策：路由。

常见的错误是把 AI 当作一个天才员工来用：把所有任务都扔给最强大的模型。但事实上，agent 工作日中约 80% 的工作量并不困难。阅读、总结、分类、提取字段、格式化。用前沿模型的价格做机械性工作，就像雇一个外科医生来换灯泡。

给每个任务分配它真正能完成的最低层级：

Rung 1 — raw volume (harvesting, extraction, mass classification): small local model on your own machine. Token cost: zero.

第一级——原始批量（采集、提取、大规模分类）：使用本地小型模型。Token 费用：零。

Rung 2 — reading and synthesis (long material, comparing sources): mid-size local model. Still zero.

第二级——阅读与综合（长文本、多源对比）：使用本地中型模型。仍然是零。

Rung 3 — hard decisions (final verdicts, judgment-heavy analysis, critical code): the frontier API, and only there. You pay for the rung, not the ladder.

第三级——困难决策（最终判断、需要大量判断的分析、关键代码）：使用前沿 API，仅限此处。你只为所需的层级付费，而非整个梯子。

Who picks the rung? A simple classification rule is enough to start — or the local model labels the task itself before executing it.

谁来选择层级？简单的分类规则就足够起步——或者让本地模型在执行前自己标注任务类型。

What a R$0 day actually is

零 token 日的真实含义

A zero-token day is a volume day: everything landed on rungs 1 and 2. The local model ate the material, agents logged everything to files, results came back ready for review. Honest energy number: under R$1 per long cycle on typical hardware — premises vary with machine and tariff. R$0 in tokens isn't R$0 absolutely; it's trading per-request cost for the fixed cost of a machine you already own.

零 token 日其实是批量处理日：所有任务都落在第一级和第二级。本地模型处理了所有材料，agent 把结果记录到文件中，最终返回可供审查的结果。诚实的能源数字：典型硬件上每个长周期低于 R$1——具体取决于机器配置和电价。Token 费用为零并不意味着绝对为零；这是在用每次请求的成本换取你已经拥有的机器的固定成本。

The two warnings gurus skip

专家们跳过的两条警告

Cheap that redoes the work three times is expensive. If the small model fails and retries, your cost per completed task climbs — price-per-token is the wrong metric to decide with. Measure cost per completed task, always.

便宜但返工三次就变成贵的了。如果小模型失败并重试，每个完成任务的成本就会上升——按 token 单价来决策是错误的指标。要始终衡量每个完成任务的成本。

Routing quality down without a reviewer manufactures error in silence. Every step down the ladder needs quality sampling: periodically, a human or a second agent checks a sample of what the cheap model produced. And some things never go to rung 1 — final decisions, anything published under your name, anything you can't verify afterward.

缺乏审核的情况下降低路由质量会在无声中制造错误。每一级下移都需要质量抽检：定期由人工或第二个 agent 检查便宜模型生成的样本。有些东西永远不会放到第一级——最终决策、任何以你名义发布的内容、任何你事后无法验证的东西。

Mechanical and voluminous → local model.

机械性且大量的任务→本地模型。

Hard decision → frontier API, only there. It's worth paying, because it's the price of being wrong less.

困难决策→前沿 API，仅限此处。值得付费，因为这是减少错误的代价。

Always measure cost per completed task, never per token.

始终衡量每个完成任务的成本，绝不以 token 数量为准。

Full disclosure: this is part of the working method behind a US$15 one-time template I sell — a second-brain vault starter with the agent rules, derived-index and search tools included, plus a short ebook on what NOT to do. The routing ladder works fine with nothing but your own setup.

完整披露：这是我出售的 US$15 一次性模板背后的工作方法的一部分——一个 second-brain 保险库启动器，包含 agent 规则、衍生索引和搜索工具，外加一本关于不该做什么的简短电子书。路由阶梯完全可以仅用你自己的设置来运作。

Second Brain Starter — US$15 one-time

Second Brain Starter——US$15 一次性

Questions about where routing breaks in practice are welcome — that's the interesting part.

欢迎提问路由在实践中哪里会失效——这才是有趣的部分。

Cross-posted from Oroboro Labs — this post is the canonical version; the original carries the full method notes.

转载自 Oroboro Labs——这篇文章是规范版本；原文包含完整的方法笔记。