The Dark Side of Vibe Coding

Here's my take on "vibe coding"
(or the things no one wants to hear):

First "AI Agents" now we read about "Vibe coding". A popular term for following your intuition and letting AI handle the rest. The term was mentioned by a post on X by Andrej Karpathy, who described it as "fully giving in to the vibes, ..., and forget that the code even exists". 

However, I think lines are starting to blur here...

What is "Vibe Coding"?

At its core, vibe coding means you tell an AI system what you want to build, and it generates most of the code. When an error pops up, you paste it back into the AI and ask for a repair. This approach is different from traditional software development and it seeks for a relaxed workflow.

What began as simple post has grown into a broader movement. Many beginners and no-coders now proudly call themselves "vibe coders." Skipping best practices and build useful prototypes with just prompting. On the other hand, this trend also gives rise to false perceptions which can then lead to fatal results, as can be seen in the following screenshot, for example.But please don't misunderstand, it's not Karpathys fault here. He never was talking about skipping the basics of coding or tell anyone to follow his vibes. I think he was more focused on bringing back the fun in coding: building some cool tools or games over the weekend without worrying about code quality, testing, or security. 

The original concept was taken a bit out of context and shaped by the community. Now "Vibe Coding" is used for almost anything related to Coding with AI

Blame Marketing.

Its all about marketing. For example many AI generators can make you think you can hand everything over to the AI. They might look beginner friendly, but they can still feel like slot machines that give you a big win and a rush of excitement at the start. Then, boom you get stuck, and every step suddenly feels like someone placed large stones in your path. 

I compare this a bit to "AI Agents". Another viral topic. AI Agents refers to more advanced AI systems that can plan, reason through multiple steps, or handle entire workflows. Many people use it whenever an AI helps with a task. That broad usage can lead to confusion. Since most systems are just... "Automation Workflows".

Who cares?

But who cares about all this terms and definitions? We are all flooded with stimuli anyway.

The worst thing you can do is point your finger at newbies and ridicule them for not knowing the difference between A and B. Translating complex concepts and processes into simple, real-world examples and visualising them is what I recommend. Leading by example instead of bashing the next generation of "Vibe Coders" is the key.

Otherwise, respect will be lost across generations.

And now? Here are my tips:

Beginners:

• Compensate for lack of programming skills with better prompting 

• Learn the basics of coding (variables, conditionals, loops, functions) 

• Learn how the web works: http, client, server, api 

• Start with Database (sql queries, crud apps)

• Use Project Planning to save tokens (PRD, Roadmap, e.g.)

• Build a prototype in a AI Code Generator and move to cursor until you hit a wall. 

• Learn! You will be faster with AI but that doesn't mean you can just sit back and relax (yet). 

Seniors:

• Combine Coding with AI. Do it and catch up, believe me here.

• Post, engage, and help answer beginner questions on X or Reddit. 
  (I often talk to senior devs on X who are 50+. Hell they are smart)

• Build workflow systems or frameworks and consider making them open source so others can learn.

• Create rules and guidelines with AI that give beginners a solid framework.

• Offer mentorship and clear advice to steer newcomers toward good coding habits.