Claude Code Lead Says 'Vibe Coding' Has Run Its Course

Boris Cherny, the head of Anthropic's Claude Code, is growing tired of the term 'vibe coding' — and he thinks the AI development community deserves a more serious label for what has become a billion-dollar industry shift. The engineer behind one of the hottest AI coding tools on the market says he is actively searching for a replacement term that better captures the scale and sophistication of modern AI-assisted software development.

The phrase 'vibe coding' was coined by OpenAI co-founder Andrej Karpathy in early 2025 to describe the emerging practice of developers using AI tools and agents to write code. It quickly went viral, becoming shorthand for any workflow where humans guide AI systems to generate software. Collins Dictionary even selected it as a word of the year candidate in November. But Cherny believes the term has outlived its usefulness.

Key Takeaways

• Boris Cherny, creator of Claude Code, says the term 'vibe coding' no longer fits the reality of AI-assisted development
• Claude Code and OpenAI's Codex are each generating billions of dollars in revenue for their parent companies
• The term was originally coined by OpenAI co-founder Andrej Karpathy in early 2025
• Cherny asked Claude itself to suggest a replacement — it proposed 'agentic engineering'
• The debate reflects a deeper tension between AI coding as a novelty vs. a professional-grade tool
• Collins Dictionary named 'vibe coding' a word of the year in November 2025

Why 'Vibe Coding' Feels Too Casual for a Billion-Dollar Industry

Cherny's frustration stems from a fundamental mismatch between language and reality. When Karpathy first introduced the concept, AI coding tools were still experimental curiosities — fun to play with, occasionally useful, but far from production-ready. The word 'vibe' captured that playful, exploratory energy perfectly.

Fast forward to today, and the landscape looks radically different. Claude Code and OpenAI Codex are generating millions of lines of production-quality code. These tools are not toys; they are infrastructure. Enterprises across the Fortune 500 rely on them daily, and they contribute billions of dollars to the revenue streams of Anthropic and OpenAI respectively.

In that context, describing the work as 'vibing' feels reductive. It is the equivalent of calling cloud computing 'internet dabbling' or referring to machine learning as 'computer guessing.' The terminology undermines the seriousness of the technology and the engineering discipline required to use it effectively.

Cherny himself has previously stated that Claude Code has essentially 'solved' programming for him personally — a bold claim that underscores just how far beyond casual experimentation these tools have evolved.

The Search for a Better Term

In a move that perfectly illustrates the AI-first mindset at Anthropic, Cherny turned to Claude itself for naming suggestions. The chatbot's top recommendation was 'agentic engineering,' a term that Karpathy had also floated as a more technical alternative.

However, the phrase has struggled to gain traction. Here is why the naming problem is harder than it looks:

• 'Agentic engineering' sounds too academic and jargon-heavy for mainstream adoption
• 'AI-assisted coding' is accurate but generic and uninspiring
• 'Copilot coding' is too closely associated with Microsoft's GitHub Copilot brand
• 'Prompt-driven development' focuses too narrowly on one interaction pattern
• 'Autonomous programming' overstates current capabilities and raises concerns

The ideal term needs to be catchy enough for social media, precise enough for technical discussions, and broad enough to encompass the full spectrum of AI coding workflows — from simple autocomplete suggestions to fully autonomous multi-file code generation.

This is not just a branding exercise. Language shapes perception, and perception shapes investment, adoption, and regulation. If AI coding continues to be framed as casual 'vibing,' it may struggle to earn the institutional credibility it needs to become a standard part of enterprise software development.

The Billion-Dollar Stakes Behind the Naming Debate

The financial context makes Cherny's frustration entirely understandable. Anthropic's AI coding products have become a critical revenue driver for the company, which has raised over $7 billion in funding and is valued at roughly $60 billion as of early 2025.

OpenAI, meanwhile, launched Codex as a dedicated coding agent built on top of its latest models. Both companies are locked in an intense competition to dominate the AI development tools market, which analysts estimate could be worth $45 billion or more by 2028.

