i. 人们使用 vibe 编码构建什么类型的应用? ii. 创建了哪些类型的应用(例如,登陆页面、游戏等)? iii. vibe 编码是否完全适合完成一个项目,还是对于更复杂的项目如游戏需要转向手动编码?
i. What kind of apps do people build using vibe coding?
ii. What types of applications are created (e.g., landing pages, games, etc.)?
iii. Is vibe coding fully appropriate to complete a project, or does it require transitions to manual coding for more complex projects like games?| i. What kind of apps do people build using vibe coding? |
| :--- |
| ii. What types of applications are created (e.g., landing pages, games, etc.)? |
| iii. Is vibe coding fully appropriate to complete a project, or does it require transitions to manual coding for more complex projects like games? |
b. 期望——是完全成功/部分成功,还是仅仅是探索?
i. 人们进入 vibe 编码时是期望完全成功,还是抱着探索的态度? ii. 用户通常期望通过 vibe 编码获得完全成功,还是试验看看这项技术能达到什么程度?
i. Do people go into vibe coding expecting to fully succeed, or are they taking an exploratory attitude?
ii. Are users generally expecting complete success with vibe coding, or do they experiment to see how far the technology can be pushed?| i. Do people go into vibe coding expecting to fully succeed, or are they taking an exploratory attitude? |
| :--- |
| ii. Are users generally expecting complete success with vibe coding, or do they experiment to see how far the technology can be pushed? |
意图
a. 初始意图?
i. 在 vibe 编码中,应用的意图是如何形成和完善的? ii. 意图是在编码开始前预先形成,还是通过与 AI 的迭代交互逐步演变的? iii. 意图的哪些部分是事先确定的,意图如何在不同的提示和阶段中传播?
i. How is the intention for an app formed and refined in vibe coding?
ii. Is the intent pre-formed before coding begins, or does it evolve through iterative interactions with the AI?
iii. Which parts of the intent are established beforehand, and how does intent spread across various prompts and phases?| i. How is the intention for an app formed and refined in vibe coding? |
| :--- |
| ii. Is the intent pre-formed before coding begins, or does it evolve through iterative interactions with the AI? |
| iii. Which parts of the intent are established beforehand, and how does intent spread across various prompts and phases? |
b. 意图如何以及为何被完善?
i. 哪些 AI 工作流程资源(例如,探索、寻找替代方案)支持并完善这一意图?
ii. AI 本身是否通过填补空白来帮助塑造最终意图?与 AI 的反复互动如何影响最终输出?
iii. vibe coding 的迭代和对话性质如何影响用户的创造力和解决问题的策略?
iv. 这一过程是鼓励更多的探索,还是导致对 AI 建议的依赖?
i. What AI workflow resources (e.g., exploration, seeking alternatives) support and refine this intent?
ii. Does the AI itself contribute to shaping the final intent by filling in gaps? How does the back-and-forth interaction with the AI shape the final output?
iii. How does the iterative and conversational nature of vibe coding affect users' creativity and problem-solving strategies?
iv. Does this process encourage more exploration or lead to a reliance on the AI's suggestions?| i. What AI workflow resources (e.g., exploration, seeking alternatives) support and refine this intent? |
| :--- |
| ii. Does the AI itself contribute to shaping the final intent by filling in gaps? How does the back-and-forth interaction with the AI shape the final output? |
| iii. How does the iterative and conversational nature of vibe coding affect users' creativity and problem-solving strategies? |
| iv. Does this process encourage more exploration or lead to a reliance on the AI's suggestions? |
工作流程
a. 工作流程的阶段?
i. vibe 编码工作流程的阶段有哪些? ii. 不同阶段(例如编程、故障定位、调试与规划和测试)如何分配时间和精力? iii. 这些阶段与更传统的 AI 增强编程工作流程以及先前识别的 AI 工作流程(如迭代目标满足)相比如何?
i. What are the stages of the vibe coding workflow?
ii. How is time and effort allocated across different stages (e.g., programming, fault localization, debugging versus planning and testing)?
iii. How do these stages compare with those in more traditional AIenhanced programming workflows and prior identified AI workflows such as iterative goal satisfaction?| i. What are the stages of the vibe coding workflow? |
| :--- |
| ii. How is time and effort allocated across different stages (e.g., programming, fault localization, debugging versus planning and testing)? |
| iii. How do these stages compare with those in more traditional AIenhanced programming workflows and prior identified AI workflows such as iterative goal satisfaction? |
b. 工具和技术组合?
i. 不同工具在 vibe 编码的背景下如何协同工作? ii. 其他资源(如文档、网络搜索等)如何整合到 vibe 编码工作流程中?
i. How do different tools work together in the context of vibe coding?
ii. How are other resources (such as documentation, web searches, etc.) integrated into the vibe coding workflow?| i. How do different tools work together in the context of vibe coding? |
| :--- |
| ii. How are other resources (such as documentation, web searches, etc.) integrated into the vibe coding workflow? |
提示
a. 需要做哪些准备工作?
i. 创作者如何设置一个氛围编码会话(例如,编辑提示指令与使用默认设置)?
b. 明确的提示策略?
i. vibe 编码中的提示策略有哪些? ii. 在编码过程中,采用了哪些方法来启动和引导 AI?
i. What are the prompting strategies in vibe coding?
ii. What methods are employed to initiate and guide the AI during the coding process?| i. What are the prompting strategies in vibe coding? |
| :--- |
| ii. What methods are employed to initiate and guide the AI during the coding process? |
Top Level Category Subcategories Elaboration
Goals a. What do they build? "i. What kind of apps do people build using vibe coding?
ii. What types of applications are created (e.g., landing pages, games, etc.)?
iii. Is vibe coding fully appropriate to complete a project, or does it require transitions to manual coding for more complex projects like games?"
b. Expectation - for full/partial success, or just exploration? "i. Do people go into vibe coding expecting to fully succeed, or are they taking an exploratory attitude?
ii. Are users generally expecting complete success with vibe coding, or do they experiment to see how far the technology can be pushed?"
Intentions a. Initial intention? "i. How is the intention for an app formed and refined in vibe coding?
ii. Is the intent pre-formed before coding begins, or does it evolve through iterative interactions with the AI?
iii. Which parts of the intent are established beforehand, and how does intent spread across various prompts and phases?"
b. Intention refined? How and why? "i. What AI workflow resources (e.g., exploration, seeking alternatives) support and refine this intent?
ii. Does the AI itself contribute to shaping the final intent by filling in gaps? How does the back-and-forth interaction with the AI shape the final output?
iii. How does the iterative and conversational nature of vibe coding affect users' creativity and problem-solving strategies?
iv. Does this process encourage more exploration or lead to a reliance on the AI's suggestions?"
Workflow a. Stages of workflow? "i. What are the stages of the vibe coding workflow?
ii. How is time and effort allocated across different stages (e.g., programming, fault localization, debugging versus planning and testing)?
iii. How do these stages compare with those in more traditional AIenhanced programming workflows and prior identified AI workflows such as iterative goal satisfaction?"
b. Portfolio of tools and technologies? "i. How do different tools work together in the context of vibe coding?
ii. How are other resources (such as documentation, web searches, etc.) integrated into the vibe coding workflow?"
