PR已自动化，审查仍未跟上

Member-only story Artificial Intelligence AI Agent Software Engineering Developer Tools Github The PR Is Automated. The Review Still Isn’t. Coding agents can produce more PRs than teams can confidently review. The missing discipline is reviewability. Shoaib Ahmed Quraishi 8 min read · 8 hours ago -- Listen Share Coding agents are no longer just sitting inside the editor waiting to complete the next line. They can take a task, work in the background, modify files, run checks, summarize the change, and return with a pull request. GitHub’s Copilot coding agent can be assigned issues and open PRs. GitHub’s agents panel is built around launching agent tasks from GitHub workflows. OpenAI’s Codex app presents a similar shift: supervising multiple long-running coding agents in parallel. If agents make implementation cheaper, they also make it easier to create more review work. The PR arrives with a clean summary and green checks, but someone still has to decide whether the change should be trusted. That is where the discussion around coding agents can become too narrow. CI is useful evidence, but it is not the same as understanding the change. The bottleneck has moved from generation to review. The discipline that matters now is not only prompting. It is reviewability: structuring delegated work so a human can confidently decide whether the result is correct. Press enter or click to view image in full size CODING AGENT → PULL REQUEST → review gap → HUMAN REVIEW