Massimo, excellent article. I agree with all sentiments.
I am evaluating a hobby application that is fairly robust with a mixture of vibe coding as well as collaboration with the AI tool (my mind first then ai).
The collaboration approach is pretty good and can be impressive. But the moment you lose site of any of the codebase, I could see issues compound quickly.
Ai has created extraneous folders and duplicate code files numerous times. New models completely changed requirements by making erroneous assumptions but it goes ahead and implements it.
The code style and patterns are all over the place even when you tell it. This is my second project evaluation for vibe coding and it is far and I mean far from feasible for "good", "important" applications. When it works well it is impressive. But all it takes is a bad iteration and it creates a mountain. This is inevitable with larger codebases and even medium sized ones.
I would ask biz decision makers a simple question. "Are you comfortable with vibe coders working on your mission critical applications?" Where risk and reward have a monetary impact?
You might even have a new wave of script kiddies getting hired and at first it is a revolution. Then the issues become apparent down the road. Who will fix it then?
I think we are stepping into uncertain territory. The leaders in the industry should be aware of these things and not make decisions based upon ideas driven by social media.
The behaviour you describe is exactly what makes Vibe Coding still considered too immature for industrial and professional use.
In my experience, it can only be used with very good results in the prototyping phases and only when you consider the idea of throwing away a good part of the prototype at the end of testing and redesigning the application with traditional methods.
In summary: perfect for a lightning prototype, today unsuitable for professional development.
Maybe in the future many of the problems we have today will be overcome.
Massimo, excellent article. I agree with all sentiments.
I am evaluating a hobby application that is fairly robust with a mixture of vibe coding as well as collaboration with the AI tool (my mind first then ai).
The collaboration approach is pretty good and can be impressive. But the moment you lose site of any of the codebase, I could see issues compound quickly.
Ai has created extraneous folders and duplicate code files numerous times. New models completely changed requirements by making erroneous assumptions but it goes ahead and implements it.
The code style and patterns are all over the place even when you tell it. This is my second project evaluation for vibe coding and it is far and I mean far from feasible for "good", "important" applications. When it works well it is impressive. But all it takes is a bad iteration and it creates a mountain. This is inevitable with larger codebases and even medium sized ones.
I would ask biz decision makers a simple question. "Are you comfortable with vibe coders working on your mission critical applications?" Where risk and reward have a monetary impact?
You might even have a new wave of script kiddies getting hired and at first it is a revolution. Then the issues become apparent down the road. Who will fix it then?
I think we are stepping into uncertain territory. The leaders in the industry should be aware of these things and not make decisions based upon ideas driven by social media.
Hi Andy, thanks for your comment.
The behaviour you describe is exactly what makes Vibe Coding still considered too immature for industrial and professional use.
In my experience, it can only be used with very good results in the prototyping phases and only when you consider the idea of throwing away a good part of the prototype at the end of testing and redesigning the application with traditional methods.
In summary: perfect for a lightning prototype, today unsuitable for professional development.
Maybe in the future many of the problems we have today will be overcome.