40% of pull requests in September were drafted by an AI agent. In March it was… · @ReignSwords · TwKit

@ReignSwords

40% of pull requests in September were drafted by an AI agent. In March it was 17%. You don't write code anymore. You review it. And agents write code that passes tests while being quietly wrong. My 5-point checklist for reviewing an agent's PR — steal it: 1) Read the diff, not the summary. Agents write confident summaries of diffs they butchered. 2) Check file scope. Does the diff touch files the task never mentioned? That's where the bugs hide. 3) Hunt the fake-green tests. A test that can't fail is decoration — delete one line and watch. 4) Grep for secrets before approving. Agents love hardcoding keys "temporarily." 5) Ask for the why in one sentence. If it can't explain the change simply, it doesn't understand it either. The bottleneck moved. It used to be typing speed. Now it's judgment speed. Fellow AI builders — follow for more. My rule: never trust a PR whose tests can't fail. What's your #1 agent-PR red flag? 👇 🔵 Fellow Premium members, let's connect! I post about AI & tech — follow for more 👥

曝光 47 · 评论 10 · 点赞 1 · 书签 0