Vibe Coding
Vibe Coding is an AI-assisted software development approach centered on natural language interaction, proposed by OpenAI co-founder Andrej Karpathy. Its core concept involves developers describing intent and requirements to large language models, which then generate code rather than requiring manual line-by-line writing. The approach emphasizes human-AI collaboration, with developers assuming roles of requirement description, direction guidance, and result verification. It aims to lower programming barriers and improve development efficiency while placing new demands on code quality control and system architecture comprehension.
Core Facts
Generative AI refers to artificial intelligence systems that can automatically generate text, images, code, and other content based on input prompts.
The two root causes of the 'AI flavor' in AI-generated writing are training data bias and AI's lack of genuine creative intent.
The concept of vibe coding was first proposed by OpenAI co-founder Andrej Karpathy in February 2025.
Timeline (last 90 days)
Projects suitable for getting started with Vibe Coding include landing pages, calculators, task trackers, invoice tools and small business tools.
In Vibe Coding, using Git archiving (commits) can avoid being forced to start over from scratch after AI makes large-scale changes that introduce hard-to-locate bugs.
The lower bound of Vibe Coding depends on precision of expression, and the upper bound depends on precision of requirements; even without writing code, one can use AI to complete a demonstration of a full web application (such as a Hangzhou one-day trip planner).
The five key actions of Vibe Coding are "think, frame, save, capture, feed": first let the AI ask about requirements, define boundaries, archive with Git, capture on-site details, and feed specific information.
When fixing a bug, you should change only the smallest scope: when adding a feature, only add the feature; when fixing a problem, only fix the problem; do not mix them in the same conversation, so as to improve verifiability and rollback capability.
In Vibe Coding frontend debugging, handing screenshots to an AI Agent is an inefficient approach, because screenshots can only convey visual appearances and cannot provide DOM-structure-level diagnostic information.
Vibe coding is better suited to prototype validation and internal tools; if used in a production environment, human review of the generated code's security, performance, and maintainability is still required.
Browser ports of classic games such as Halo, The Simpsons: Hit & Run, and GTA have been produced through vibe coding combined with AI decompilation techniques.
Vibe Coding has lowered the barrier to programming, enabling people who do not master technical stacks such as HTML, CSS, and JavaScript to create web pages or small applications through AI programming assistants.
The practical goal of the course is to guide students in personally writing a mobile web page that can be opened and viewed directly on a smartphone.
40 more timeline events
All Facts (20)
Generative AI refers to artificial intelligence systems that can automatically generate text, images, code, and other content based on input prompts.
90%VerifiedThe two root causes of the 'AI flavor' in AI-generated writing are training data bias and AI's lack of genuine creative intent.
80%VerifiedThe concept of vibe coding was first proposed by OpenAI co-founder Andrej Karpathy in February 2025.
80%VerifiedVibe Coding（氛围编程）是Andrej Karpathy在2025年初提出的概念，指开发者完全通过自然语言描述需求让AI生成全部代码
80%VerifiedTraditional MVP development typically takes weeks to months, whereas Vibe Coding compresses this cycle to hours or days.
80%VerifiedKarpathy formally introduced the concept of Vibe Coding on social media in early 2025.
80%VerifiedVibe Coding是指开发者以自然语言描述意图、由AI完成大部分代码生成的编程范式
80%VerifiedVibe Coding has significantly lowered the barrier for non-technical personnel to participate in software development, enabling product managers to quickly build prototypes and tools.
80%VerifiedVibe Coding is the ultimate extension of prompt engineering in the software development context.
80%VerifiedThe Vibe Coding concept was proposed by Andrej Karpathy in early 2025.
80%VerifiedThe Vibe Coding concept was first proposed by former Tesla AI Director and OpenAI co-founder Andrej Karpathy in early 2025.
80%VerifiedIn the process of repeatedly patching bugs, AI introduces redundant logic, duplicate code, and contradictory implementations, causing technical debt to accumulate far faster than in traditional development.
80%VerifiedThe core philosophy of Vibe Coding is that developers no longer care about how the code is written, only whether the results generated by AI are correct.
80%VerifiedVibe Coding and GitHub Copilot code completion are fundamentally different: in the former, AI is the primary productive force, while in the latter, manual coding remains dominant.
80%VerifiedThe concept of Vibe Coding was proposed by OpenAI co-founder Andrej Karpathy
80%VerifiedThe core philosophy of Vibe Coding is to shift the focus from 'how to write code' to 'how to describe requirements,' which is highly aligned with prompt engineering.
80%VerifiedDeep human involvement when using AI coding tools—actively making corrections, understanding code logic, and maintaining control over direction—remains indispensable.
75%VerifiedGoogle AI Studio supports a Vibe Coding feature, allowing users to quickly generate application prototypes through natural language descriptions.
75%VerifiedVibe Coding downgrades 'knowing how to write code' from a necessary condition to a bonus, making product thinking, user insight, and requirements-definition ability the new core competencies.
75%VerifiedIf AI coding tools could truly produce working results on the first attempt, users should no longer need to carefully craft prompts in order to get good results.
70%