Prompting a. Any setup work? i. How do creators set up a vibe coding session (e.g., editing prompt instructions vs. using default settings)?
b. Explicit prompting strategies? "i. What are the prompting strategies in vibe coding?
ii. What methods are employed to initiate and guide the AI during the coding process?"| Top Level Category | Subcategories | Elaboration |
| :--- | :--- | :--- |
| Goals | a. What do they build? | i. What kind of apps do people build using vibe coding? <br> ii. What types of applications are created (e.g., landing pages, games, etc.)? <br> iii. Is vibe coding fully appropriate to complete a project, or does it require transitions to manual coding for more complex projects like games? |
| | b. Expectation - for full/partial success, or just exploration? | i. Do people go into vibe coding expecting to fully succeed, or are they taking an exploratory attitude? <br> ii. Are users generally expecting complete success with vibe coding, or do they experiment to see how far the technology can be pushed? |
| Intentions | a. Initial intention? | i. How is the intention for an app formed and refined in vibe coding? <br> ii. Is the intent pre-formed before coding begins, or does it evolve through iterative interactions with the AI? <br> iii. Which parts of the intent are established beforehand, and how does intent spread across various prompts and phases? |
| | b. Intention refined? How and why? | i. What AI workflow resources (e.g., exploration, seeking alternatives) support and refine this intent? <br> ii. Does the AI itself contribute to shaping the final intent by filling in gaps? How does the back-and-forth interaction with the AI shape the final output? <br> iii. How does the iterative and conversational nature of vibe coding affect users' creativity and problem-solving strategies? <br> iv. Does this process encourage more exploration or lead to a reliance on the AI's suggestions? |
| Workflow | a. Stages of workflow? | i. What are the stages of the vibe coding workflow? <br> ii. How is time and effort allocated across different stages (e.g., programming, fault localization, debugging versus planning and testing)? <br> iii. How do these stages compare with those in more traditional AIenhanced programming workflows and prior identified AI workflows such as iterative goal satisfaction? |
| | b. Portfolio of tools and technologies? | i. How do different tools work together in the context of vibe coding? <br> ii. How are other resources (such as documentation, web searches, etc.) integrated into the vibe coding workflow? |
| Prompting | a. Any setup work? | i. How do creators set up a vibe coding session (e.g., editing prompt instructions vs. using default settings)? |
| | b. Explicit prompting strategies? | i. What are the prompting strategies in vibe coding? <br> ii. What methods are employed to initiate and guide the AI during the coding process? |
顶层类别
子类别
详细说明
c. 颗粒度高/低级别?
i. 提示和迭代的颗粒度是多少? ii. 在氛围编码过程中,迭代步骤的细化程度如何? iii. 通过自然语言提示传达的美学意图如何被生成式人工智能转化(或误解)为技术规范和代码?
i. What is the granularity of prompts and iterations?
ii. How fine-grained are the iterative steps in the vibe coding process?
iii. How is aesthetic intent communicated through natural language prompts translated (or misinterpreted) by generative AI into technical specifications and code?| i. What is the granularity of prompts and iterations? |
| :--- |
| ii. How fine-grained are the iterative steps in the vibe coding process? |
| iii. How is aesthetic intent communicated through natural language prompts translated (or misinterpreted) by generative AI into technical specifications and code? |
d. 单目标/多目标提示?
i. 用户是同时处理多个问题,还是一次只处理一个问题?
e. 模式:语音/文本/图像?
i. 不同模式(语音、文本、截图等)在情绪编码中扮演什么角色? ii. 创作者如何将各种输入模式融入情绪编码过程? iii. 代码复用是如何管理的,例如通过借用在线示例来引导 AI?
i. What is the role of different modes (voice, text, screenshot, etc.) in vibe coding?
ii. How do creators incorporate various input modes into the vibe coding process?
iii. How is code reuse managed, for instance by borrowing online examples to steer the AI?| i. What is the role of different modes (voice, text, screenshot, etc.) in vibe coding? |
| :--- |
| ii. How do creators incorporate various input modes into the vibe coding process? |
| iii. How is code reuse managed, for instance by borrowing online examples to steer the AI? |
调试
a. 针对目标程序,他们采用了哪些调试策略?
i. 在 vibe 编码中采用了哪些调试策略? ii. 创作者如何调试通过 vibe 编码生成的代码?
i. What debugging strategies are employed in vibe coding?
ii. How do creators approach debugging code generated through vibe coding?| i. What debugging strategies are employed in vibe coding? |
| :--- |
| ii. How do creators approach debugging code generated through vibe coding? |
挑战
a. 实现目标时除了代码错误之外还面临哪些挑战?
i. 在 vibe 编码过程中遇到了哪些挑战和障碍? ii. 在整个过程中出现了哪些技术、概念或工作流程相关的障碍?
i. What challenges and barriers are encountered during vibe coding?
ii. What technical, conceptual, or workflow-related obstacles arise throughout the process?| i. What challenges and barriers are encountered during vibe coding? |
| :--- |
| ii. What technical, conceptual, or workflow-related obstacles arise throughout the process? |
b. 如何应对?
i. 采用了哪些策略来应对这些挑战?
专业知识
a. 何时使用专业知识?
i. 专家如何在氛围编码中运用他们的专业知识? ii. 专家以何种方式运用他们的专业知识?
i. How do experts deploy their expertise in vibe coding?
ii. In what ways do experts deploy their expertise?| i. How do experts deploy their expertise in vibe coding? |
| :--- |
| ii. In what ways do experts deploy their expertise? |
b. 使用了哪些类型的专业知识?
i. 相关的专业知识是否仅限于开发者知识,还是包括对系统需求、端到端设计以及更广泛技术生态系统的理解? ii. 在这种情况下,技术前瞻性的态度是否可能与传统编码技能同样重要?
i. Is the relevant expertise strictly developer knowledge, or does it include understanding system requirements, end-to-end design, and the broader technical ecosystem?
ii. Could a tech-forward attitude be as significant as traditional coding skills in this context?| i. Is the relevant expertise strictly developer knowledge, or does it include understanding system requirements, end-to-end design, and the broader technical ecosystem? |
| :--- |
| ii. Could a tech-forward attitude be as significant as traditional coding skills in this context? |
c. 何时切换到手动操作?
i. 从何时起,实践者会从氛围编码转向手动编辑(代码、文本、图像、工件)?
ii. 在工作流程的哪个阶段适合切换到传统的手动编辑?
i. When do practitioners transition from vibe coding to manual editing (code, text, image, artifact)?
ii. At what point in the workflow is it appropriate to jump into traditional manual editing?| i. When do practitioners transition from vibe coding to manual editing (code, text, image, artifact)? |
| :--- |
| ii. At what point in the workflow is it appropriate to jump into traditional manual editing? |
信任
a. 对信任、过度信任、依赖有何反思?
i. 氛围编码中的信任本质是什么?
ii. 用户如何建立对 AI 生成结果的信任,以及过度信任或过度依赖该技术存在哪些风险?
i. What is the nature of trust in vibe coding?
ii. How do users develop trust in AI-generated outputs, and what are the risks of overtrust or overreliance on the technology?| i. What is the nature of trust in vibe coding? |
| :--- |
| ii. How do users develop trust in AI-generated outputs, and what are the risks of overtrust or overreliance on the technology? |
vibe 编码的定义与表现
a. 程序员如何定义氛围编码?
i. 何时一项活动被视为 vibe 编码,而不仅仅是使用 AI 工具? ii. vibe 编码是否不仅包括代码生成,还包括调试、规划和测试? iii. vibe 编码在 AI 辅助编程活动的范围中处于何种位置?
i. When is an activity considered vibe coding versus simply using AI tools?
ii. Does vibe coding include not only code generation but also debugging, planning, and testing?
iii. Where does vibe coding fit on the spectrum of AI-assisted programming activities?| i. When is an activity considered vibe coding versus simply using AI tools? |
| :--- |
| ii. Does vibe coding include not only code generation but also debugging, planning, and testing? |
| iii. Where does vibe coding fit on the spectrum of AI-assisted programming activities? |
b. YouTube/Twitch 的表现如何影响生产?
i. 氛围编码的表演性方面(例如,YouTube 演示)如何影响生成式人工智能在编程中的实践和认知? ii. 公开演示可能如何影响工具选择、提示策略以及用户体验的构建?
i. How does the performative aspect of vibe coding (e.g., YouTube demonstrations) influence the practice and perception of generative AI in programming?
ii. How might public demonstrations affect tool choice, prompting strategies, and the framing of users' experiences?| i. How does the performative aspect of vibe coding (e.g., YouTube demonstrations) influence the practice and perception of generative AI in programming? |
| :--- |
| ii. How might public demonstrations affect tool choice, prompting strategies, and the framing of users' experiences? |
Top Level Category Subcategories Elaboration
c. Granularity high/low level? "i. What is the granularity of prompts and iterations?
ii. How fine-grained are the iterative steps in the vibe coding process?
iii. How is aesthetic intent communicated through natural language prompts translated (or misinterpreted) by generative AI into technical specifications and code?"
d. Single/multiobjective prompts? i. Do users address multiple issues at once or handle one problem at a time?
e. Modes voice/text/images? "i. What is the role of different modes (voice, text, screenshot, etc.) in vibe coding?
ii. How do creators incorporate various input modes into the vibe coding process?
iii. How is code reuse managed, for instance by borrowing online examples to steer the AI?"
