The routine of modern developers is changing rapidly. If earlier programming meant writing accurate instructions for the computer and their multiple repetition, now it is enough to describe the desired result in the usual language – and artificial intelligence itself converts the industrial plate into a working code.
This new approach was called “Waib Coding”. It became possible thanks to powerful language models created by Openai and Anthropic, as well as special tools like Cursor Composer, Github Copilot and Replit Agent. Now even people without special education can create programs, just talking with a neural network.
The term itself appeared thanks to the former Openai researcher Andrei Karpati. On the second of February, he published a post on the social network X, where he described this method as a complete immersion in the development stream, in which the programmer “releases” control and “forgets about the existence of code”. According to him, the process boils down to the fact that the developer “watches something, says something, starts something and copies something”-and somehow all this works.
In contrast to traditional programming, where each line of code is thoroughly verified, with a wyb-coding, the developer acts intuitively. When an error occurs, its text is simply copied back into the dialogue with AI, accept the proposed changes and repeat the process until the program is earned. Carpathy with humor admits that he does not even seek the necessary sections of the code himself – instead, asks the II to perform the simplest tasks, for example, “reduce the indentation on the side panel.”
The popularity of the new method is growing: Cursor already has 40,000 paid users, and GitHub Copilot used 1.3 million people in February 2024. Wayb-coding is especially active in games developers. Recently, Peter Young from Microsoft demonstrated how, with the help of voice commands and AISSISTROS CURSOR and CLAUDE 3.7, Sonnet created a simple three -dimensional shooter with a zombie – he simply described the desired changes, and artificial intelligence embodied them in the code.
An independent developer and researcher of AI Simon Willison sees in Wibe-Coating an excellent tool for the quick creation of prototypes and testing ideas. However, it warns against the use of this method in industrial development, where the quality and understandability of the code is critical. “Wyb -coding is fun until you have to engage in wyb -laying,” another developer, bin Sout.
Ironically remarked.
The main problem of the method is that AI can generate error code or even refer to non -existent functions and libraries. However, this approach has a built -in verification mechanism: an incorrect code simply will not work. This creates the natural boundaries of the applicability of wyb-conding and helps to cut off erroneous solutions.