Andrej Karpathy, co-founder of OpenAI, recently coined the term “vibe coding” in a tweet that sparked online debates. Vibe coding refers to a new approach to programming where developers fully embrace AI-powered coding assistants, focusing on the intent and forgetting about the code itself. As a tech journalist with 15 years of experience, I decided to try my hand at vibe coding.
I’ve always had an interest in programming but often struggled to complete projects due to the time and effort required to learn traditional coding methods. Late last year, I used Claude 3.5 Sonnet, an AI coding assistant, to replicate my personal website. I provided the AI with my site’s content and a screenshot, and it generated a functional website using HTML, CSS, and JavaScript.
With a bit of tweaking, the AI-generated site was even better than the original WordPress version. Following Karpathy’s approach, I didn’t focus on understanding the code deeply. Instead, I told the AI what to do, copied the results into VSCode, and reloaded my browser to see the changes.
This method was enjoyable enough that I’ve since tackled several other coding projects, including a dice roller web app, a tool to simplify rolling dice, and a puzzle game for Pico-8. Notably, these projects were in different programming languages—HTML/JavaScript, Python, and Lua—none of which I’m proficient in.
Vibe coding: Embracing AI in programming
I’ve also experimented with C# while initially building the puzzle game in Unity before deciding to switch to Pico-8. While vibe coding is fast, fun, and approachable, it does come with risks. Critics worry that it could lead to an era of AI-driven mediocrity in software, similar to the low-quality content seen in blogs, forums, artwork, videos, and music.
AI-generated code can contain flaws that aren’t immediately obvious, and inexperienced programmers may struggle to spot bugs and security issues. However, many people, like myself, see vibe coding as a fun way to tackle simple projects previously out of reach. Others, including software engineers, use it to build prototypes, demonstrate proofs of concept, and learn.
In a way, vibe coding reminds me of the early days of blogging and social media, which paved the way for my career in tech journalism. With AI assistance, people interested in programming don’t necessarily need formal training. Anyone familiar with a laptop can ask ChatGPT about the tools they need, how to install them, and generate the code for their projects.
While mistakes will be made and some generated code will be subpar, vibe coding has the potential to open programming to more people who lack the patience for traditional learning methods. As AI continues to evolve, it will be interesting to see how this new approach shapes the future of software development.
Image Credits: Photo by Fotis Fotopoulos on Unsplash
Cameron is a highly regarded contributor in the rapidly evolving fields of artificial intelligence (AI) and machine learning. His articles delve into the theoretical underpinnings of AI, the practical applications of machine learning across industries, ethical considerations of autonomous systems, and the societal impacts of these disruptive technologies.
