围绕 Jev 搭的编码 agent 循环:选文件、定动作、验结果
作者演示一个以 Jev 为核心的编码 agent 工作流,每步结果后决定继续、重试还是求助,减少无效动作。
Your AI agent is still working. But is it getting anywhere? Opening more files. Calling more tools. Rewriting the same code. Every action costs time. Every bad decision creates more work. This is the coding loop I’d build around Jev: → Pick the relevant files. → Choose the next action. → Make a focused change. → Check permissions. → Test the result. → Bring the evidence back to you. After each result, decide: continue, retry, or ask for help. The video illustrates the concept with simulated activity. The goal is simple: fewer wasted steps between “build this” and something you can actually use.
这条是引用着另一条写的:x.com/dain0x/status/2102363438370480337