Debugging a. What debugging strategies for their target program? "i. What debugging strategies are employed in vibe coding?
ii. How do creators approach debugging code generated through vibe coding?"
Challenges a. What challenges (besides code bugs) in achieving goals? "i. What challenges and barriers are encountered during vibe coding?
ii. What technical, conceptual, or workflow-related obstacles arise throughout the process?"
b. How addressed? i. What strategies are adopted to address these challenges?
Expertise a. When is expertise used? "i. How do experts deploy their expertise in vibe coding?
ii. In what ways do experts deploy their expertise?"
b. What kinds of expertise used? "i. Is the relevant expertise strictly developer knowledge, or does it include understanding system requirements, end-to-end design, and the broader technical ecosystem?
ii. Could a tech-forward attitude be as significant as traditional coding skills in this context?"
c. When switch to manual work? "i. When do practitioners transition from vibe coding to manual editing (code, text, image, artifact)?
ii. At what point in the workflow is it appropriate to jump into traditional manual editing?"
Trust a. Any reflections on trust, overtrust, reliance? "i. What is the nature of trust in vibe coding?
ii. How do users develop trust in AI-generated outputs, and what are the risks of overtrust or overreliance on the technology?"
Definition of vibe coding and performance a. How programmers define vibe coding? "i. When is an activity considered vibe coding versus simply using AI tools?
ii. Does vibe coding include not only code generation but also debugging, planning, and testing?
iii. Where does vibe coding fit on the spectrum of AI-assisted programming activities?"
b. How performance for YouTube/Twitch skews production? "i. How does the performative aspect of vibe coding (e.g., YouTube demonstrations) influence the practice and perception of generative AI in programming?
ii. How might public demonstrations affect tool choice, prompting strategies, and the framing of users' experiences?"| Top Level Category | Subcategories | Elaboration |
| :--- | :--- | :--- |
| | c. Granularity high/low level? | i. What is the granularity of prompts and iterations? <br> ii. How fine-grained are the iterative steps in the vibe coding process? <br> iii. How is aesthetic intent communicated through natural language prompts translated (or misinterpreted) by generative AI into technical specifications and code? |
| | d. Single/multiobjective prompts? | i. Do users address multiple issues at once or handle one problem at a time? |
| | e. Modes voice/text/images? | i. What is the role of different modes (voice, text, screenshot, etc.) in vibe coding? <br> ii. How do creators incorporate various input modes into the vibe coding process? <br> iii. How is code reuse managed, for instance by borrowing online examples to steer the AI? |
| Debugging | a. What debugging strategies for their target program? | i. What debugging strategies are employed in vibe coding? <br> ii. How do creators approach debugging code generated through vibe coding? |
| Challenges | a. What challenges (besides code bugs) in achieving goals? | i. What challenges and barriers are encountered during vibe coding? <br> ii. What technical, conceptual, or workflow-related obstacles arise throughout the process? |
| | b. How addressed? | i. What strategies are adopted to address these challenges? |
| Expertise | a. When is expertise used? | i. How do experts deploy their expertise in vibe coding? <br> ii. In what ways do experts deploy their expertise? |
| | b. What kinds of expertise used? | i. Is the relevant expertise strictly developer knowledge, or does it include understanding system requirements, end-to-end design, and the broader technical ecosystem? <br> ii. Could a tech-forward attitude be as significant as traditional coding skills in this context? |
| | c. When switch to manual work? | i. When do practitioners transition from vibe coding to manual editing (code, text, image, artifact)? <br> ii. At what point in the workflow is it appropriate to jump into traditional manual editing? |
| Trust | a. Any reflections on trust, overtrust, reliance? | i. What is the nature of trust in vibe coding? <br> ii. How do users develop trust in AI-generated outputs, and what are the risks of overtrust or overreliance on the technology? |
| Definition of vibe coding and performance | a. How programmers define vibe coding? | i. When is an activity considered vibe coding versus simply using AI tools? <br> ii. Does vibe coding include not only code generation but also debugging, planning, and testing? <br> iii. Where does vibe coding fit on the spectrum of AI-assisted programming activities? |
| | b. How performance for YouTube/Twitch skews production? | i. How does the performative aspect of vibe coding (e.g., YouTube demonstrations) influence the practice and perception of generative AI in programming? <br> ii. How might public demonstrations affect tool choice, prompting strategies, and the framing of users' experiences? |
