AI and Vibecoding helped me create my own software.

I am not a encoder. I can't write a single Python, Javascript or C ++ line. Except for a brief period in my adolescence when I built websites and played with Flash animations, I have never been a software engineer, nor I host the ambitions to renounce journalism for a career in the technology industry.

And yet, in recent months, I have been encoding a storm.

Among my creations: a tool that transcribes and summarizes the long podcasts, a tool to organize my social media markers in a search database, a website that tells me if a furniture furniture will fit into the trunk of my car and an application called Lunchbox Buddy, which analyzes the content of my refrigerator and helps me decide what to do the lunch of my son's school.

These creations are possible thanks to artificial intelligence, and a new trend of the “vibecoding”.

VibeCoding, a term popularized by the IA researcher <a target="_blank" class="css-yywogo" href="https://karpathy.ai/” title=”” rel=”noopener noreferrer” target=”_blank”>Andrej KarpathyIt is a useful tachigraphy for the way in which today's tools allow non -technical fans to build applications and websites that work completely, just writing indications in a text picture. You don't have to know how to encode Vibecode, just having an idea and a little patience is usually enough.

“It's not really encoding,” Mr. Karpathy <a target="_blank" class="css-yywogo" href="https://x.com/karpathy/status/1886192184808149383″ title=”” rel=”noopener noreferrer” target=”_blank”>wrote This month. “I see only things, I say things, execute things and copy things to hit, and it works mainly.”

My own Vibecoding experiments have been aimed at doing what I call “software for one”: small applications already measure specific problems in my life. These are not the types of tools that a large technology company would build. There is no real market for them, their characteristics are limited and some of them only work.

But building software in this way, describing a problem in a prayer or two, then seeing a powerful model of going to work by building a personalized tool to solve it, is an amazing experience. It produces a sensation of vertigo ai, similar to what I felt after using chatgpt for the first time. And it is the best way I have found to demonstrate to the skeptics the skills of today's ai models, which can now automate large basic computer programming fragments, and soon it can be capable of similar feats in other fields.

Ia coding tools have existed for years. The previous ones, like Github Copilot, were designed to help professional encoders to work faster, partly at the end of their code lines in the same way as Chatgpt completes a sentence. I still needed to know how to code to make the most of them and intervene when the ai ​​stuck.

But during the past year or two, new tools have been built to take advantage of the most powerful models that even allow neophytes to program as professionals.

These tools, which include cursor, Replis, Bolt and Levable, work similarly. Given a user's message, the tool presents a design, decides the best software packages and programming languages ​​to use, and gets to work by building a product. Most products allow limited free use, with payments that unlock better characteristics and the ability to build more things.

For a non -programmer, VibeCoding can feel like witchcraft. After writing your prompt and mysterious lines of code, and a few seconds later, if everything goes well, a work prototype arises. Users can suggest adjustments and reviews, and when they are happy with it, they can implement their new product on the web or run it on their computers. The process can take only a few minutes, or up to several hours, depending on the complexity of the project.

This is how I looked when I asked Bolt to build an application that could help me pack a school lunch for my son, according to a photo loaded with my refrigerator's content:

The application first analyzed the task and broke it into components. Then it came to work. It generated a basic web interface, chose a image recognition tool to identify food in my refrigerator and developed an algorithm to recommend meals based on those articles.

If ai needed to make a decision, whether I would like the application to list the nutritional facts of the foods it recommended, for example, it took me with several options. Then it would go out and encode a little more. When he hit a hitch, he tried to purify his own code or retreated to the pace before he had stuck and proven a different method.

Approximately 10 minutes after having entered my warning, Lunchbox Buddy, which is what the ai ​​had decided to call my application, was ready. You can try it for yourself here. (The version I created incorporates a ai image recognition tool that costs money to use; for this public web version, I have replaced it with a simulated image recognition function so as not to accumulate a large invoice).

Not all my Vibecoding experiments have succeeded. I have been fighting for weeks to build a “entry tray car” tool capable of responding to my emails automatically, in my writing style. I have found obstacles when I try to integrate workflows of ai in applications such as Google Photos and the voice notes of iOS, which are not designed to play well with third -party accessories.

And, of course, the ai ​​occasionally makes mistakes. Once, when I tried to build a website for a tire store in my neighborhood, the ai ​​formed false criticism of the store's Yelp page and added them to a testimonial page. On another occasion, when I tried to turn a long history that I had written on an interactive website, the ai ​​included approximately half of the text and left the other half.

Vibecoding, in other words, still benefits from having humans who supervise robots, or at least floating nearby. And it is probably better for hobbing projects, not essential tasks.

That might not be true for much longer. Many ai companies are working on software engineering agents that could completely replace human programmers. Already, he is achieving <a target="_blank" class="css-yywogo" href="https://the-decoder.com/code-competition-codeforces-bans-ai-code-as-as-it-reaches-new-heights-that-cannot-be-overlooked/” title=”” rel=”noopener noreferrer” target=”_blank”>World class scores In competitive programming tests, and several large technological companies, including Google, have subcontracted a large part of their engineering work to artificial intelligence systems. (Sundar pichai, executive director of Google, <a target="_blank" class="css-yywogo" href="https://fortune.com/2024/10/30/googles-code-ai-sundar-pichai/” title=”” rel=”noopener noreferrer” target=”_blank”>He recently said The code generated by ai constituted more than a quarter of all new codes implemented in Google).

If I were a junior programmer, the type of ai seems more likely to replace, I could panic about my labor perspectives. But I am only a guy who likes to play and build tools that improve my life in small ways. And vibecoding, or real coding, is an area where ai is unequivocally improving.

Since he talked about my Vibecoding experience in my podcast last week, I have heard of dozens of other people who have been building their own tools with ai assistance. The colleagues have told me about the nutrition applications that they have created to help them meet their diets, or the tools they are using to summarize the email bulletins they obtain. Readers have sent websites in which they have built Egg price monitoringor Zillow scraped listings in Los Angeles A Discover rental instances After Palisades fire.

Few of these tools change the world for their own right. The new and remarkable thing is that with some keys of keys, fans can now build products that engineers would have previously required.

I am not Pollyannaish on ai, or blind for the effects that IA coding applications could have in society if they continue to improve. I think it is possible that an ai that automates the creation of useful software can also automate the creation of malicious code, or even lead to autonomous cyber attacks. And I worry that software engineering is only the first white collar profession to experience the work replacement effects of ai tools.

But for now, building applications to automate annoying tasks or that require a lot of time in my life seems like a good use of ai as any other. So I'm going to continue vibrating, at least until my son can pack her own lunch.