My Thoughts on Vibe Coding

Vibe Coding, it's all the rage lately. In the ever-evolving landscape of software development, it seems like there's a new buzzword popping up every other week. But Vibe Coding I kept hearing about it, people building entire 3D video games, letting the AI models write all the code for them! Is this the future? Is it just a gimmick? Honestly, I had to dive in and see what all the fuss was about.

Before we get started enjoy a short little demonstration.

What is Vibe Coding?

The term "Vibe Coding" was coined by Andrej Karpathy, describing a new approach to software development where developers collaborate with AI to create applications. This method emphasizes using AI as a coding partner, letting it handle substantial portions of code generation while the developer guides the overall direction and ensures quality.

Vibe coding is more than just a trendy term. Many people are having great success with many different projects. Here's are a few examples:

Blobberman!

It's a twist on the classic Bomberman but where your bombs paint the map like Splatoon to win.

Penetration testing tool

You put in your website URL, and it spins up a multi-agent AI swarm to test for security issues.

These projects demonstrate how vibe coding isn't just limited to games or graphics - developers are successfully building everything from security tools to innovative game mechanics, showing the versatility and potential of AI-assisted development.

My personal experience with Vibe Coding

In my development workflow, I use a combination of AI tools to enhance my coding workflow. Here are some of the tools I use:

• Cursor: A dedicated AI-first code editor
• VS Code with GitHub Copilot: Microsoft's popular editor enhanced with AI
• Trae: An AI programming assistant
• V0:
• Firebase Studio:

You can think of these IDE and tools as the front end of how you interact with the models like the following:

• GPT-4
• Claude 3.5 and 3.7
• DeepSeek
• Gemini

One of the great advantages of these modern AI coding tools is their integration capabilities. When chatgpt first came I did a lot of copying and pasting to get the code I wanted. Now I can just guide the model to write the code for me. It's fascinating to watch AI agents take control of the IDE, autonomously coding out substantial portions of the application while I supervise. This level of integration makes the development process feel more fluid and natural, almost like pair programming with an AI that understands my intentions and can execute them at scale.

The Challenges of Vibe Coding

While the potential of vibe coding is exciting, it's important to acknowledge its current limitations and challenges:

Inconsistency in Output

One significant challenge I've encountered is the lack of consistency. Even with identical prompts, context, and project requirements, the AI can generate notably different code versions each time. This inconsistency can make it difficult to maintain a coherent codebase, especially in team environments.

Code Security Concerns

Vibe coding raises security concerns. The AI can generate code that is not secure, which can lead to vulnerabilities and security breaches. This is a significant concern for developers who want to build secure applications.

Developer Oversight Remains Critical

As discussed in this detailed thread, vibe coding isn't a magic solution. It requires experienced developers to review, refactor, and maintain quality standards. The AI is a powerful tool, but it shouldn't replace fundamental software engineering practices.

The reality is particularly stark for non-developers. One user shared their frustrating experience trying to build an internal tool: despite using advanced AI models like Claude 3.7, they encountered persistent issues with basic functionality like logging. The AI would confidently provide solutions that simply didn't work, leading to a cycle of trial and error without resolution. This highlights a crucial point: without the foundational knowledge to validate and debug AI-generated code, users can end up with partially functioning applications and no clear path forward.

This experience echoes a broader truth: while vibe coding can accelerate development for experienced programmers, it's not yet a reliable solution for non-coders to build production-ready applications. The AI's confident-sounding but potentially incorrect solutions can be especially misleading for those who lack the technical background to verify its output.