在 YT22 中,我们看到一个协商过程,创作者积极评估 AI 的建议,删除了一些 AI 生成的功能,比如完成率框,接受了其他功能,同时识别出缺失的功能,如登录按钮。类似地,在 YT21 中,测试应用程序揭示了意想不到的需求,比如在测试过程中发现需要一个停止按钮,这直接源于体验应用程序的行为并识别出一个最初计划中未包含的可用性问题。
AI 也在 vibe 编码过程之外被咨询,用于指导和迭代意图,比如征求功能创意(TW1)。还观察到其他传统的意图细化来源,如将已有作品和替代商业产品作为设计参考(YT21)。
我们观察到对 AI 工作上下文的刻意管理。当转向制作动画箭头时,YT15 打开了一个新的“作曲家”窗口(本质上是 Cursor 内的新聊天线程)以创建一个新的上下文。类似地,在开始新阶段之前,YT22 关闭了所有标签页以“清除 AI 的上下文”。
YT21 开发者经常通过代码片段提供上下文,并通过函数名、变量名和 API 名等具体代码元素来精确细致地引导模型。他们还使用编号列表来构建复杂指令,并明确说明期望的结果和约束条件。此外,他们提供具体示例,例如粘贴 API 预期的 JSON 响应示例以指导实现。该开发者还在新提示中明确引用模型的先前输出,以提供纠正性反馈。在我们的数据集中,该开发者独特地在代码编辑器内使用内联提示进行局部修改(即典型于早期代码助手的较小范围、自动补全式交互),以实现有针对性的调整。
YT15 中的创作者使用了精确和不精确的语言来传达美学。在请求箭头重新定位时,创作者给出了具体指示,如“不要使用斜线箭头”和“箭头头部应指向盒子中间”。然而,创作者也使用了不精确的描述词,比如将空间“显著”扩大,依赖于 AI 的理解而非提供确切的尺寸。创作者还发出了一个高级提示,提到了主题和颜色,但未具体说明细节,将这些留给模型的“解读”。同样,YT22 中的程序员在样式设计上也使用了具体与模糊指令的混合,比如“将日期移到下方并使用非常小的字体”和“不要使用粗体,缩小对话间距”。这些提示结合了具体操作和带有主观色彩的描述词,如“非常小”和“缩小”,在保持方向控制的同时,留给模型一定的解读空间。
YT22 有意修改高级提示,聚焦于更小的部分(“只做第一阶段”),以减少幻觉现象,揭示了提示粒度如何影响 AI 性能的心理模型。他们的调试提示始终是单一目标(粘贴具体错误并请求修复),即使是功能请求也集中在特定元素上,比如“制作一个包含所有通过优化的简历的页面”,而不是同时请求多个功能,体现了对可控范围和顺序解决问题的偏好。YT22 还揭示了极细粒度提示可能被认为效率低下的情况;创作者经常进行手动编辑,而不是提示进行微小调整。
尽管有 AI 辅助,创作者仍然进行手动代码审查和编辑。在 YT21 中,程序员经常扫描并阅读 AI 生成的代码,以评估其是否符合预期并在测试前识别潜在问题。同样,在 YT22 中,创作者频繁检查生成的代码,滚动浏览文件并查看差异,以理解 AI 输出的逻辑。在 TW1 中,创作者检查 AI 生成的代码以确认其可行性,在某些情况下,如果输出“接近”预期,他们会接受代码并打算手动修复。
程序员会关注特定的函数调用和 LLM 写的代码注释(例如 YT21)。他们寻找任务成功的关键指标,比如确保调用了正确的 API 或使用了正确的标识符(YT21),如果在快速扫描中某些地方引起注意,会更详细地检查代码行。尤其是对于结构化代码,如 HTML 或 React 组件,
程序员可以快速评估整体结构。
有经验的程序员可以通过观察 AI 生成的代码,直观地感知其是否使用了合适的抽象层次。当看到 AI 为图表编写低级的 HTML/JS/CSS 原语时,程序员立刻认出这种模式,感觉像是在“重新发明轮子”,并推测“应该有一些库能做这些事情”(YT15)。
超越 AI 即时建议,理解其在整个项目中的影响,是专业知识的重要应用。专业知识使程序员能够将 AI 生成的代码与代码库的其他部分或外部文档进行交叉核对。他们可能会检查某个特定文件,以确保 AI 使用了正确的 API 路径(YT21),或者将 AI 使用的参数名称与不同文件中的函数进行交叉核对(YT21,YT15)。
专家能够通过代码在脑海中形象化最终产品,而无需运行应用程序进行测试,这也支持了对 AI 生成代码的快速扫描以进行验证。例如,YT21 反复查看 HTML,以确保生成的代码出现在正确的部分,仅通过观察代码结构就能立即发现数据将显示的位置或标题将显示的位置。
此外,快速扫描的过程不仅仅是让程序员能够对当前轮次 AI 生成的编辑中的潜在错误做出反应。在阅读 AI 生成的代码时,程序员还表现出一种主动的态度。程序员有意识地为审查后续周期的 AI 输出做准备。通过熟悉生成代码的整体结构,注意关键标识符、组件以及模型所做的实现决策,他们为评估未来的输出和差异做好准备,保持代码库随着增长仍具备“快速浏览性”。他们可能会参考 API 文档或组件结构,以了解模型在某些区域应生成的内容(YT21,YT22)。
理解 AI 的能力和局限性是另一个挑战。在 YT15 中,创作者对某个特定的 AI 模型(Claude 3.5 Sonnet)是否能按预期制作动画表示不确定,后来在体验不佳后切换到了另一个模型(o3 mini)。在 TW1 中,创作者觉得某些模型在特定工作流程中表现更好,表明将合适的 AI 能力匹配到合适的任务需要对模型有经验和了解。TW1 还表示,“LLMs 会帮你完成大部分工作,但不会帮你完成全部”,认为 AI 能力存在普遍的内在局限性。
AI 幻觉和未能完全遵守提示指令造成了障碍。在 YT21 中,模型有时生成带有不存在属性的代码,开发者需要通过错误信息和代码检查来识别这些问题。TW1 中的开发者表达了挫败感,
专家知识同样指导问题识别和解决方案开发。在 YT22 中,遇到空白屏幕时,创作者迅速检查浏览器控制台并识别出“trpc 错误”,展示了读取和解读错误信息的专业能力。在 YT21 中,程序员通过立即识别错误信息中的数据库问题,展示了调试技术的专业知识,有效利用浏览器开发者工具诊断问题,形成关于根本原因的假设,并插入打印语句追踪代码执行。在 YT21 中,程序员展示了对 API 端点、数据结构及应用不同部分如何通信的理解,使其能够识别 API 调用中的问题。同样,在 YT22 中,创作者发现应用试图连接 PostgreSQL 而非 Supabase,并准确知道如何手动编辑环境文件以纠正配置。在 YT22 中,创作者通过将 AI 生成的 React 组件与文档进行对比,评估代码是否符合预期标准,展示并应用了其对 UI 库的知识。
专业知识指导质量控制和特征对齐。在 YT22 中,创作者识别并决定移除那些虽然没有错误或干扰,但与预期设计不符的 AI 生成特征。在 TW1 中,创作者展示了对生成代码何时会增加技术债务从而“降低开发者效率”的理解,并主动解决这些问题。
最后,专家还运用他们的知识来判断何时在 AI 指导和手动编辑之间切换(将在第 3.7.3 节中详细讨论)。
因此,我们观察到专业知识并未被取代,而是被重新引导:从直接编写代码转向评估、指导和完善 AI 生成的解决方案。专家更多地扮演导演、审阅者和编辑的角色,而非逐行编写代码,但他们的技术知识在整个开发过程中依然至关重要。
3.7.2. 在 vibe 编码中使用了哪些类型的专业知识?
如前所述,我们研究的所有视频中的程序员都显然具备软件开发方面的专业知识。YT21 利用了一系列专业技能,包括对数据库和编程语言的具体技术知识,但同样重要的是他们的调试能力、对 API 调用的理解以及软件架构概念。类似地,YT22 展示了对其技术栈(Next.js、TypeScript 等)的技术知识。他们根据项目需求做出明智的库选择决策,并能识别 AI 建议何时符合或偏离需求。
除了“传统”的编程专长(代码质量和可维护性、故障定位和调试)之外,我们研究的程序员还表现出两个额外领域的专长:人工智能专长(理解模型、提示实践、上下文窗口等概念,以及 AI 代码生成的局限性),以及产品管理专长(将目标形成并转化为产品功能)。或许由于原型设计与生产的融合(如第 3.2.2 节所述),vibe 编码需要(或至少在很大程度上依赖于)跨越技术实现、工具理解和产品愿景的多方面技能组合。
在这些视频中,我们看到创作者将传统编码知识与对新工具和工作流程的适应相结合。有效的 vibe 编码者似乎能够在直接编写/编辑代码、为 AI 工具制定有效提示、评估 AI 输出以及理解更广泛的产品和用户需求之间自如切换,这表明技术基础和“技术前瞻”思维模式都是该情境下专长的重要组成部分。
此外,我们推测这些程序员也在培养一种可以称为环境能力的感觉。这是一种能力感,程序员觉得自己能够处理以前不会考虑的任务,因为他们可以使用一个 AI 系统,该系统可能只需投入很少的努力就能完成任务,无论他们自己是否具备相关能力。
开发者在多种不同情况下以及出于多种原因,从 AI 辅助的 vibe 编码转向手动操作。他们可能出于效率、调试、优化以及在模型难以“引导”的特定环节选择手动工作。vibe 编码过程中转向手动操作不仅是技术需求的体现,也反映了个人偏好、专业水平以及对人机交互的期望。
对于那些提示开销超过直接修改所需努力的简单编辑,开发者通常选择手动操作。在 YT21 中,程序员尝试用 AI 进行内联编辑,但在确定 AI 辅助对于如此小的改动并不必要后,决定手动完成这行简单的修改。在 TW1 中,开发者阐述了她在何时选择手动工作与何时使用 AI 辅助的决策过程。在一个例子中,她在认为手动工作更高效时切换到手动操作,并指出“IntelliJ 的自动补全和搜索仍然比 LLMs 更快”。这促使她“与 AI 竞速”寻找代码位置,有时在找到所需内容更快时会中断 AI 生成过程。
这种细化同样体现在项目计划和文档中。在 YT22 中,开发者手动纠正了 AI 生成文档中的错误(将“Next.js 14”改为“Next.js 15”),并删除了实施计划中的冗余步骤。
调试是多个视频中手动干预的常见触发因素。在 YT21 中,程序员在遇到错误时几乎立即转为手动操作,使用传统的调试技术,如浏览器开发者工具、控制台检查和假设形成。在一个实例中,他们基于错误信息提出了解决方案,并在没有 AI 帮助的情况下实施了修复(YT21)。YT22 显示了类似的模式,开发者手动编辑环境文件以修复数据库连接错误,并直接干预以纠正运行时问题。这些例子表明,当出现错误时,开发者通常依赖自身专业知识和传统调试方法,而非 AI 工具。
然而,YT15 展现了一个相反的案例,开发者明确表示在遇到错误时“不会自己尝试修复”,并引用了遵循 Karpathy 准则。因此,这些视频确实表现出在手动操作方法上的个体差异。YT15 的创作者有意避免手动调试,几乎完全依赖在整个工作流程中通过重新提示来进行修改。相比之下,YT21 的程序员则经常进行手动代码审查,并根据对正确性和项目需求的理解拒绝 AI 的建议。
在 YT15 中,我们看到一位开发者通过对 AI 生成的代码进行最小程度的检查就接受它,从而表现出初步的信任。随着后续成功的输出,信任进一步发展。当模型成功生成复杂动画时,创作者表现出惊讶(“如果这真的有效,那感觉有点疯狂”),随后在成功时强化了信任(“哇,[…] 这正是我所要求的”)。这表明显著的积极结果可以显著增强对 AI 能力的信任。
YT21 向实践者展示了一种看似校准良好的信任心态。这位开发者表达了他可以“快速通过 AI 完成 80-90%80-90 \% 的工作进度”。然而,这种信任是有条件的,正如他明确表示的那样,“我喜欢审查更改,因为这在某种程度上帮助我保持控制权。”过度信任的风险被积极地减轻,因为开发者经常拒绝使用不存在的 API 或他认为维护性较差的方法的 AI 建议。所有参与者都表现出频繁(尽管是轻量级的)代码审查。因此,审查更改对于氛围编码来说似乎很重要,不仅是为了保持对代码的理解,也为了保持主动权、作者所有权和信任。
用户通过实验、验证和适应的过程建立对 AI 生成输出的信任。当用户未能批判性地评估 AI 输出或缺乏识别有问题代码的专业知识时,过度信任的风险就会出现。在这些视频中,我们看到开发者大多数情况下似乎成功避免了过度信任。他们对 AI 输出保持批判态度,通过测试和审查验证结果,并具备足够的专业知识来识别有问题的代码。
3.9. 定义和执行氛围编码
3.9.1. 程序员如何定义氛围编码?