The revenue breakdown tells the story:

• Anthropic reportedly earns a significant portion of its API revenue from coding-related use cases through Claude Code
• OpenAI's Codex operates as a cloud-based coding agent running in sandboxed environments
• Google's Gemini Code Assist is aggressively expanding its enterprise footprint
• Microsoft's GitHub Copilot surpassed 1.8 million paid subscribers earlier this year

With this much money on the line, the industry needs language that communicates value and reliability to CIOs, CTOs, and procurement teams — not just excitement to developers on social media.

How AI Coding Has Evolved Beyond 'Vibes'

The original concept of vibe coding described a specific workflow: a developer describes what they want in natural language, lets the AI generate the code, and accepts or rejects the output based on whether the result 'feels right' — hence the 'vibe.' There was minimal code review, limited testing, and a heavy reliance on intuition.

Today's AI coding tools operate at a fundamentally different level. Claude Code functions as a command-line agent that can navigate codebases, run tests, fix bugs, and execute multi-step development tasks autonomously. It does not just generate code snippets — it understands project structure, dependencies, and architectural patterns.

Similarly, OpenAI's Codex can read entire repositories, write code across multiple files, and validate its own output by running test suites in sandboxed environments. These are not 'vibe' tools. They are engineering systems.

The gap between early 2025 vibe coding and late 2025 agentic development is comparable to the gap between a pocket calculator and a spreadsheet application. Both do math, but they operate at entirely different scales of complexity and capability.

Industry Context: A Maturing Market Needs Mature Language

Cherny's push for better terminology reflects a broader maturation happening across the AI industry. As AI tools move from experimental novelties to mission-critical infrastructure, the language around them inevitably evolves.

We have seen this pattern before in tech. 'Surfing the web' gave way to 'digital transformation.' 'The cloud' evolved from a vague metaphor to a precisely defined set of infrastructure services. 'Big data' eventually yielded to more specific terms like 'data engineering' and 'analytics infrastructure.'

AI coding is at a similar inflection point. The technology has matured faster than the vocabulary, and the old terminology now acts as a constraint on how seriously the industry is taken.

This matters especially in regulated industries — finance, healthcare, defense — where decision-makers need to trust that AI-generated code meets rigorous quality standards. Telling a bank's chief technology officer that your team uses 'vibe coding' is unlikely to inspire confidence, regardless of how capable the underlying tools actually are.

What This Means for Developers and Businesses

For individual developers, the terminology debate may seem academic. But it has real practical implications:

• Job descriptions increasingly reference AI coding skills, and the terminology used affects how these roles are perceived and compensated
• Enterprise adoption depends on AI coding being framed as a professional discipline, not a casual experiment
• Regulatory frameworks will be shaped partly by how AI-generated code is described and categorized
• Open-source communities need shared vocabulary to build standards and best practices

Developers who want to future-proof their careers should pay attention to how this language evolves. The engineers who can articulate the difference between casual AI code generation and disciplined AI-augmented development will have a significant advantage in the job market.

Looking Ahead: What Comes After 'Vibe Coding'?

The replacement term has not yet emerged, and it may take months — or even a new breakthrough — before the industry settles on a consensus label. Cherny's public dissatisfaction with 'vibe coding' signals that Anthropic may try to lead the rebranding effort, potentially tying a new term to Claude Code's marketing.

If history is any guide, the winning term will likely come not from a corporate brainstorming session or an AI chatbot's suggestions, but from organic developer usage. The phrase that sticks will be the one that developers naturally reach for when describing their daily workflow with these tools.

Until then, 'vibe coding' remains the default — even if its most prominent practitioner wishes it would go away. The irony is not lost on the community: the man who says AI has solved coding for him cannot solve the much simpler problem of finding the right two words to describe it.

One thing is clear: whatever we end up calling it, AI-assisted software development is no longer a vibe. It is a discipline, a business, and increasingly, the default way software gets built.