How to Manage Coding Agents
InfoQ reports that Anthropic Claude Code engineer Addy Osmani published an essay arguing that engineers who frequently use coding agents are already taking on management work. Drawing on Claude Code, he recommends delegating tasks while retaining judgment, using automated tests and independent reviews to validate results, limiting manual intervention, and watching for orchestration costs when running multiple agents in parallel.
The essay offers a practical workflow for developers using coding agents, emphasizing that effective use depends on validation mechanisms, human judgment and coordination rather than prompt engineering alone.

编译|林绮蓓
编辑|蔡芳芳
以前,工程师的一天可能从打开 IDE、阅读代码开始。如今,一些深度使用 Coding Agent 的工程师,每天要做的第一件事,可能是查看几个 AI Agent 的工作进度:哪些任务已经完成、哪些测试没通过、哪些 PR 等待审查,以及哪些工作需要重新分配。
代码是 AI 写的,但工程师的工作似乎并没有因此变得简单。
