Vibe Coding - Real Life Stories - Part 1
- Massimiliano Turazzini
- Mar 16
- 9 min read
Imagine sitting down with a coffee in your hand, writing a couple of sentences, and watching an entire application take shape before your eyes. Design, database, authentication, documentation, secure code – ready in less time than it takes for a pointless call on Teams.
Sci-fi? No, it's Vibe Coding.
The term "vibe coding" was coined by Andrej Karpathy, an AI engineer who worked at Tesla and OpenAI. In February 2025, Karpathy tweeted about his new way of coding, where you "give yourself over to the vibes" and power of AI, letting the AI handle the details of your code. Vibe coding represents a paradigm shift: instead of writing code line by line, you interact with the AI iteratively and conversationally, describing your goal and refining the code generated through ongoing dialogue.
I have been experimenting with this new way of working for months now and I can finally tell you what it means to live this experience from the inside as someone who has been in the world of development since she was 13 years old.
Vibe Coding: Write a Prompt, Get a Software
The scenario is this: you open an AI tool, describe what you want in natural language and it spits out code. Nice, clean, organized. If you need to modify it, just add details in subsequent prompts. If there are errors, they are corrected with another input.
And, in some iteration, you have your working software.
It seems like magic. But is it? The short answer is: it depends. 🤷
For example, the app in this video was made in 25 minutes. Below I'll tell you how.
If you know what you're doing, it's a superpower!
If you are inexperienced, it's a double-edged sword. I leave you with a recent post of mine on the roles of AI. As always, the difference is made by those in command.
AI is a tool that amplifies the skills of those who use it. Those who have experience and a clear vision of the project can achieve incredible results. Those who start from scratch risk creating code monsters that collapse at the first wrong line of input.
So what happens? We learn! By making mistakes, correcting, and trying again. And here comes the best part: coding is becoming accessible to everyone.
Brief personal history
Many years ago, the 90s were coming to an end, when I was still a 'bubbez', given my phase of divulgation of application development in Microsoft, I was asked to do a two-month workshop to a C-Level team of a large Italian financial company. Objective? To teach them the basics of databases with Access, the newborn from Microsoft, and programming, always in Access, with VBA. The 'vibes' at that time were that thanks to visual tools 'anyone' could start programming. However, explaining the concept of forms, events, macros, code, at that time was not easy. Explaining it to managers with an average age between 40 and 55 was even less so.
But in two months we managed to develop together an app to aggregate and reclassify the balance sheets of 5 companies in the group.
I remember perfectly that they had chosen on their own, consciously, to slip into a black hole like that of programming. While at the beginning, they were all disoriented, after a few weeks, they all started to show me their procedures, their queries, and the reports they were making, and a project emerged that lasted for a few years. Only a couple of them went on to develop further tools in Access. But I will never forget the enthusiasm, the desire to get involved, the difficulty of understanding how a database, even a simple one like that of Access, is much more complex than an Excel file, how much their eyes shone when they became autonomous in extracting data that was very precious to them that previously had to take weeks for the 'EDP' to prepare something tangible for them. A lot of time has passed since then, IT has become much more complicated, and I have rarely found managers, at any level, capable and willing to develop their own tools.
In the meantime, I had also stopped programming and, I confess, I was happy about it. Because even though development had always been my passion since I was 12, everything had become very complicated and, probably, my patience and my mind were in decline.
But today those vibes are back, amplified
When GPT-3 came out, in addition to asking it for 'the usual things' I immediately wanted to get it to produce code. But it was all too complicated, again. Even with ChatGPT, with the various subsequent versions that greatly improved the quality of the code, we were in 'Copy&Paste Development': you ask ChatGPT (or one of its competitors) to produce some code, you copy it into an editor, you test it, you check for errors. You copy the errors and paste them back into the chat, it responds with new code, which we try, without reading it, to paste on the other side, all in an infinite, low-productivity cycle.
Then came the copilots that provide, within a development environment, a chat that reads the code that is written in real time, can modify it, can complete it and suggest changes. More or less interesting tools like Copilot by GitHub, Cursor, up to extremely powerful plugins like Cline or Roo Code that are, at least for me, disarming for how much they manage to give me again the ability to generate code and even very complex applications without me necessarily being an expert in that language or that technology.
What did I do?
Dashboard for soft mobility analysis with Green Ride
App to generate structured content in multiple languages and following very strict brand voice lines
Chat with LLM to understand how they work behind the scenes.
Personal assistants who have access to my camera, microphone, and screen, whom I can verbally question about what I'm doing to get answers, suggestions, and advice.
Real-time video analysis that recognizes faces, people, and animals, connected to the external camera of my house. The next goal is that they recognize me and let me enter the house without me having to use the keys, and that they tell me if the burglar alarm has gone off at night, when it rains, because the cat has passed in front of the sensor to take shelter.
Analysis of all my historical internet bookmarks, classification and creation of an archive that I can query via LLM
Everything you find on AI Cookbook on 01Net was done basically this way: creating Podcasts, transcribing meetings, working with long texts and much much more. I'll stop...