因此,Vibe 编码表现为一种编程方法,其特点是与 AI 工具进行对话式、迭代式的交互,模型承担了编码工作的主要部分。从 YT15 中,我们看到 vibe 编码通过口语化、相对随意但详细的提示定义,并且模型是代码创建和完善的主要执行者。同样,在 TW1 中,一位程序员直接将其定义为“让 AI 做繁重的工作”,并描述了工作流程,即 AI 创建一个功能,然后“把它全部润色一遍”。在 YT21 中,程序员将 vibe 编码描述为“只是和应用聊天,说做这个,做那个”。
TW1 将 vibe 编码呈现为灵活且富有创造力,类比为“一幅 Bob Ross 的画作”,在那里“你可以
基于这些观察,vibe 编码在编程活动中对 AI 依赖的光谱上似乎处于比“传统”AI 辅助编码(例如 Sarkar 等人(2022a)或 Barke 等人(2023)所记录的初代 GitHub Copilot)更高的位置。它不仅仅是使用 AI 完成离散任务或代码补全,而是将代码的创建和修改大量委托给 AI 系统。然而,这并非完全放手不管。如 YT21 所示,程序员可能依赖 AI 生成代码,但在必要时会迅速切换到手动调试和问题解决。同样,YT22 强调人类监督仍然重要,程序员坚持阅读和审查 AI 生成的代码。
评估和调试的过程也表现出差异。早期的 AI 辅助编程需要通过检查、执行、文档查阅和静态分析来进行验证(Sarkar 等,2022a;Barke 等,2023),而 vibe 编码则优先考虑快速、有针对性的检查。如第 3.5.1 节所述,程序员快速浏览代码差异,寻找熟悉的模式或关键词(眼动—
球状检查、扫视)而不是逐行审查。关键是,传统的手动调试技能在 vibe 编码中仍然至关重要。当出现错误时,vibe 编码者经常回归传统方法:分析错误信息,使用浏览器开发者工具或终端,并自行形成关于错误及其解决方案的假设。然后,模型可能被要求实现程序员确定的具体修复,尽管有时只是将原始错误信息提供给模型进行修复。这种由人类诊断和规划、AI 执行修复的工作流程,与早期研究中用户理解和修复 AI 代码的体验(Vaithilingam 等,2022)形成对比。它也不同于早期工具探索模式中更为显著的显式验证策略,如详细检查和文档查阅(Barke 等,2023)。
程序员的专业知识在两种范式中都至关重要,但在 vibe 编码中的应用有所不同。早期研究发现,有经验的用户在“加速”阶段(生成代码以实现明确意图)花费更多时间(Barke 等,2023)。在 vibe 编码中,专业知识不仅对编码本身至关重要,还对与 AI 的战略性互动以及管理更广泛的开发环境至关重要。专家运用他们的知识选择工具和模型,配置项目,快速进行视觉代码评估,当 AI 失败时手动诊断复杂的错误,并战略性地决定何时在 AI 驱动和手动工作之间切换。他们能够快速识别冗余或不需要的 AI 输出,并在多个文件或外部工具(如文档或终端输出)之间综合信息。这种专业水平不再是逐行编写代码(如早期工具可能加速的那样),而更多是协调模型和其他工具在复杂开发环境中的运作(Lee 等人(2025)称之为“任务管理”)。
Vibe 编码被视为第一代 AI 辅助编程的进化,它依托大型语言模型的对话和生成能力,同时仍然需要大量的人类专业知识和判断力。它代表了从将代码生成模型作为高级自动补全或搜索工具,向更集成且功能更强大的开发工具的转变。之前的研究已经指出了将 AI 辅助编程与通过搜索重用、编译或通过规范编程之间的类比存在的不足(Sarkar 等,2022a);而在 vibe 编码的情况下,这些类比甚至更不成立。
我们推测性地提出,vibe coding 的氛围与格式塔心理学(例如,Kaldis (2013))之间存在联系,二者都以整体感知和涌现理解为特征。vibe coding 以其快速、迭代的特性,鼓励程序员将 IDE 中 AI 生成的注释、代码和代理行为视为一个整体,依赖于持续的“氛围检查”,这与格式塔原则相对应,即对世界的感官体验是作为有组织的整体结构(而非部分)来感知的。
从 MET 的视角来看,vibe 编码环境中的 AI 工具可以被解读为一种新的“物”或中介材料。因此,程序员的认知过程不再是通过操作代码语法来实现,而是通过与 AI 界面互动、制定提示语并评估 AI 生成的材料输出(生成的代码及其行为)来完成。这种将 AI 视为“物”的解读表明,vibe 编码涉及的是物质参与对象的转变(从代码转向 AI),而非完全放弃,程序员的专业技能现在转向于驾驭这一新的物质基底。
有趣的是,YT21 中的一个例子揭示了 AI 可能如何改变程序员与代码物质性的关系的另一种复杂性。具体来说,由于程序员使用了 AI,他们似乎能够使用更“原生”的 JavaScript,而较少依赖像 React 这样的高级库,或像 TypeScript 这样抽象程度较高的语言。高级库和语言当然是针对在特定物质基底中工作的感知限制而直接创建的抽象。关于早期 AI 辅助编程的研究指出,这些抽象中体现的人类专业知识是早期模型成功应用的关键因素,这些模型一次生成的代码行数相对较少(Sarkar 等,2022a),相比之下,LLM 代理在 vibe 编码过程中进行的是跨代码库的广泛变更。
值得在此稍作停留,思考这个例子对手动编程与 AI 辅助编程的影响。对于手动程序员来说,高级库非常有价值,因为它们降低了接触代码的物质障碍,因此对他们的工作流程至关重要。对于手动程序员而言,使用原生 JavaScript 编写代码可能冗长、乏味且容易出错。而对于使用上一代 AI 代码生成工具的 AI 辅助程序员来说,高级库的表达能力极大地增强了在那些模型所能处理的小上下文和输出规模内所能实现的功能。
本研究是对 vibe 编码的首次实证分析。vibe 编码是一种新兴的编程范式,开发者主要通过自然语言提示与代码生成 AI 交互来编写和编辑代码,而非直接操作代码。通过对 YouTube 和 Twitch 上精选的边思考边讲解视频进行框架分析,我们考察了程序员在这种新型编程方式中如何设定目标、执行工作流程以及运用专业知识。
我们发现,vibe 编码是传统 AI 辅助编程的有意义演进,其特点是开发者通过提示、评估、调试和完善的循环过程,实现迭代式目标满足。该工作流程始终涉及将代码生成策略性地委托给 AI,同时通过快速评估和有针对性的干预保持人工监督。高效的 vibe 编码者采用多样且混合的提示策略,涵盖从高层次、模糊且主观的指令到详细、细致且技术性的规范。
我们的观察挑战了“vibe 编码消除了编程专业知识需求”的观点。相反,我们观察到专业知识的部署发生了重新分配。传统的编码知识被转向提示和上下文管理、快速代码评估、错误识别与解决,以及决定何时在 AI 辅助和手动干预之间切换。因此,对 AI 工具的信任是细致的、动态的且有条件的,通过反复验证建立,而非盲目接受。
这些发现可能对知识工作具有更广泛的影响。vibe 编码代表了一种物质脱离的早期表现形式,实践者通过 AI 中介来协调内容生产,而非直接操作。然而,我们的分析表明,这类工作流程
Barke, S., James, M. B., 和 Polikarpova, N.(2023)。有根的副驾驶:程序员如何与代码生成模型互动。ACM 编程语言学报,7(OOPSLA1):85111。
Drosos, I. 和 Guo, P. J.(2021)。主播教授编程、艺术与游戏:认知学徒制、偶发的可教时刻与隐性专家知识。载于 2021 年 IEEE 视觉语言与以人为本计算研讨会(VL/HCC),第 1-6 页。
Drosos, I., Sarkar, A., Xu, X., Negreanu, C., Rintel, S., 和 Tankelevitch, L. (2024)。 “就像一只会回应的橡皮鸭”:通过参与式提示研究理解生成式人工智能辅助的数据分析工作流程。载于第三届人机交互工作研讨会年会论文集,CHIWORK '24,美国纽约。计算机协会出版。
Ensmenger, N. L. (2012)。计算机男孩的崛起。《计算机历史》。麻省理工学院出版社,英国伦敦。
Goffman, E. (1956)。日常生活中的自我呈现。Doubleday,苏格兰。
Goldsmith, L. J. (2021)。在应用定性研究中使用框架分析。《定性报告》,26(6):2061-2076。
Gordon, A. D., Negreanu, C., Cambronero, J., Chakravarthy, R., Drosos, I., Fang, H., Mitra, B., Richardson, H., Sarkar, A., Simmons, S., Williams, J., 和 Zorn, B. (2024)。共审计:帮助人类复核 AI 生成内容的工具。第 14 届人机交互与编程语言交叉年会论文集(PLATEAU 2024)。
Green, T. R. (1989)。符号的认知维度。人和计算机 VV ,第 443-460 页。
