EVENT DOSSIER
As AI code generation tools become more widespread, the surge in low-quality code has become a new bottleneck in engineering review. Surveys show that approximately 42% of code contributed by developers is generated by AI, and most developers do not trust its accuracy. Synthesia’s number of pull requests increased by 120% year-over-year, with 95% of them containing AI-generated content, mainly involving hidden errors and repetitive issues. To address this challenge, engineers are adjusting their review strategies: establishing detailed guidelines before coding, deploying dedicated AI agents to detect common defects, and submitting high-risk changes for manual review or requiring developers to provide explanations. Startups like CodeRabbit are taking this opportunity to raise funds, trying to fill the gap in trust. Synthesia’s Chief Technology Officer noted that extensive work is needed to train AI to avoid repetitive generation.
AI 代码生成工具虽能大幅加速开发，但产生的大量低质量代码仍需人工审查。Sonar 调查显示，开发者贡献的代码中 42% 来自 AI，且 96% 的人不信任其正确性；Synthesia 公司 pull request 数量同比增长 120%，其中 95% 含 AI 生成内容，主要问题包括隐蔽错误和代码重复。为应对挑战，工程师正调整审查策略：在 AI 编码前制定详细规范、部署专用 AI 代理检测常规缺陷、将高风险变更提交人工审核或要求开发者辩护其产出。CodeRabbit 等初创公司借此机会融资，试图填补信任缺口。Synthesia 首席技术官 Peter Hill 指出，AI 生成代码导致审查成为新瓶颈，且需大量工作训练 AI 以避免重复。