Anthropic Agent架构指南：子代理与主循环的取舍权衡
Banger guide by Anthropic.
Agent开发者必读，Anthropic官方给出了从子代理迁移到主循环的具体评估方法与避坑指南，直接指导工程落地。
Banger guide by Anthropic.
Anthropic 的 Banger 指南。
While frontier models have improved at context management, subagents still help in some cases (e.g., parallelizing research tasks with their own context windows).
尽管前沿模型在上下文管理方面有所提升，但在某些情况下（例如使用各自的上下文窗口并行化研究任务），子代理仍然有帮助。
I still use them, but mostly for complex research work via dynamic workflows and /goal.
我仍然使用它们，但主要用于通过动态工作流和 /goal 进行复杂的研究工作。
For most agent workflows, folding subagents into the main loop via tool calls and skills makes a ton of sense (as Anthropic recommends). That's because newer models are more persistent and better at managing large contexts. Subagents might be getting in the way now, since models can handle much of what subagents were used for. I see a similar pattern with memory (another day).
对于大多数代理工作流，通过工具调用和技能将子代理折叠到主循环中是非常合理的（正如 Anthropic 所建议的那样）。这是因为较新的模型更具持久性，且更擅长管理大型上下文。子代理现在可能会成为阻碍，因为模型能够处理原本由子代理承担的大部分工作。我在记忆方面看到了类似的模式（改天再谈）。
But you can't just get rid of subagents if you have been using them extensively. You need to run evals (cost, efficiency, etc.) and let that help you determine whether to fold them (and how) into the main agent loop.
但是，如果你之前广泛使用了子代理，就不能简单地抛弃它们。你需要运行评估（成本、效率等），并让评估结果帮助你决定是否将其（以及如何）折叠到主代理循环中。
The evals should always inform your strategy.
评估应始终为你的策略提供依据。
But keep in mind issues that can arise when scaling agent work, such as too many tool calls, context bloat, and unclear separation of concerns. Subagents are still very useful for that.
但要记住扩展代理工作时可能出现的问题，例如过多的工具调用、上下文膨胀以及关注点分离不明确。子代理在这些方面仍然非常有用。
Lastly, all of this is constantly evolving as frontier models improve. Always keep an open mind and regularly test/eval your workflows and harnesses.
最后，随着前沿模型的不断进步，这一切都在不断演变。始终保持开放的心态，并定期测试/评估你的工作流和框架。
更进一步：量化金融体系
看懂新闻只是起点——沿量化金融路径，把它变成能交付的工程能力