This is about
Code
4 minutes
19 Mar 2025
This is about
Code
4 minutes
19 Mar 2025
Does vibe coding work?
AI is often hyped as the future of software development, with a recent wave of "vibe coding," where developers rely entirely on prompting AI without touching the code themselves. But how well does this approach actually work in practice?
I recently conducted an experiment using only AI to build the Explore page for ShaderGPT. My goal was to see how far I could get relying entirely on AI-generated code, with minimal manual intervention.
The Setup
I used the popular editor Cursor which has a Composer mode, that allows AI to make changes across multiple files automatically, along with its Agent mode, which can read JavaScript errors from the console and attempt to fix them. Cursor was set to use the latest model Claude Sonnet 3.7 from Anthropic which is regarded as one of the best LLMs for coding.
The workflow was simple:
- Write a detailed prompt describing the feature I wanted.
- Let the AI generate code and apply changes.
- Review the AI’s work, fix bugs, and refine the implementation.
- Iterate until the feature worked as intended.
The results? Faster development but also a fair amount of AI-induced chaos.
Speed vs. Control
One thing became immediately clear: AI significantly sped up development. Tasks like implementing pagination, filtering, and search functionality happened much faster than if I had coded them from scratch. However, this speed came with trade-offs:
- Overly complex code: AI often introduced unnecessary complexity, requiring manual cleanup.
- Lack of design intuition: Visual tweaks were frustratingly imprecise. Asking AI to "move something a bit to the right" often led to drastic layout changes.
- Bugs that AI couldn't fix: I encountered a few issues where AI struggled to provide a working solution, requiring me to step in, debug, and then have the AI implement my fix.
Example: Fixing Visual Glitches
One of the biggest struggles was removing a blinking effect when loading the Explore page. Initially, I asked AI to "make the page fade in nicely." It responded with a solution that broke other layout elements, causing cascading issues. Fixing this required:
- Debugging the problem myself.
- Providing a detailed fix for AI to implement.
- Undoing some unnecessary AI changes.
- Iterating multiple times until it looked right.
At one point, I even tried uploading a screenshot of the intended layout, which actually caused AI to remove all its extra JavaScript components and revert to a CSS-only fade-in (something I actually preferred). But it highlighted how AI lacks design intuition and can’t "see" how things should look.
The Trade-Off: AI in Creative Development
AI is great for speeding up boilerplate and repetitive coding tasks, but for highly creative front-end development, it falls short. The lack of precise control makes it difficult to achieve a refined, bespoke design without significant manual intervention.
When AI Works Well
✅ Rapid prototyping and MVP development
✅ Implementing standard UI components and functionality
✅ Generating boilerplate code and database queries
When AI Struggles
❌ Fine-tuned visual adjustments and motion design
❌ Debugging complex interactions
❌ Writing clean, maintainable code without oversight
For now, true front-end magic still comes from experienced designers and developers who can craft a vision beyond what AI can generate.
The Future of AI-Assisted Development
While AI is undoubtedly a powerful tool, it still requires an experienced developer to guide and refine the output. For projects like those we do at 14islands—where visuals, motion, and user experience matter—a skilled developer or motion designer is still essential. AI can accelerate the process, but it can’t replace craftsmanship.
Relying on AI means giving up some control, which can introduce security risks, unnecessary dependencies, and inefficient logic. The code may not be as clean or structured as a developer would like, since AI prioritizes functionality over maintainability. Careful review and refactoring remain essential.
In the end, I found the experiment valuable but also a reminder that AI is just a tool. It’s not a replacement for human creativity. For now, true front-end magic still comes from experienced designers and developers who can craft a vision beyond what AI can generate.
Droplets
Droplets
Sign-up to get the latest insight from 14islands. We send a newsletter only once every quarter with inspirational links and creative news. It's short and sweet. You can unsubscribe it at any time.