For an 'ex-geek' like me it was a fantastic phase: I had a lot of 'AI Interns' who helped me get back to programming. That is: I did the analysis and gave the specifications, they programmed, I checked, complained that they didn't work, showed my colleagues fantastic things etc.
In the meantime, the Vibes of those 90s have really exploded for anyone who has approached these instruments.
What was missing?
Everything was going smoothly when it came to prototyping and creating personal apps. There were various solutions:
Many ran directly from the command line
Some offered a minimal desktop GUI
Others lived in Colab notebooks, perfect for quick tests.
Some even went as far as a real web interface. But they were, and are, complex worlds.
The main problem with Copy&Paste coding, even in the presence of Copilots, is that it is difficult, if you are not a technician, to release them in production environments (deploy). They all need to be released in suitable places, to have an architecture behind them that supports them, that provides security, etc.
Themes that I could have gone back to doing with Copy&Paste coding but that would have taken me a lot of time and, since they were all personal applications that ran on my Mac, or that I had to distribute to a few people, I never invested too much time in them. But I learned a lot of things on the technical side that, at least, allow me to talk to technicians with greater awareness of the potential and limits of the tools they use. (I actually wasn't rusty at all on these concepts, but I can't do everything 😀)
In the meantime...
We have seen a progressive release of new tools, increasingly simple to use and increasingly complex in terms of the results generated. Tools that also take care of putting an app into production, managing the security system behind it (authentication, user management, etc.), creating a support database, generating and hosting web functions (to send emails, do complex calculations, etc.)
I started playing with one of these, Lovable.dev, and... WOW! in 20 minutes I remade ChatGPT (version 2022 though) complete with authentication, conversation storage, choice of different templates to use.
For the article on agents, I developed the paperclip maximizer in just seven steps.
For a company project, instead of making a mockup, we developed a test application, working, in 1 hour (time required for mockups at least 1 week). This allowed us better to understand the customer's needs, the development complexities, and calibrate the proposal accordingly (then we will make the software 'as it once was')
During an AI workshop, we developed a mobile application for time tracking of farmers in the vineyard. In 15 minutes.
Another example can be found here, on 01net, where in 25 minutes I set up a structured and functional corporate feedback collection system, as the video shows.
And Lovable is not the classic American super-funded-startup. But only the European startup grew fastest in history.
The main tools (in my opinion)
Here I report the main solutions, those I have put my hands on even if for a short time. They all have a lot of side effects and interesting features.
Tools | Description | Importance in Vibe Coding |
Online coding platform with a built-in AI assistant that lets you build and deploy applications from your browser. | It offers a complete and accessible development environment, ideal for those who are new to vibe coding. | |
Code editor based on VS Code with AI chat capabilities to generate and edit code using natural language. | It allows precise control over the generated code and integrates with a popular editor like VS Code. | |
Web platform that generates full-stack applications from natural language descriptions, with integration with Supabase for database and authentication. | Simplify the creation of prototypes and MVPs, focusing on the UI and UX. | |
Codeium's AI-native IDE that combines an AI coding assistant with the ability to automatically perform tasks like editing multiple files or running commands. | It offers a “mind meld” experience with AI, managing project context and refactoring. | |
Platform that allows you to run, modify and deploy full-stack web applications directly from the browser, with the ability to install packages and run backends. | Suitable for rapid prototyping and MVP development, with an intuitive interface and the ability to edit the generated code. | |
AI coding assistant developed by GitHub and OpenAI that works as an extension in code editors and offers real-time code suggestions and a chat mode for programming questions. | Helpful for learning to program and getting help writing code, with seamless integration into popular editors. | |
VS Code plugin that acts as an AI assistant for software development, capable of creating and editing files, running commands in the terminal, and using the browser. | It offers a collaborative approach to development, allowing you to customize AI behavior and integrate external tools via MCP. | |
Cline fork with advanced features, such as multi-model support, AI-powered automation, and experimental capabilities. | It allows you to create “agent teams” with customized prompts for different roles (e.g. QA engineer, product manager) and provides granular control over AI behavior. |
What are Plugins? 'Third party' applications that install in an existing environment and enhance it through the use of AI Agents specialized in programming. The two most powerful are Cline and Roo Code. Both open source, the second born from the first as a 'fork' (they may have argued, who knows). The fact is that they are the ones I like the most.
Check out a Deep Vibe Coding moment from a few days ago on my Mac where I'm using three different tools at once (just to wow you guys 😀).
The beautiful part
Simplify access to coding, make English (or the language you like the most) the best programming language allows anyone
you can explain it
have a critical sense
be logical
have something to do to quickly produce software, even if only at the prototype level
The fact that software development is easier does not mean that programmers will die, but that more people can approach this world.
Those who are better programmers than me (99.999999999%) now have wonderful tools at their disposal to enhance their work and be even better than before.
But... there are a few side effects.
I'll talk about it in the next few days. In the meantime, start playing it so we can see if you have the same impressions as me.
See you soon!!!
Massimiliano
Comments