Vibecoders

In my current org we‘re all vibecorders/AI-driven developers, but I recently started noticing that a some devs have difficulties of maintaining their own AI-slop. The AI-slop works in most cases, but it may happen that some subtle corners cases breaks it. It usually requires deep dive and comprehensive investigation. AI can only give advice in such cases if one knows most parts of the answer/s themselves to ask the right questions. The biggest issue here is that managers value such AI approach because of the very fast execution and basically say it’s how the things should be done. How often do you see vibecoders cannot fix their own AI slop?