Fuse 产品工厂用 Jev 挑代码评审人
作者描述 Fuse 的 agent 流水线：从客户反馈聚类到设计评审、代码评审、监控与自动修错，其中代码评审环节接入 Jev 来选懂相关代码且有空的人类评审者。
I’m responsible for the product factory at Fuse ⚡ Customer feedback → design → design review → code gen → code review → deployment → monitoring → error auto-fix → feedback into design and code review ↺ ➡️ Customer feedback An agent clusters customer feedback by topic and sentiment, turning recurring frustrations and requests into feature ideas. ➡️ Design review Design is a lot more objective than people think. Every choice needs a rationale: no two primary buttons competing for attention, colours that adhere to the design system. This agent comments on those. ➡️ Code review This agent reads the requirements, then asks specialist agents to check for bugs, duplicated logic and unnecessary code. It links to stable engineering rules checked into the repo. Those can be challenged by opening a PR. Low-risk changes can skip human review. Riskier changes require one or two human approvals. The agent selects reviewers who know the relevant code and are available (plugging in Jev for this one). ➡️ Monitoring After deployment, an agent watches for errors and performance regressions, tracing problems back to the changes that introduced them. ➡️ Error auto-fix The auto-fixer investigates the root cause and opens a fix PR. ➡️ Feedback into the next iteration Human reviewers go after the code review agent. Issues they catch that the agent should have found become regression cases. The reviewer is also dinged for production errors and performance regressions it should have caught.
这条是引用着另一条写的：x.com/alanchanguk/status/2102035127534534938