Vibe coding: when AI takes over the code and the developer defines the problem

The birth of a new concept

In 2025, Andrej Karpathy, co-founder of OpenAI, coined the term “vibe coding” to describe the practice of creating software through conversations with artificial intelligence. In this approach, the programmer describes what they want to build, and the AI generates most of the necessary code.

Specification takes precedence over syntax

With AI tools capable of generating functional code snippets, the quality of the output depends more on the clarity of the instructions than on mastery of a specific programming language. A generic request, such as creating a financial control app, leaves many questions unanswered: what data will be recorded, how expenses will be categorized, who will have access to the information, and what should happen in case of an error.

A detailed description enables the professional to define the objective, identify the target audience, set constraints, and establish criteria for validating the final product. After the code is generated, they test its functionality and adjust the direction as needed.

Technical expertise remains essential

Vibe coding was initially associated with experimental, short-term projects where developers didn’t need to monitor every line of code produced by AI. For simple prototypes, this approach may suffice. However, in systems handling sensitive data, financial transactions, security, or large user bases, the ability to review code, identify flaws, and understand the architecture remains crucial.

New skills come into play

The division of tasks shifts: while AI handles implementation, humans define the problem and success criteria. This change values professionals who can articulate their needs precisely, provide context, evaluate versions, and suggest improvements.

For tech professionals, learning to craft effective commands becomes an additional skill. The barrier to creating a first software product is lowered when natural language guides the process, but transforming an idea into a reliable product still requires human judgment.