“Sorry, I can’t help you with this problem, I don’t know how to program.” How many times have we said or heard this phrase? The boundary between those who can and those who cannot create software has always been clear, insurmountable. Until now. The vibecoding (a term coined by the researcher Andrew Karpathy, which is not exactly a dessus having co-founded OpenAI) is upending this paradigm with the force of an earthquake.
I myself have spent the last few months building apps that analyze the contents of my refrigerator, transcribe podcasts, and organize my social media bookmarks. All without writing a single line of code. Theartificial intelligence has transformed programming from a technical discipline into an almost conversational process, where the idea counts more than the syntax and logic algorithms. Of course, not all that glitters is gold, and of course not everything is perfect. But it is surprising, and not a little.
Confessions of a (non)programmer
Forgive the bad expression, but I have to say it: I am a bean with programming. I have never written a line of Python. No, I do not know JavaScript. I cannot distinguish C++ from a particularly complicated mathematical equation. Apart from a few moments in my adolescence in which I tinkered with basic, html, Asp, php, websites and Flash animations (ah, the good times!), I have never been a software engineer nor do I aspire to leave the worlds of advertising and journalism for a career in tech.
And yet, in the last few months I’ve found myself programming like a madman. Or rather, having someone else program like a madman for me: a tireless digital assistant who doesn’t ask for coffee breaks and doesn’t complain when I change my mind for the fifth time. It’s like having a technological genie: I make a wish and it grants it, even if sometimes it interprets my requests in… creative ways.
And I'm not talking about those “do-it-yourself” apps with drag-and-drop interfaces that let you create simplified versions of whatever you have in mind. No, I'm talking about complex, personalized tools that are tailor-made to solve specific problems in my daily life. All thanks to this new trend called vibecoding.
What is Vibecoding (and Why You Should Care)
The term “vibecoding” might sound like yet another tech neologism destined to disappear in a few months. But behind this word lies a trend that is radically changing the way we create software.
Put simply, the vibecoding is about building working applications by simply describing what you want to an AI. “Computer! Program this & that!” (how much do I love Star Trek?) No technical knowledge, no programming languages, no frustration over that damn missing semicolon that drives the compiler crazy. Just an idea, and of course the patience to guide the AI through the creative process.
As he himself wrote Carpathia: “It’s not really programming. I see things, I say things, I do things, and I copy-paste things, and it mostly works.” A description that would make any programming purist cringe, but it perfectly captures the essence of this new approach: the idea counts more than the code, the vision more than the technique. It's a bit like if, instead of learning a foreign language, we suddenly had a simultaneous translator so good that we could appear perfectly fluent. No, the example doesn't fit.
My digital creations laboratory
Among my creations in the last few months, I can count tools that would make any professional developer smile (or maybe cringe). I built an app that transcribes and summarizes long podcasts, for those times when I don't have time to listen to three hours of conversation to extract the 10 minutes (or specific topics) that really interest me.
Also: I created a tool that organizes my social media bookmarks into a searchable database, because apparently saving interesting content without ever revisiting it is one of my favorite hobbies. I even developed a web app that tells me if a piece of furniture will fit in the trunk of my car (yes, I’m moving), saving me that awkward scene at the furniture store where I find myself doing rough calculations with my arms outstretched. While my wife looks at me the way any wife should look at her husband: like a moron.
And I'm still the first bean, which makes you understand that vibecoding, perhaps, makes a lot of sense to exist.
How Vibecoding Works
The tools of AI coding have existed for years. The first ones, like GitHub Copilot, were designed to help professional programmers work faster by completing their lines of code the way ChatGPT completes a sentence. You still needed to know how to code to make the most of them and intervene when the AI got stuck.
In recent years, new tools have emerged that are built to leverage more powerful AI models, allowing even novices to program like professionals. Names like Cursor, repeat, Bolt e loveable have become the new best friends of “accidental developers”.
The process is almost hypnotic: You enter a description of the problem you want to solve, and the AI gets to work. Mysterious lines of code scroll across the screen, and a few seconds later (if all goes well) a more or less working prototype emerges. You can suggest changes and revisions, and when you’re happy, you can publish your new product to the web or run it on your computer.
My refrigerator experiment
I want to tell you how a small, COMPLETELY USELESS app was born (in the sense that I could do this thing by hand, but I wanted to see the effect it has), so you can understand how surprising and at times disconcerting this process is.
