The Paradigm Shift范式转变
Traditional software engineering focuses on syntax correctness. Vibe coding inverts this, prioritizing review and iteration. The human moves from “Writer” to “Manager.” 传统软件工程注重语法正确性。Vibe coding 将其反转，优先考虑审查和迭代。人类的角色从“编写者”变为“管理者”。
⚖️ Effort Allocation精力分配
In Vibe Coding, the heavy lifting of syntax generation is offloaded. The “Tax” of coding shifts from typing/looking up docs to Reading, Reviewing, and Verifying. 在 Vibe Coding 中，语法生成的繁重工作被卸载。编程的“税收”从输入/查找文档转变为阅读、审查和验证。
Traditional Coding传统编程
80% Writing / 20% Reviewing80% 编写 / 20% 审查
Vibe Coding
20% Prompting / 80% Reviewing & “Vibe Checking”20% 提示 / 80% 审查与“氛围检查”
“I just write code in English now... I basically just monitor the vibes.” — Andrej Karpathy
Developer Time Distribution (Estimated)开发者时间分配（估算）
The “Vibe” Loop“Vibe” 循环
Vibe coding isn’t just “asking ChatGPT.” It’s an iterative loop similar to the OODA loop (Observe-Orient-Decide-Act). Click the steps below to explore the workflow. Vibe coding 不仅仅是“问 ChatGPT”。它是一个类似于 OODA 循环（观察-判断-决策-行动）的迭代过程。点击下方步骤探索工作流程。
1. Prompting the Intent
Instead of writing `def sort_list(x):`, the developer writes “Make the list sorting robust to null values and case-insensitive.” The focus is on business logic and edge cases, describing the 'shape' of the solution.
Expert Tip专家建议
“Treat the AI like a junior developer who types fast but makes weird mistakes. Be specific about constraints.”
Expert Perspectives专家观点
What the industry leaders are saying.行业领袖们在说什么。
The “Vibe” Trade-off“Vibe” 权衡
Vibe coding accelerates velocity but introduces “Illusion of Competence.” Adjust the slider to see how leaning too heavily on AI changes the risk profile. Vibe coding 加速了开发速度，但引入了“能力幻觉”。调整滑块查看过度依赖 AI 如何改变风险状况。
Projected Outcome预期结果
At this level, you are reviewing generated code carefully. Velocity is high, but you maintain a mental map of the codebase.在这个水平，你仔细审查生成的代码。速度很快，但你保持对代码库的全局认知。
> Code is clean, but subtle bugs may hide in edge cases.> 代码很干净，但边界情况中可能隐藏着细微的 bug。
Deep Dive Analysis深度分析
The term, popularized by Andrej Karpathy, refers to a workflow where the developer focuses on the high-level system behavior and relies on LLMs to handle the syntax. It is distinct from “No Code” because the output is still code (Python, JS, etc.), but the developer interacts with it primarily through English prompts and diff reviews.
The “Vibe” aspect refers to the quick visual or behavioral check a developer does to ensure the code “feels” right—runs without errors, looks correct on screen—often replacing rigorous TDD (Test Driven Development) in the early stages of prototyping.
这个术语由 Andrej Karpathy 推广，指的是开发者专注于高层系统行为，依赖 LLM 处理语法的工作流程。它与“无代码”不同，因为输出仍然是代码（Python、JS 等），但开发者主要通过英语提示和 diff 审查与之交互。
“Vibe” 方面指的是开发者进行的快速视觉或行为检查，确保代码“感觉”正确——没有错误地运行、在屏幕上看起来正确——这通常在原型设计早期替代了严格的 TDD（测试驱动开发）。
The primary criticism of Vibe Coding is that it encourages a superficial understanding of the codebase. If a developer only checks if the app “looks” right (the vibe), they may miss subtle logic errors, security vulnerabilities, or memory leaks that don’t manifest immediately.
Senior engineers argue that reading code is harder than writing it. By generating massive amounts of code, Vibe Coding increases the “reading load” significantly, potentially leading to “Review Fatigue” where developers just accept AI suggestions without understanding them.
对 Vibe Coding 的主要批评是它鼓励对代码库的表面理解。如果开发者只检查应用“看起来”是否正确（氛围），他们可能会遗漏细微的逻辑错误、安全漏洞或内存泄漏。
资深工程师认为阅读代码比编写代码更难。通过生成大量代码，Vibe Coding 显著增加了“阅读负担”，可能导致“审查疲劳”，开发者不理解就接受 AI 建议。
As LLMs improve, the role of a software engineer is predicted to shift further towards System Architecture and Product Management. The ability to articulate *what* needs to be built (in English) will become more valuable than knowing *how* to implement a specific sorting algorithm in C++.
随着 LLM 的改进，软件工程师的角色预计将进一步转向系统架构和产品管理。能够用英语表达“需要构建什么”的能力将比知道“如何”用 C++ 实现特定排序算法更有价值。