Diego Corbi : 31 July 2025 10:57
“I need a system to manage my church events: volunteers, registration, community event planning.”.
Two weeks after writing this prompt, John Blackman, a 91-year-old retired electrical engineer, had developed a complete application. Event management system, volunteer recruitment, API integration for car search, etc.
All working, for less than $350.
The most incredible detail? He had never developed software in his life. He simply conversed with Claude and Replit, describing his needs, as if he were talking to an assistant.
Welcome to the era of vibe coding.
The term “vibe coding” originates from a tweet by Andrej Karpathy, co-founder of OpenAI, who described this phenomenon as “a new kind of coding where you get completely carried away by the vibes, embrace exponential growth, and forget that the code even exists.
But what does it really mean?
Well, it’s “simple”.
Instead of writing code, line by line, you describe to an AI what you want to achieve using natural language. You explain the “vibe” – the idea, the goal, the spirit of the project – and let the artificial intelligence translate it into executable code.
It’s like having a conversation with a super-expert colleague who never sleeps.
You tell him “I want an app that does X” and he, in a timeframe that would make even the most robust software engineer fall off his chair, shows you the working result.
This revolution was inevitable.
Already with the release of ChatGPT in November 2022, it was clear that language models knew how to write code. But today, we’ve gone much further: the tools have evolved, the models have improved. The quality of the software produced and the additional features of AI-powered IDEs are truly astonishing.
The LLMs have been trained on billions of lines of code from GitHub, Stack Overflow (does it still exist?!), and open source documentation. This makes them “almost perfect” translators from human language to programming language.
If you’re interested in learning more about the evolution of Vibe Coding through the eyes of one of the Italians who is right at the center of this revolution, I’ll leave you the link to an interview by Marcello Ascani with Michele Catasta, president of Replit: https://www.youtube.com/watch?v=KsIJqywDO3w&t=1219s
The workflow is almost banal: from idea to implementation without going through syntax.
We can talk about a circular workflow:
Describe → Generate → Test → Refine → Repeat.
The ecosystem of tools available to “vibe-coders” is expanding rapidly.
Here are the main categories.
ChatGPT, Gemini, Claude are the starting point. Perfect for generating simple scripts or quick prototypes. Just ask and receive the code ready for “copy-paste”.
I must give you a warning, though: if you have no development skills, this vibe methodology Coding is not for you.
This is because interfaces provide you with “working” (and even quality) code, but then you will have to assemble it yourself to create your application.
If you are already “up in the air” when it comes to development environments, packages to install, and dependencies to consider, this approach is not for you.
But don’t worry, let’s move on.
Cursor is probably one of the most famous tools. It allows you to modify entire projects simply by describing what you want to change. You select a file and write: “refactor this script using classes instead of functions” and it does it. It also allows you to integrate many third-party tools (such as Figma, GitHub, etc.), especially useful for developing digital products, via MCP servers, in a fraction of a second.
GitHub Copilot is Microsoft’s answer to the coding assistant landscape. Starting as a simple chat plugin within Visual Studio Code, it has now become a full-fledged programming assistant, operating both in vibe-coding mode and as an expert copilot.
It is possible to define various agents and different MCP services to automate processes, Leveraging the power of the highest-performing AI models like those from OpenAI, Google, and Anthropic. Today, it integrates with the most widely used IDEs, such as Visual Studio Code, Visual Studio, the JetBrains suite, and many others.
Why are many companies choosing it? Microsoft guarantees that everything sent and generated is owned by the user and remains confined within the company tenant. I’d say not bad, especially at a time when the opt-out from retraining AI models with sent data is in the spotlight.
In this category, Augment Code also deserves a mention. It’s an AI agent that integrates into your favorite IDE and offers all the features found in tools like Cursor and GitHub Copilot.
These tools are very powerful and are the real weapons available to the software engineer who wants to integrate AI into their development process. However, even here, if you don’t know at least the basics of software development, these tools aren’t the most suitable for you (although alternatives like Amazon KIRO could fit right in between the two worlds. But that’s another story).
Let’s move on and get to the real protagonists of the Vibe Coding universe.
Finally, let’s talk about the tools that every vibe-coder dreams of working with. We can divide them into two parts. categories.
This category, just to name the most important, definitely includes Bolt.new and Lovable.dev, which allow you to build and deploy complete apps using only prompts. You can even import designs from Figma and transform them into code. But it doesn’t end there. They also have native integration with PostGres database services, such as Supabase, which allow you to integrate databases and Edge functions via text input.
I’ve mentioned the two most famous ones. But there are many others emerging, such as Base44, which I may discuss in a future article.
This category includes all those tools that can be used both in a pure vibe-coding mode (without looking at a single line of code) and in a developer mode (as you would do with an IDE like Cursor). Clearly, without the customization that you could get with a real IDE that you use locally on your computer.
Replit is certainly the most performing tool in this category. It provides you with a real development environment, which runs on a cloud container, in which you can develop manually, get help from an AI assistant (their Replit Assistant) or completely delegate the development to their Replit Agent which, on the Based on the requirements provided, it will develop a full stack app for you.
Personally, as a Product Manager with entry-level programming skills, this is the Vibe Coding tool I use the most.
