Does Vibe Coding Work?

Does Vibe Coding Work?

A lot of software developers hates vibe coding. Partially because the discipline creates bad code that's difficult to read and maintain. My LinkedIn profile is scattered with examples of "vibe coding gone bad", and senior software developers apparently can't say enough bad things about it.

On the other side those who already loves it cannot praise it enough. Vibe coding allows "citizen" to create fairly complex functionality, that kind of seems to work, and I can only assume it gives a very strong feeling of empowerment to those using it.

So with these two opposing views, I asked myself the following question ...

Can the dumbest vibe coder on earth create something useful with "vibe coding"?

How the system works

Magic Cloud is 100% open source, and you can rapidly get started with it using its Docker files as long as you've got a MacBook with M2 or M3, or a Linux system of some sort. Once you install it and start using it, you'll see something resembling the following.

Vibe Coding Dashboard

In the above screen, you can either click the "Quick Start" button, or give it any command you wish. The system can;

  • Module & File Management

    • List, create, and delete modules
    • List, create, modify, and delete files and folders within modules
    • Retrieve file information and load file contents
    • Hyperlambda Automation
  • Generate or modify Hyperlambda code based on your requirements

    • Execute Hyperlambda code or files directly
    • Database Operations
  • List, create, and delete SQLite databases

    • Execute SQL queries (with or without returning results)
    • Retrieve database schema (DDL)
    • API & Documentation
  • Generate OpenAPI (Swagger) specifications for modules

    • Plugin Management
  • List and install backend plugins

    • User & Role Management
  • Create, delete, and list users and roles

    • Add or remove users from roles
    • List roles for a user
    • Task Scheduling & Automation
  • Create, list, retrieve, delete, and schedule tasks

    • Generate random strings (e.g., for passwords)
    • Email & Communication
  • Send emails (supports HTML/Markdown)

    • AI & Machine Learning
  • Create, list, and delete machine learning types

    • Crawl websites for RAG (Retrieval Augmented Generation) training data
    • Vectorize types for semantic search
    • Create, search, update, and delete RAG training snippets
    • Create AI functions (tools) for types
  • Web & Data Integration

    • Scrape URLs for content
    • Search the web and retrieve URLs

But we're constantly expanding upon it. If you've got custom requirements, you can contact us here.

Citizen Software Development

The above has been our industry's "hole grail" for decades. I wrote an article about it in 2010 called; "Teaching grandma to create software". The article was published in one of the largest IT magazine in Norway, and ended up having hundreds of comments. Some of those commenting thought I was crazy, and claimed that "software development is too complex for 'normal people'".

Ignoring the fact that these are the same arguments as we saw 500 years ago when we translated the Bible into English, it's been an elusive goal, and nobody has really managed to actually do it. Even in today's gen-AI world, with LLMs producing code at hundreds of lines per minute, there always seems to be "that one bug the AI simply can't figure out".

How Magic differs

The Magic difference is in its dense line count. Typically, you can take any C#, Java, or Python solution, port it to Hyperlambda, and end up with 5% of the original codebase. This implies 5% of the tokens, 5% of the complexity, and 20 times more readable code.

This implies that the LLM's problem is much simpler than with traditional code. Basically, you'll need a 20x smarter LLM to create something in Python as you'd need for Hyperlambda. This implies that it's literally 20 times easier for me to create a "perfect vibe coding platform" than for anybody else not using Hyperlambda.

This is true to the extent that our Hyperlambda generator is fine tuned on top of GPT-4o-mini in fact! We don't even need a strong AI model to generate "perfect Hyperlambda".

The results

Vibe coding is not perfect, not even in Magic. It messes up every now and then. But if you're waiting for perfect code, I suspect you'll never be satisified. However, asking for perfect code is the wrong question. What you should be asking yourselves is the following ...

Is this useful?

And the answer to that question is; Yes, indeed!

So is vibe coding the end of software development? I suspect not. There will always be room for really smart people, helping others, with whatever it is they want help for - Including sanity checking vibe code for that matter. So even though vibe coding is not perfect, it's still extremely useful, but it's not a software developer replacement.

The task from the above video for instance, is something I'd spend a couple of hours on doing myself. The system was perfectly capable of solving it 100% automatically, without me knowing anything about software development from before. And I sanity checked the code afterwards and it was 100% perfect!

If you need a vibe code platform, you can install the open source version, or contact us below if you need a hosted and managed solution.

Thomas Hansen

Thomas Hansen

I am the CEO and Founder of AINIRO.IO, Ltd. I am a software developer with more than 25 years of experience. I write about Machine Learning, AI, and how to help organizations adopt said technologies. You can follow me on LinkedIn if you want to read more of what I write.

This article was published 11. Aug 2025

Open Source Vibe Coding Magic

As far as I know, Magic Cloud is the only open source 'vibe coding platform' in existence. In this article, I will show you what it can do.

GPT 5 Launches Today

10AM Pacific Time OpenAI will launch GPT version 5, during a live broadcast. Hopefully we will all be given access to it indiscriminately.

Open Source OpenAI Models Released

OpenAI has taken a lot of hits because of choosing to release their models as 'closed source'. Today that changed. Meet gpt-oss-120b and gpt-oss-20b.

Copyright © 2023 - 2025 AINIRO.IO Ltd