Kaldis, B. (2013)。格式塔心理学。载于 Kaldis, B.(编),《哲学与社会科学百科全书》,第 2 卷,第 383-386 页。SAGE 出版公司。
Kery, M. B. 和 Myers, B. A.(2017)。探索探索性编程。载于 2017 年 IEEE 视觉语言与以人为本计算研讨会(VL/HCC),第 25-29 页。IEEE。
Knuth, D. E.(1992)。文体化编程。语言与信息研究中心出版讲义。斯坦福大学语言与信息研究中心,加利福尼亚州斯坦福。
Laugier, S.(2013)。为什么我们需要普通语言哲学。芝加哥大学出版社。
Lee, H.-P. H., Sarkar, A., Tankelevitch, L., Drosos, I., Rintel, S., Banks, R., 和 Wilson, N. (2025)。生成式人工智能对批判性思维的影响:来自知识工作者调查的自我报告认知努力减少及信心效应。《2025 年 CHI 人机交互大会论文集》,CHI '25,美国纽约。计算机协会。
Lee, M. J. L., Sarkar, A., 和 Blackwell, A. F. (2024)。使用 Copilot 进行标识符命名的可预测性:混合主动编程工具的案例研究。发表于第 35 届程序设计心理学兴趣小组年会论文集(PPIG 2024)。
Malafouris, L. (2019)。心智与物质参与。《现象学与认知科学》,18(1):1-17。
Reif, J. A., Larrick, R. P., 和 Soll, J. B. (2025)。使用人工智能的社会评价惩罚证据。《美国国家科学院院刊》,122(19):e2426766122。
Ritchie, J. 和 Spencer, L.(2002)。应用政策研究的定性数据分析。载于《定性数据分析》,第 173-194 页。Routledge 出版社。
Sarkar, A.(2023a)。计算机应该易于使用吗?质疑用户界面设计中的简洁原则。载于 2023 年 CHI 人机交互大会扩展摘要,CHI EA '23,美国纽约。计算机协会。
Sarkar, A.(2025)。《人工智能可能写了这篇文章:知识工作中阶级主义侮辱语的诞生》。载于《CHI 人机交互大会扩展摘要集》,CHI EA '25,美国纽约。计算机协会出版。
Sarkar, A., Gordon, A. D., Negreanu, C., Poelitz, C., Srinivasa Ragavan, S., 和 Zorn, B.(2022a)。《用人工智能编程是什么体验?》载于第 33 届编程心理学兴趣小组年会论文集(PPIG 2022)。
Sarkar, A., Ragavan, S. S., Williams, J., 和 Gordon, A. D.(2022b)。《终端用户在电子表格中遇到 lambda 抽象:阿波罗之弓还是阿喀琉斯之踵?》载于 2022 年 IEEE 视觉语言与以人为本计算研讨会(VL/HCC),第 1-11 页。
Schilke, O. 和 Reimann, M.(2025)。《透明度困境:人工智能披露如何侵蚀信任》。组织行为与人类决策过程,188:104405。
Srinivasa Ragavan, S., Sarkar, A., 和 Gordon, A. D. (2021)。电子表格理解:猜测、放弃与回访作者。载于 2021 年 CHI 人机交互大会论文集,CHI '21,美国纽约。计算机协会出版。
Tankelevitch, L., Kewenig, V., Simkute, A., Scott, A. E., Sarkar, A., Sellen, A., 和 Rintel, S. (2024)。生成式人工智能的元认知需求与机遇。载于 CHI 人机交互大会论文集,CHI '24,美国纽约。计算机协会出版。
Vaithilingam, P., Zhang, T., 和 Glassman, E. L. (2022)。期望与体验:评估由大型语言模型驱动的代码生成工具的可用性。载于 2022 年 CHI 人机交互大会扩展摘要,CHI EA '22,美国纽约。计算机协会出版。
Weinberg, G. M. (1971)。计算机程序设计心理学。纽约:Van Nostrand Reinhold 出版社。
吴勇,Kropczynski, J.,Shih, P. C.,和 Carroll, J. M.(2014)。探索 GitHub 及其他平台上的软件开发者生态系统。载于第 17 届 ACM 计算机支持协作工作与社会计算会议伴随出版物论文集,页 265-268。
https://www.youtube.com/ watch?v=5k2-NOh2tk0
What is "Vibe Coding"? Here's how I do it...| https://www.youtube.com/ watch?v=5k2-NOh2tk0 |
| :--- |
| What is "Vibe Coding"? Here's how I do it... |
https://www.youtube.com/ watch?v=HBwXs99LFlw
VIBE CODING 3 min demo l Cursor + o3-mini + SuperWhisper| https://www.youtube.com/ watch?v=HBwXs99LFlw |
| :--- |
| VIBE CODING 3 min demo l Cursor + o3-mini + SuperWhisper |
YT17
https://www.youtube.com/ watch?v=Nt2Lkdy3f5Y “Vibe” 编码 AI 代理为我做销售(OpenAI 代理 SDK)
https://www.youtube.com/ watch?v=Nt2Lkdy3f5Y
?Vibe? Coding AI Agent to do sales for me (OpenAI Agents SDK)| https://www.youtube.com/ watch?v=Nt2Lkdy3f5Y |
| :--- |
| ?Vibe? Coding AI Agent to do sales for me (OpenAI Agents SDK) |
2025年3月17日
00:40:04
致敏
YT18
https://www.youtube.com/ watch?v=_QOvocOFLbo
2025年3月16日
00:22:28
敏感化,发展
氛围编码 - 编码与放松
2025年3月15日
YT19
https://www.youtube.com/ watch?v=TwSYePsdfOk
00:27:28
致敏
回忆中的进行中工作:Vibe 编码加密阿尔法检测代理 #1
ID URL and video title Date published Duration Research phases
YT7 https://www.youtube.com/ watch?v=dan3QfN3CDU 06-Feb-25 00:33:47 Sensitising
Karpathy Vibe Coding Full Tutorial with Cursor (Zero Coding)
YT8 28-Feb-25 00:02:35 Sensitising
"https://www.youtube.com/ watch?v=g84CGmelvSU
Vibe Coding: Launch Your SaaS with AI (Cursor, Supabase, & Stripe)"
YT9 https://www.youtube.com/ watch?v=GwhFjhMF 60A 06-Feb-25 00:10:56 Sensitising
Vibe coding using ChatGPT
YT10 https://www.youtube.com/ watch?v=ulJrdYXLo9I 04-Mar-25 00:11:52 Sensitising
Vibe Coding a FULL Game ?? (AI Coding)
YT11 "https://www.youtube.com/ watch?v=faezjTHA5SU
Complete Guide to Cursor For NonCoders (Vibe Coding 101)"
YT12 https://www.youtube.com/ watch?v=JEU07S2WyDs 08-Mar-25 00:21:03 Sensitising, development
Vibe Coding in 2025: No Typing, No Stress, Just AI!
