Despite the misleading connotations of "vibe coding," Ng emphasizes that AI-assisted coding is intellectually demanding and requires deep technical understanding. Speaking at the LangChain Interrupt conference, he described it as "a deeply intellectual exercise" that can be as taxing as traditional coding methods.12 This perspective challenges the notion popularized by OpenAI co-founder Andrej Karpathy that developers can simply "fully give in to the vibes" and "forget the code even exists."2

Ng maintains that learning to code remains essential in the AI era, comparing it to literacy in modern society.3 He argues that coding knowledge enhances one's ability to communicate effectively with AI tools, create more precise prompts, and interpret errors correctly.1 Rather than fearing AI will make programming obsolete, Ng suggests that AI tools should be viewed as empowering extensions that require foundational programming knowledge to use effectively, similar to how high-level programming languages build upon fundamental software engineering principles.45

The term "vibe coding" has evolved far beyond Andrej Karpathy's original definition, creating significant misconceptions about AI-assisted programming. Karpathy initially described it as an approach for "throwaway weekend projects" where one might "fully give in to the vibes" and "forget that the code even exists"1 – specifically for low-stakes, experimental work. Simon Willison emphasizes that vibe coding is "not the same as professional AI-assisted development" and should be understood as "playful and low-stakes"2.

Many professionals worry about this terminology bleeding into broader discussions about AI in software development. As one critic noted, labeling all AI-assisted programming as "vibe coding" risks "flattening important distinctions"2 between casual experimentation and serious engineering. Even Karpathy appears uncomfortable with how the term has expanded, now carefully distinguishing between "real coding" and pure "vibe experiments"2. This confusion undermines the reality that responsible AI-assisted coding still demands rigor: "reviewing code, testing it, understanding it, explaining it"2 – precisely the "intense intellectual challenge" that Andrew Ng describes3.

Despite the rise of AI coding assistants, fundamental coding skills remain essential in the AI era. Google's head of research has emphasized that learning to code is still crucial even as AI generates over a quarter of Google's new code1. The ability to review, understand, and evaluate AI-generated code requires developers to possess strong programming fundamentals – as one expert puts it, "you have to know good code to approve good code"2.

Learning to code in the AI era offers distinct advantages beyond just writing software:

• It develops critical and logical thinking applicable across diverse industries3

• It trains humans to understand how computers process information3

• It enables effective collaboration with AI tools through better prompt engineering4

• It provides the foundation for evaluating AI outputs and identifying potential errors15

• It remains essential for designing, implementing, and refining AI-driven solutions6