I asked to Bolt to build an app that could help me prepare my daughter's lunch box based on a photo of the contents of my refrigerator.
The app first analyzed the task and broke it down into parts. Then it went to work generating a basic web interface, choosing an image recognition tool to identify the foods in my refrigerator, and developing an algorithm to recommend meals based on those ingredients.
When the AI needed me to make a decision—whether I wanted the app to list the nutritional values of the foods it was recommending, for example—it would present me with several options. Then it would go back to writing code. When it hit a snag, it would try to debug its code or go back to the previous step and try a different method.
About 10 minutes after I entered my request, the app was ready. It suggested a generic turkey breast sandwich. Not exactly a culinary revelation, I admit, but considering that the alternative would have been to open the refrigerator and stare at its contents for 15 minutes before opting for… the turkey breast sandwich, I’d say it’s also pointless, in my case. But, again, it’s surprising. And surely someone with more imagination than me can make great things out of it. So, is it all great? Well, well.
When things don't go as planned
Not all of my experiments vibecoding have been crowned with success. For weeks I have been trying to build an “inbox autopilot” that can automatically reply to my emails, mimicking my writing style. I have run into snags when trying to integrate AI workflows into apps like Google Photos e iOS Voice Memos, which are not designed to work well with third-party add-ons.
And, of course, AI sometimes makes mistakes. Once, when I tried to turn a long article I’d written into an interactive website, the AI included about half the text and left out the other half.
Il vibecoding, in other words, it still benefits from human supervision, or at least a human presence nearby. And it’s probably best for hobbyist projects, not essential tasks. It's a bit like when you let a teenager drive: it's better to be in the passenger seat, ready to intervene.
This, however, may not be true for much longer.
The Future of Programmers (and Non-Programmers)
Many AI companies are working on “software agents” that could completely replace human programmers. AI is already gaining world class scores in competitive programming tests, and several large tech companies, including Google, have outsourced a large portion of their engineering work to AI systems.
Sundar Pichai, CEO of Google, recently stated that AI-generated code makes up more than a quarter of all the new code implemented at Google. A statistic that, I'm sure, helps Google's programmers sleep soundly.
If I were a junior programmer (the type AI seems most likely to replace) I might be panicking about my job prospects. But I'm just a guy who likes to tinker and build tools that improve my life in small ways. And the vibecoding It's an area where AI is unquestionably improving.
The Revenge of the Amateur Creators
Since I shared my vibecoding experience on some of Futuro Prossimo’s channels, I’ve gotten emails from a dozen other people who have built their own AI-assisted tools. Fellow “fellows” (some even better, I must say) have told me about the nutrition apps they’ve created to help them stick to their diets, or the tools they’re using to summarize the newsletters they receive via email.
The interesting thing is that none of them would have been able to create these tools without AI, unless they had decided to pursue a new career as developers.
Like my turkey breast sandwich app, few of these tools are transformative in and of themselves. What is new and noteworthy is that with a few keystrokes, amateurs can now build products that previously would have required small development teams of “serious” people. It’s as if suddenly anyone could build a custom car in their garage, without having to study mechanical engineering. Nice, and also bad when you think about it. Bad for whom?
The Implications of a “Cowardly” World from Vibecoding
I am not naive about AI, nor blind to the effects that AI coding apps could have on society if they continued to improve. I think it is possible that an AI that automates the creation of useful software could also automate the creation of malicious code, or even lead to autonomous cyber attacks. And it worries me, obviously and very much, that software engineering is only the first white-collar profession to experience the effects of job replacement of AI tools.
It’s like having a genie in a bottle that can grant you any wish, but with the knowledge that they might interpret your requests in disturbing ways. “I want an app that helps me manage my time better” could turn into a surveillance system that sends you passive-aggressive notifications every time you spend more than 5 minutes on social media.
For now, building apps to automate annoying or time-consuming tasks in my life seems like as good a use of AI as any. So I’ll keep vibecoding, at least until my daughter can make her own turkey sandwich. Or until AI decides it has better things to do than help me decide whether to put an apple or a banana in her lunchbox.
And to the professional developers out there, I want to say: your jobs are safe. For now. Because if there’s one thing I’ve learned from my vibecoding journey, it’s that there’s still a gulf between what an amateur can build with the help of AI and what a team of expert engineers can create. My word.
But that gulf is narrowing. And it's doing so faster than many of us thought possible.
