I am not a coder. I cannot write a single line of Python, JavaScript or C ++. I have never been a software engineer at all, except for a brief period in my teenage and tampering with flash animations, nor did I harass the ambitions to quit journalism for career in the technical industry.
And yet, for the last several months, I have been coding a storm.
Between my compositions: A tool that transfers long podcasts and summarizes, a tool to organize my social media bookmark into a searchable database, a website that tells me whether a piece of furniture will fit into the trunk of my car and an app called Lunchbox Buddy, which helps to analyze my fridge and packed for my son’s school lunch.
These works are all possible for artificial intelligence, and a new AI trend known as “vibkoding”.
Vibecoding, a word that was popularized by AI researcher Andrej karpathiIn this way, the useful shorthand is that today’s AI equipment allows even non -inexpected enthusiasts to create fully working apps and websites, just by typing the prompt in a text box. You do not have to know how to code the vibcode – just an idea, and a little patience, usually sufficient.
“This is not really coding,” Mr. Karpathi wrote This month. “I just see the goods, say the goods, run the stuff, and copy the paste, and it mostly works.”
The purpose of my own vibecoding experiments says “Software for One” – Small, BESPOKE apps that solve specific problems in my life. These are not the type of equipment that will create a large technology company. There is no real market for them, their characteristics are limited and some of them only work.
But in this way the creation of software-describing a problem in one or two, then to see a powerful AI model is a man-up experience to work for the construction of a custom tool to solve it. This creates a sense of AI vertigo, as I first felt after using the chat. And this is the best way I have displayed to doubt the capabilities of today’s AI model, which can now automate large parts of basic computer programming, and soon be able to similar tricks in other areas.
AI coding equipment exists for years. Earlier, like Githib Copilot, the professional kodar was designed to help work faster, similarly by completing your lines of the code, the way the chat completes a sentence. You still need to know how to code to get the most of them, and when AI is stuck, steps.
But in the last one or two years, new tools have been created to take advantage of the more powerful AI model that enable even neophytes to perform programs like professionals.
These devices, including cursor, ripalat, bolt and cute, all work in the same way. Given the user’s signal, the tool comes with a design, decides to use the best software package and programming languages, and works for the manufacture of a product. Most products allow limited free use, with paid levels that unlock the ability to build better features and more things.
For a non-programmer, vibcoding may feel like sorce. After typing in your prompt, the mysterious lines of the code fly in the past, and after a few seconds, if everything goes well, a working prototype emerges. Users can suggest twicks and modifications, and when they are happy with it, they can deploy their new product on the web, or run it on their computer. Based on the complexity of the project, this process can take just a few minutes or several hours.
Here it seemed when I asked Bolt to create an app that could help me pack a school lunch for my son, based on the picture uploaded of my fridge contents:
The app first analyzed the work and broke it into the component parts. Then it started working. This generated a basic web interface, chose an image recognition tool to identify foods in my fridge and developed an algorithm to recommend food based on those items.
If AI needed to take a decision – I wanted the app to list the nutritional facts of foods that were recommended, for example – it inspired me with many options. Then it will stop and there will be some more code. When it collides with a snag, he tried to debug his own code, or before he stuck and supported the step before trying a different method.
About 10 minutes later I entered my Prompt, Lunchbox Buddy – which AI decided to call my app – was ready. You can try it for yourself Here(The version I have created includes an AI image recognition tool, which spend money to use; for this public web version, I have replaced it with a fake image recognition feature, so I do not rack a huge bill.)
All my vibcoding experiments have not succeeded. I have been struggling for weeks to create a “inbox autopylyt” equipment capable of my automatically answering my email in my writing style. I have faced obstacles while trying to integrate AI work flow in apps such as Google Photo and iOS Voice Memo, which are not designed to play well with third-party add-on.
And, of course, AI sometimes makes mistakes. Once, when I tried to create a website for a tire shop in my neighborhood, AI made a fake review from the shop’s yelp page and added them to a launch page. Another time, when I tried to turn a long story, which I wrote in an interactive website, AI included about half a lesson and left the other half.
Vibecoding, in other words, still benefits humans to look after robots, or at least hovering nearby. And it is perhaps the best for hobby projects, not the necessary task.
This may not be true for a very long time. Many AI companies are working on software engineering agents that can completely change the human programmer. Already, AI is receiving World class score On competitive programming tests, and many large technical companies including Google have outsourced a large part of their engineering work for the AI system. (Beautiful Pichai, Chief Executive of Google, Recently said AI-Janit Code was made more than one-fourth of all the new codes posted on Google.)
If I have been a junior programmer – the way AI is likely to change the most – I can panic about my job prospects. But I am just a man who likes to tinker, and manufactures devices that make my life better than small ways. And vibecoding – or real coding – is an area where AI is unwanted improving it.
Since talking about my vibkoding experience at my podcast last week, I have heard from dozens of others who are creating their own equipment with AI aid. Co -workers have told me about the nutrition apps that they have created to help them stick to their diet, or they are using devices they are using to achieve email newsletters. Readers have sent to the websites they have created Track the price of eggsOr scrap the district listing in Los Angeles Find examples of rent-gauzing Fire after pulses.
Some of these devices are world-up in themselves. What is new and notable that with some keystrokes, amateur can now manufacture products that already require teams of engineers.
I am not polynamon about AI, or blind to the effects that can be on the AI coding apps society if they improve. I think it is possible that an AI that automatically makes the manufacture of useful software can automate the construction of malicious code, or even lead the autonomous cyber attack. And I worry that software engineering is only the first white collar profession to experience the labor-replacement effects of the AI Tool.
But for now, it is a good use of AI in any way to construct apps to automate the tasks that take annoying or time -taking tasks in my life. So I am going to keep vibcoding – at least until my child can pack his lunch.
,
#Vibkoding #helped #software
