The "Vibe Coding" Mirage: Why It's Not the Solution You Think for Your Project

Lately, I get a lot of inquiries like: "Hey, I've seen that with AI I can now build a website without knowing how to code, what do you think?"

I always respond the same way: it depends on what you want to do.

If you're looking to build a professional website, an e-commerce that actually works, or any application that real people will use, then we need to talk. Because there's a big difference between "making something that works" and "making something that works well."

What is "vibe coding"

"Vibe coding" is programming by feelings. You write in natural language what you want, the AI generates code for you, you paste it, and if it seems to work, that's it. You don't review the code, you don't understand what it does, you just trust that the AI has done it right.

Sounds great, right?

The problem is that this only works for very simple things. For anything serious, it's a trap.

Why it seems to work (but doesn't)

AI is very good at creating things that look good. It can make you a beautiful interface, with colors that work well, buttons in the right place, forms that look professional.

But when you start using the application for real, problems begin:

• Forms work on desktop but break on mobile
• Menus don't respond well on some devices
• Pages load slowly because the code is poorly optimized
• There are functions that seem to work but are broken inside
• Security holes

It's like buying a car that looks good on the outside, but when you take it to the mechanic, they tell you the repair will cost more than what you paid for it. They recommend throwing it all away and buying a new one. The same thing happens with code generated by "vibe coding": many times it's worth starting from scratch.

What I've seen in the last year

I've been working in web development with Drupal for several years. In the last year, since "vibe coding" became popular, I've encountered problematic cases.

I've reviewed websites partially made with AI by interns or people who had no knowledge about Drupal specifically. The result was always the same: something that looked like a decent visual mockup, but didn't work as the client expected.

The main problem was that the person who had built it didn't know how to fix the errors that the AI had created. They needed someone who actually understood the code to solve it... or redo everything from scratch.

Patterns that repeat:

Security problems: Forms that don't validate data properly, databases exposed without protection.

Duplicated code: AI repeats the same code in different places. When you want to change something, you have to change it in 20 different places.

Fragile integrations: Connections with payment systems or external services break with any small change. They have bugs that are easy to exploit like auto-incremental numeric IDs in URLs without validation.

Poor performance: The website works well with 10 users but collapses with 100. It doesn't take into account the cache system or best practices. Excessively complex and redundant code.

Using code when it's not necessary: Instead of using and configuring community modules from drupal.org, they prefer to use more random AI code and make it poorly maintainable.

Where AI does help

I'm not going to tell you to avoid AI completely. I use it every day, but intelligently:

For learning: It's excellent for explaining concepts, searching for code examples, suggesting different approaches, or understanding existing project code.

For quick prototypes: If you want to test an idea quickly, it can help you.

For simple modules (Drupal): If you need to create a calculator, a block with basic logic, or a simple module, AI saves you a lot of time. Before, I could spend two hours creating something simple, now I have it in 15 minutes. But it only works when you know exactly what you need and how it should be structured.

For repetitive tasks: Generating test data, creating basic structures, automating simple tasks. For example, YAML files for Drupal migrations from the migrate module.

The key is to use it for specific and simple parts, not to build entire applications.

How I work with AI

To be transparent, let me explain when I use AI and when I don't:

In personal projects: Here I do rely heavily on AI and try new things. These are projects where I want to test ideas quickly with minimum acceptable quality. I don't need perfection, I prefer to spend as little time as possible and see if they're viable.

With some clients: There are clients who allow me to use tools like Claude. I rely on them to work faster. But careful: AI doesn't work alone. It simply helps me with repetitive and monotonous tasks like refactoring part of the code, implementing tests, replicating similar files (migrations).

What AI does NOT do for me: My real work. For products going to production that need minimum performance and security, you need an expert human to review that code.

Practical recommendations

If you decide to use AI for your project:

Maintain control: Use AI as an assistant, not as a substitute. Always review and understand the code before using it. Never leave it in autonomous agent mode and send code to production without reviewing it.

Start small: First try with simple functionalities to see how it behaves.

Invest in testing: You need automated tests that verify everything works as expected.

Document everything: Make sure you understand what each part does. If something breaks, you'll need to know how to fix it.

My final recommendation

If your project is simple and experimental, go ahead with AI. But if you're investing serious money or it's something important for your business, find someone with experience to supervise the process.

The difference lies in knowing when to use the tool and when not to. For simple modules, calculators, or basic functionalities, AI is great. For complex architectures, user management, or systems that handle sensitive data, it's better to have someone who understands the implications supervise it.

AI is a very powerful tool, but like any tool, you need to know how to use it well.

I hope this perspective helps you make better decisions about when and how to use AI in your development projects.