Finally, we come to the development assistants dedicated mainly to professional developers.
Claude Code by Anthropic and Codex by OpenAI can read and understand the entire codebase before making changes, maintaining context memory between sessions. different.
You can ask them to identify a bug, explain how the main functions that characterize your software work or ask them to modify entire files.
There are many of them and, often, they are very abused or victims of exaggerated hype.
It must be said, however, that, compared to the past, we are facing an unprecedented transformation, both in the world of coding and product management.
So let’s take a look at examination of the main weapons that this “movement” allows us to use.
What used to take weeks is now done in hours.
A practical example, taken from the Replit vibe coding community, is that of Content Genie.
The protagonist of this story is a marketer who, despite having no technical programming skills, had an idea: automating the generation of “ideas” for content starting simply from YouTube URLs.
In other words: transform hours and hours spent watching videos, looking for inspiration for its content, into an automation that performs this process for it.
After just 30 minutes of conversation with an AI – using tools like Replit – the project came to life. The result? A process that previously required hours of manual work now happens in a few moments thanks to vibe coding.
You no longer need to know how to program to create software (so some say).
Designers, marketers, and entrepreneurs can transform their ideas into working prototypes (no more than that, for now) by interacting with AI.
The case of John Blackman is emblematic: at 91 years old, with no programming experience, he built a complex event management system for his church using Claude and Replit. His system manages:
All in two weeks, for less than $350.
But now you’re probably wondering: can I generate a product for an enterprise company using (pure) Vibe Coding?
The answer is: absolutely not.
But let’s remember that we are only at the beginning and we are quite far from the maturity of this technology.
By freeing yourself from manual coding, you can focus on:
The AI becomes your personal tutor.
It corrects you, suggests improvements, guides you through concepts complex problems without having to spend hours and hours searching for information online.
Speaking of searching for solutions online, have you happened to see the traffic curve of the StackOverflow site lately?!
Unfortunately, it seems that with the explosion of generative AI and the introduction of tools like these, things aren’t going so well…
But it’s not all roses and flowers. Like any revolution, this one also carries significant risks that, if ignored, can lead you into serious trouble.
Let’s explore them together.
AI-generated code may look perfect on the surface but hide inefficiencies, vulnerabilities, or questionable architectural choices. It’s like a building that may seem magnificent on the surface, but is actually built on shaky foundations (Developers, rest assured for now).
The speed of code generation can lead to the belief that the code doesn’t need review. “AI wrote it, it must be fixed”. But this bias is very dangerous: an overlooked vulnerability could expose you to a data breach of millions of your users’ data.
Vibe coding can accumulate technical debt without you realizing it. Every shortcut, every piece of code that isn’t aligned with the overall architecture adds up, and you’ll pay the price as soon as you release your application into production.
AI doesn’t excel at advanced debugging. If most of the code is generated by AI, who will fix the complex bugs that usually appear after the first few weeks of use in production? How do you understand a system if you have no idea how it was built and what foundations it stands on?
We can honestly say that, in a professional context, Vibe Coding does not replace programmers.
It transforms them.
From Coders to Visionaries
Developers will become “orchestra conductors” who:
Essential new skills
Hybrid Human-AI teams
In the future we will see teams where each developer has a dedicated “AI instance” which:
We are increasingly moving towards complex systems where an AI Orchestrator coordinates other specialized AIs. A master agent distributes tasks to specific AIs for frontend, backend, database, testing.
Visual builders will integrate (and are already doing so) with LLMs. You can say “when the user clicks this button I want X to happen” and the AI will generate the necessary logic, overcoming the current limitations of no-code.
This technological progress, and the resulting disintermediation from developers, could give rise to new professional figures such as:
The key is to find the balance between speed and quality. But how do we do that?
With some best practices and a very interesting emerging approach, which developers like Omar Diop and Gianluca Carucci are carrying forward: the Vibe Engineering.
Emerging Best Practices
The first time I heard this word was in One of the posts from the vibeEngineers newsletter on SubStack.
****Its fundamental axiom is not to abandon oneself completely to AI, but to combine:
It’s essentially like when we talk about supervised autonomous driving: the car takes you “by itself” where you decided to go. You maintain control of the destination.
Vibe engineering represents an evolution of vibe coding that combines AI-powered creative flow with the principles of software engineering: architecture, product development, domain-driven design, team topology.
Everything you need to build products that not only work, but grow, maintain, and scale over time.
Vibe Coding represents a fundamental paradigm shift. We’re not just automating code writing: we’re redefining what it means to be a developer and a product manager.
The code of the future will increasingly be developed through a conversation between different intelligences – human and artificial. The quality of the result will depend on the quality of the questions, the honesty of the answers, and the wisdom of the listener.
Will it lead to less skilled developers or simply more efficient ones? It’s still too early to tell.
But we’re only at the beginning.
Soon, models will become more powerful, tools more integrated, and human-machine collaboration methods more fluid.
So, the real question is not whether Vibe Coding will change the world of software development.
It’s already doing so.
The question you need to ask yourself is: what “vibe” do you want to convey to your AI to build the future?
Diego Corbi