#ai #coding excitement-frustration-abandonment
The bad news:
Persistent quality decline: Static analysis warnings rose ~30% and stayed elevated. Increased complexity: Code complexity increased over 40%, beyond what codebase growth alone would explain
The even worse news (as is with Ai slop): The author notes this creates a potential feedback loop: if popular open-source repositories (likely used as training data) are degrading in quality, future AI models trained on this code may produce even worse output.
Bottom line: The responsibility for code quality remains with human developers.
https://arxiv.org/pdf/2511.04427
Via @robbowley.net https://blog.robbowley.net/2025/12/04/ai-is-still-making-code-worse-a-new-cmu-study-confirms/
Caveats:
They can only see repos where someone committed #Cursor config files (not how much it was actually used), the control group probably uses other AI tools anyway (so this measures Cursor vs. other AI, not AI vs. nothing), and it’s all open-source projects where developers can just abandon tools when frustrated, which might not happen in big organizations.
#ai #coding #cursor #aislop