复杂应用需将任务规格转为环境与评估
Correct.
AI 导读
正确。 不过，对于复杂应用，你的 jobs-to-be-done 规格应尽快转化为环境 + 评估（最好可验证）。这对复杂的既有系统或工作流往往很难。 这种设置让你能针对自己的用例对 harness + 模型进行爬山优化。 否则你迭代 agent 的能力就会受限。在较简单的应用里，这不太成问题。
正文
当前显示原文，中文全文译文尚未就绪。
Correct.
However , for complex applications, your jobs-to-be-done spec should quickly be translated to environment + evals (Ideally verifiable). this is often very hard for complex pre-existing systems or workflows.
This setup lets you hill-climb the harness + model on your use cases.
else your ability to iterate on the agent becomes limited. in simpler applications this is less of a concern.