YT13 https://www.youtube.com/ watch?v=i0gWDz9EUgI 11-Feb-25 00:13:48 Sensitising
Vibe Coding 101 "03-Mar-25
00:10:22
Sensitising"
YT14 https://www.youtube.com/ watch?v=yGZtu_6E118
Vibe Coding in Databutton AI FREE AI Code Editor for Full stack development in Claude Sonnet 3.7
YT15 https://www.youtube.com/ watch?v=irBnUAq3MAw 17-Feb-25 00:17:45 Sensitising, development, analysis
VIBE CODING with CURSOR - Draw w/ HTML and JS
TW2 https://www.twitch.tv/ coolaj86/video/2399572024 07-Mar-25 05:57:50 Sensitising
Vibe Coding Hangout: Live w/ Grok3 Unfurl-as-a-Service:Link Unshortener
TW3 https://www.twitch.tv/ vadimnotjustdev/video/ 2402985837 11-Mar-25 02:41:35 Sensitising
Vibe Coding a React Native Game
YT16 15-Feb-25 00:03:20 Sensitising
"https://www.youtube.com/ watch?v=HBwXs99LFlw
VIBE CODING 3 min demo l Cursor + o3-mini + SuperWhisper"
YT17 "https://www.youtube.com/ watch?v=Nt2Lkdy3f5Y
?Vibe? Coding AI Agent to do sales for me (OpenAI Agents SDK)" 17-Mar-25 00:40:04 Sensitising
YT18 https://www.youtube.com/ watch?v=_QOvocOFLbo 16-Mar-25 00:22:28 Sensitising, development
Vibe Coding - Code & Chill 15-Mar-25
YT19 https://www.youtube.com/ watch?v=TwSYePsdfOk 00:27:28 Sensitising
WIP at Recall: Vibe Coding Crypto Alpha Detection Agents #1 | ID | URL and video title | Date published | Duration | Research phases |
| :--- | :--- | :--- | :--- | :--- |
| YT7 | https://www.youtube.com/ watch?v=dan3QfN3CDU | 06-Feb-25 | 00:33:47 | Sensitising |
| | Karpathy Vibe Coding Full Tutorial with Cursor (Zero Coding) | | | |
| YT8 | | 28-Feb-25 | 00:02:35 | Sensitising |
| | https://www.youtube.com/ watch?v=g84CGmelvSU <br> Vibe Coding: Launch Your SaaS with AI (Cursor, Supabase, & Stripe) | | | |
| YT9 | https://www.youtube.com/ watch?v=GwhFjhMF 60A | 06-Feb-25 | 00:10:56 | Sensitising |
| | Vibe coding using ChatGPT | | | |
| YT10 | https://www.youtube.com/ watch?v=ulJrdYXLo9I | 04-Mar-25 | 00:11:52 | Sensitising |
| | Vibe Coding a FULL Game ?? (AI Coding) | | | |
| YT11 | https://www.youtube.com/ watch?v=faezjTHA5SU <br> Complete Guide to Cursor For NonCoders (Vibe Coding 101) | | | |
| | | | | |
| YT12 | https://www.youtube.com/ watch?v=JEU07S2WyDs | 08-Mar-25 | 00:21:03 | Sensitising, development |
| | Vibe Coding in 2025: No Typing, No Stress, Just AI! | | | |
| YT13 | https://www.youtube.com/ watch?v=i0gWDz9EUgI | 11-Feb-25 | 00:13:48 | Sensitising |
| | Vibe Coding 101 | 03-Mar-25 <br> 00:10:22 <br> Sensitising | | |
| YT14 | https://www.youtube.com/ watch?v=yGZtu_6E118 | | | |
| | Vibe Coding in Databutton AI FREE AI Code Editor for Full stack development in Claude Sonnet 3.7 | | | |
| YT15 | https://www.youtube.com/ watch?v=irBnUAq3MAw | 17-Feb-25 | 00:17:45 | Sensitising, development, analysis |
| | VIBE CODING with CURSOR - Draw w/ HTML and JS | | | |
| TW2 | https://www.twitch.tv/ coolaj86/video/2399572024 | 07-Mar-25 | 05:57:50 | Sensitising |
| Vibe Coding Hangout: Live w/ Grok3 Unfurl-as-a-Service:Link Unshortener | | | | |
| TW3 | https://www.twitch.tv/ vadimnotjustdev/video/ 2402985837 | 11-Mar-25 | 02:41:35 | Sensitising |
| | Vibe Coding a React Native Game | | | |
| YT16 | | 15-Feb-25 | 00:03:20 | Sensitising |
| | https://www.youtube.com/ watch?v=HBwXs99LFlw <br> VIBE CODING 3 min demo l Cursor + o3-mini + SuperWhisper | | | |
| YT17 | https://www.youtube.com/ watch?v=Nt2Lkdy3f5Y <br> ?Vibe? Coding AI Agent to do sales for me (OpenAI Agents SDK) | 17-Mar-25 | 00:40:04 | Sensitising |
| YT18 | https://www.youtube.com/ watch?v=_QOvocOFLbo | 16-Mar-25 | 00:22:28 | Sensitising, development |
| | Vibe Coding - Code & Chill | 15-Mar-25 | | |
| YT19 | https://www.youtube.com/ watch?v=TwSYePsdfOk | | 00:27:28 | Sensitising |
| | WIP at Recall: Vibe Coding Crypto Alpha Detection Agents #1 | | | |
表 2 - 来源表(续)
ID
网址和视频标题
发布日期
持续时间
研究阶段
YT20
https://www.youtube.com/ watch?v=Pe8ghwTMFlg ?? VS Code - 代理模式升级!
