Detailed Analysis
Boris Cherny, the engineer at Anthropic who created Claude Code, has publicly expressed frustration with the term "vibe coding," a phrase that has saturated the AI and software development discourse since early 2025. The term, originally coined by AI researcher and former OpenAI scientist Andrej Karpathy, describes an approach to software development in which practitioners largely surrender precise control to AI systems, accepting generated code with minimal scrutiny and relying on intuition and iteration rather than rigorous engineering discipline. Cherny's objection — coming from the individual who built one of the most prominent AI-assisted coding tools on the market — carries considerable weight and reflects a growing tension within the professional software community over how AI-augmented development is being characterized and understood.
Cherny's discomfort with the phrase is significant precisely because of his vantage point. Claude Code, Anthropic's agentic command-line coding tool, is designed to handle complex, multi-step engineering tasks autonomously, from writing and editing files to executing terminal commands and navigating large codebases. The tool is explicitly aimed at professional developers working on real-world software projects, not casual users generating throwaway scripts. For Cherny, the "vibe coding" framing likely undersells the depth of human judgment, architectural oversight, and iterative verification that responsible AI-assisted development actually requires — flattening a sophisticated collaborative practice into something that sounds closer to reckless improvisation.
The debate over terminology matters beyond semantics because it shapes how organizations, educators, and policymakers think about AI's role in software engineering. If "vibe coding" becomes the dominant frame, it risks normalizing an approach where code quality, security, and maintainability are deprioritized in favor of speed and convenience. This has real consequences: AI-generated code that is not carefully reviewed can introduce subtle bugs, security vulnerabilities, or architectural debt that compounds over time. Cherny's pushback implicitly argues that the most powerful use of tools like Claude Code is not abdication of engineering responsibility but rather an elevation of it — using AI to handle lower-level execution so that human engineers can focus on higher-order design and judgment.
Broader trends in AI development reinforce why this framing debate is consequential. As AI coding assistants proliferate — from GitHub Copilot to Cursor to Claude Code to Google's Gemini Code Assist — the industry is actively negotiating what it means to be a competent software engineer in an AI-augmented world. There is a genuine bifurcation emerging between developers who use AI as a force multiplier for expert-level work and those who use it as a crutch that bypasses the need to develop deep expertise at all. Cherny's stance aligns him firmly with the former camp and signals that Anthropic, at least at the level of its tooling philosophy, is building products intended for serious engineering workflows rather than for the lowest-friction path to running code. His objection to "vibe coding" is, in this sense, also a statement about what Claude Code is and is not meant to be.
Read original article →