Whether you are streaming a show, paying bills online, or sending emails, each of these operations depends on a computer program running behind the scenes. The process of writing a computer program is called encoding. Until recently, most computer code was written by humans at least initially. However, with the advent of generative artificial intelligence, this has begun to change.
Now, just like you can ask Chatgpt to rotate recipes for your favorite dishes or write sonnets in Lord Byron's style, you can now ask the generated AI tools to write computer code for you. Openai co-founder Andrej Karpathy previously led the AI efforts at Tesla, which he recently called "atmosphere coding".
For a full beginner or non-technical dreamer, writing code based on resonance rather than clearly defined information may be like a superpower. You don't need to master programming languages or complex data structures. A simple natural language prompt will solve the problem.
How it works
Vibe encoding relies on standard patterns in technical languages, which AI systems use to piece together original code from its training data. Any beginner can use an AI assistant, such as GitHub Copilot or cursor chat, put some tips, and then get the system to work. Here is an example:
“Create an active and interactive visual experience that reacts to music, user interaction or real-time data. Your animation should include smooth transitions and colorful, lively visuals with an engaging experience. Animation should have organic music, interactions with music, user interactions or reactions to users’ interactions or real-time data and allow experience with JavaScript and Reactive, and allow experience with JavaScript and Reactivity, and allow use of JavaScript and Reactive, and allow use of JavaScript and other reactions.
But AI tools do this without really mastering specific rules, edge cases, or security requirements about the software. This is far from the process behind developing production-grade software, which must balance the tradeoffs between product demand, speed, scalability, sustainability, and security. A skilled engineer writes and views code, runs tests and creates security barriers before they go live.
But despite the lack of structured processes that can save time and reduce the skills required for coding, there are trade-offs. With Vibe encoding, most of these stress testing practices have disappeared, leaving the system vulnerable to malicious attacks and personal data breaches.
And there is no simple fix: If you don't know every or any line of code written by your AI agent, you can't fix the code when that code breaks. Or worse, as some experts have pointed out, you won't notice when it fails silently.
AI itself does not have the ability to perform this analysis. It identifies what a "working" code usually looks like, but it does not necessarily diagnose or resolve deeper problems that the code may cause or aggravate.
[embed]https://www.youtube.com/watch?v=p7lrycivxga[/embed]
Why it matters
Vibe encoding may just be a flickering phenomenon that will disappear soon, but it may also discover deeper applications for experienced programmers. This practice can help skilled software engineers and developers turn an idea into a viable prototype faster. It can also enable novice programmers and even amateur coders to experience the power of AI, perhaps inspiring them to pursue the discipline more deeply.
Vibe encoding may also mean that natural languages can be a more feasible tool for developing certain computer programs. If so, it will respond to early website editing systems, called Wysiwyg Editors, who promise designers that “what you see is what you get” or “drag and drop” website builders, which makes it easy for anyone with basic computer skills to start a blog.
At the moment, I don't think Vibe encoding will replace experienced software engineers, developers or computer scientists. The discipline and art are much more subtle than AI can handle, and the risk of passing on "Vibe Code" with the advantages of legitimate software is too great.
However, as AI models improve and become more adept at incorporating context and taking risks into account, practices such as Vibe encoding may lead to further blurring of the boundaries between AI and human programmers.