Posts

It is arduous to read and remember a whole prompt written in natural language. For example, Claude’s system prompt is 16,000 words long. If the subject described in the prompt is constantly changing (e.g., API specification), the prompt must be continuously updated, which is a demanding task. This problem is very similar to the issue of inconsistency between traditional software specifications and their implementation. Because specifications are written in natural language, if the person in charge forgets, changes in the implementation may not be reflected in the specification, and discovering such inconsistencies also presents a challenge.

Before vibe coding, there was a no-code solution that ultimately failed. The fundamental reason was the limits of complexity and the difficulties of maintenance. While it was possible to build simple prototypes, creating complex services, ensuring security, and maintaining systems were impossible without understanding the code. Vibe coding, so far, is following a similar path. This phenomenon stems from the inherent ambiguity of natural language. A sentence can be interpreted in multiple ways, and the interpretation itself is natural language, so it is ambiguous too.

Vibe coding is a new way of coding in which developers supervise coding agents using natural language. However, the way of interacting with agents is still chat-style. Chat is not a suitable method of interaction for vibe coding. In chat, what is said is quickly lost. Once lost, it is practically impossible to retrieve. But in vibe coding, what is said to the language model is a kind of code specification, which is very important.