For Vibe Coding, I like to call that type of programming "Solution Searching" as opposed to "real" programming which would be "Problem Solving" under this context.
For the purpose of this article, I'm defining Vibe Coding as prompting the Generative AI to create a code for you, and prompting it to correct the code when you encounter errors without actually double-checking what the AI does.
So, I'm not a true programmer, I only dabbled in gamedev for a while and I know programming basics, I haven't used my programming skills for an actual job before, but that's exactly why I know this fact:
Programming isn't writing code. Programming is Problem Solving!
"Real" programmers don't only code. They learn the system they'll work on, have an objective to accomplish with that system. They accumulate as much knowledge and tools as possible and use them to solve problems they encounter.
With enough experience, a programmer gain an intuitive sense of what works for what purpose. This allows them to create a new solutions to unexpected problems. Even generative AI can be one of these tools, but they learn how to guide it to enhance their problem solving.
So, a real programmer, even if they asked AI to help them solve the problem. They'll be able to follow the logic and try to learn from it.
AI generated code can reach the same result, but its approach is entirely different.
Instead of knowledge and tools. The AI accumulates as many Problem-Solution pairs as possible, and when asked, it tries to bring out the best "Solution" to any problem they encounter. What makes generative AI good is how it's able to find patterns, even better than humans sometimes... So, it's able to adjust the "solution" to fit the problem, but this isn't fool-proof... If they encounter a new problem that's completely outside of their Problem-Solution database, they'll just spout nonsense!
Not only that, for current coding generation models, if the problem isn't sufficiently explained in the prompts, it'll also spout nonsense!
All of this, and the biggest limitation for now is the Context Window. Coding generation models can only "see" a limited number of lines at a given time. They also become less accurate as more code is given as context. It's currently impossible to generate big projects with them!
Right now, new models are developed every week to tackle all of these challenges.
We're still in early development regarding those, and as they say "AI now is at the worst it'll ever be."
So when a programming-user depends entirely on Vibe Coding, they'll probably create functioning and nice-looking applications. But these will be rehashes of popular programs that has already been made. That's when the generated code manages to actually produce a functioning application, or if the errors were easy to fix.
With good prompting, and understanding of how application work, they'll be able to Vibe Code more complex applications with less time spent on trial and error. But that requires a knowledge most people who resort to vibe coding, including myself, lack, and it's (at least right now) impossible when leaving the AI to code on auto-pilot.
Another challenge is the security issue. Unless you know what you're doing, Vibe Coding will result in fragile applications that can easily get hacked or broken. Especially if the "base solution" the AI code generator came up with is made up of various codes that function well together, but are equipped with incompatible security measures...
Due to these problems, it's never recommended to create production-ready applications with Vibe Coding. But I'm excited to know how these challenges will be tackled in 2026.
What do you think?
@ahmadmanga/re-leothreads-2iby3q44v
@ahmadmanga/re-leothreads-phnveujc
