When I reject AI code even if it works
Summary
The piece argues that as AI-assisted coding accelerates, the real bottleneck becomes reviewing AI-generated code. The author describes cognitive overload during review, and explains why they often reject AI-generated changes unless they can articulate the approach, ensure the diff aligns with the problem, and maintain understandable, scalable solutions. It champions human-in-the-loop review alongside AI tools and emphasizes that AI can help but cannot yet autonomously produce sustainable solutions.