https://www.youtube.com/watch?v=EAPWrpvIOjs 用 AI 进行氛围编码:观看我在这段长视频中实时修复聊天界面
https://www.youtube.com/ watch?v=EAPWrpvIOjs
Vibe Coding with AI: Watch me fix a chat interface in real-time in this long video| https://www.youtube.com/ watch?v=EAPWrpvIOjs |
| :--- |
| Vibe Coding with AI: Watch me fix a chat interface in real-time in this long video |
https://www.twitch.tv/ codesinthedark/video/ 2414451732
DAY ??x3 - Vibe coding a game using Cursor #vibejam| https://www.twitch.tv/ codesinthedark/video/ 2414451732 |
| :--- |
| DAY ??x3 - Vibe coding a game using Cursor #vibejam |
2025年3月24日
01:32:40
致敏
YT30
https://www.youtube.com/ watch?v=OZaxtm3RyCw
2025年3月30日
00:12:16
致敏
YT31
https://www.youtube.com/ watch?v=y9XEBnNvu2Q
2025年3月28日
01:12:38
致敏
YT32
2025年3月25日
00:05:19
致敏
氛围编码与作品集评审 https://www.youtube.com/watch?v=_yKDiRlToSs 非程序员的氛围编码——我用 AI 在 30 秒内构建了一个在线游戏
Vibe Coding and Portfolio Reviews
https://www.youtube.com/ watch?v=_yKDiRlToSs
Vibe Coding For Non Coders - I built an online game in 30 seconds using AI| Vibe Coding and Portfolio Reviews |
| :--- |
| https://www.youtube.com/ watch?v=_yKDiRlToSs |
| Vibe Coding For Non Coders - I built an online game in 30 seconds using AI |
ID URL and video title Date published Duration Research phases
YT20 "https://www.youtube.com/ watch?v=Pe8ghwTMFlg
?? VS Code - Agent Mode UPGRADE!" 07-Mar-25 01:32:45 Sensitising
YT21 "https://www.youtube.com/ watch?v=EAPWrpvIOjs
Vibe Coding with AI: Watch me fix a chat interface in real-time in this long video" 06-Feb-25 01:29:37 Sensitising, analysis
YT22 https://www.youtube.com/ watch?v=SNARzs6jzWY 14-Mar-25 00:32:05 Sensitising, analysis
Vibe Coding Startup Using Most Advanced AI - Phase 1
YT22b https://www.youtube.com/ watch?v=u4yqSsYGpcc
Vibe Coding Startup: Dashboard Phase 2 Part 1
YT23 https://www.youtube.com/ watch?v=DvD0b0ZezQQ 19-Mar-25 Sensitising
Vibe coding actually sucks 21-Mar-25 00:38:24
YT24 https://www.youtube.com/ watch?v=NYVaCr3T1T0 Sensitising
Vibe Coding is Actually INSANE... (Vibe Coding Tutorial for Beginners)
YT25 YT25 00:45:16 Sensitising
He makes $750 a day 'Vibe Coding' Apps (using Replit, ChatGPT, Upwork)
YT26 https://www.youtube.com/ watch?v=faPSZV5XwyI 17-Mar-25 00:21:48 Sensitising
Start Vibe Coding Like a Pro, Here's How
YT27 https://www.youtube.com/ watch?v=y3vQBBmE3Cc 20-Mar-25 00:08:31 Sensitising
YT28 AI Created My Game in 6 Hours... Vibe Coding so you don't have to
https://www.youtube.com/ watch?v=icRXv9RXhXI 27-Mar-25 01:22:39 Sensitising
Vibe Coding FULL Course + WIN MacBook Pro, PlayStation 5 ??
YT29 https://www.youtube.com/ watch?v=jTaqixu79qU 18-Mar-25 00:16:28 Sensitising
Vibe Code SaaS & Mobile Games (Grok, Bolt, Cursor, Prompts)
TW4 "https://www.twitch.tv/ codesinthedark/video/ 2414451732
DAY ??x3 - Vibe coding a game using Cursor #vibejam" 24-Mar-25 01:32:40 Sensitising
YT30 https://www.youtube.com/ watch?v=OZaxtm3RyCw 30-Mar-25 00:12:16 Sensitising
YT31 https://www.youtube.com/ watch?v=y9XEBnNvu2Q 28-Mar-25 01:12:38 Sensitising
YT32 25-Mar-25 00:05:19 Sensitising
"Vibe Coding and Portfolio Reviews
https://www.youtube.com/ watch?v=_yKDiRlToSs
Vibe Coding For Non Coders - I built an online game in 30 seconds using AI" | ID | URL and video title | Date published | Duration | Research phases |
| :--- | :--- | :--- | :--- | :--- |
| YT20 | https://www.youtube.com/ watch?v=Pe8ghwTMFlg <br> ?? VS Code - Agent Mode UPGRADE! | 07-Mar-25 | 01:32:45 | Sensitising |
| YT21 | https://www.youtube.com/ watch?v=EAPWrpvIOjs <br> Vibe Coding with AI: Watch me fix a chat interface in real-time in this long video | 06-Feb-25 | 01:29:37 | Sensitising, analysis |
| YT22 | https://www.youtube.com/ watch?v=SNARzs6jzWY | 14-Mar-25 | 00:32:05 | Sensitising, analysis |
| | Vibe Coding Startup Using Most Advanced AI - Phase 1 | | | |
| YT22b | https://www.youtube.com/ watch?v=u4yqSsYGpcc | | | |
| | Vibe Coding Startup: Dashboard Phase 2 Part 1 | | | |
| YT23 | https://www.youtube.com/ watch?v=DvD0b0ZezQQ | 19-Mar-25 | | Sensitising |
| | Vibe coding actually sucks | 21-Mar-25 | 00:38:24 | |
| YT24 | https://www.youtube.com/ watch?v=NYVaCr3T1T0 | | | Sensitising |
| Vibe Coding is Actually INSANE... (Vibe Coding Tutorial for Beginners) | | | | |
| YT25 | YT25 | | 00:45:16 | Sensitising |
| | He makes $\$ 750$ a day 'Vibe Coding' Apps (using Replit, ChatGPT, Upwork) | | | |
| YT26 | https://www.youtube.com/ watch?v=faPSZV5XwyI | 17-Mar-25 | 00:21:48 | Sensitising |
| Start Vibe Coding Like a Pro, Here's How | | | | |
| YT27 | https://www.youtube.com/ watch?v=y3vQBBmE3Cc | 20-Mar-25 | 00:08:31 | Sensitising |
| YT28 | AI Created My Game in 6 Hours... Vibe Coding so you don't have to | | | |
| | https://www.youtube.com/ watch?v=icRXv9RXhXI | 27-Mar-25 | 01:22:39 | Sensitising |
| | Vibe Coding FULL Course + WIN MacBook Pro, PlayStation 5 ?? | | | |
| YT29 | https://www.youtube.com/ watch?v=jTaqixu79qU | 18-Mar-25 | 00:16:28 | Sensitising |
| Vibe Code SaaS & Mobile Games (Grok, Bolt, Cursor, Prompts) | | | | |
| TW4 | https://www.twitch.tv/ codesinthedark/video/ 2414451732 <br> DAY ??x3 - Vibe coding a game using Cursor #vibejam | 24-Mar-25 | 01:32:40 | Sensitising |
| | | | | |
| YT30 | https://www.youtube.com/ watch?v=OZaxtm3RyCw | 30-Mar-25 | 00:12:16 | Sensitising |
| YT31 | https://www.youtube.com/ watch?v=y9XEBnNvu2Q | 28-Mar-25 | 01:12:38 | Sensitising |
| | | | | |
| YT32 | | 25-Mar-25 | 00:05:19 | Sensitising |
| | Vibe Coding and Portfolio Reviews <br> https://www.youtube.com/ watch?v=_yKDiRlToSs <br> Vibe Coding For Non Coders - I built an online game in 30 seconds using AI | | | |
https: / /www.youtube.com/ watch?v=5qwucCaHpWY
BUILDING A GAME IN 7 DAYS| https: / /www.youtube.com/ watch?v=5qwucCaHpWY |
| :--- |
| BUILDING A GAME IN 7 DAYS |
2025年3月21日
00:12:16
致敏
YT34
https://www.youtube.com/ watch?v=78jina4V7j4 用 Lovable 一击秒杀排名出租网站并直播(与 Alex 的 Vibe 编码)
https://www.youtube.com/ watch?v=78jina4V7j4
One Shotting a Rank and Rent Site With Lovable and Going Live (Vibe Coding With Alex)| https://www.youtube.com/ watch?v=78jina4V7j4 |
| :--- |
| One Shotting a Rank and Rent Site With Lovable and Going Live (Vibe Coding With Alex) |
ID URL and video title Date published Duration Research phases
YT33 "https: / /www.youtube.com/ watch?v=5qwucCaHpWY
BUILDING A GAME IN 7 DAYS" 21-Mar-25 00:12:16 Sensitising
YT34 "https://www.youtube.com/ watch?v=78jina4V7j4
One Shotting a Rank and Rent Site With Lovable and Going Live (Vibe Coding With Alex)" 02-Apr-25 00:26:16 Sensitising
YT35 "https://www.youtube.com/ watch?v=X-xiJgkqnok
Episode #503 - Vibe Coding" 01-Apr-25 00:17:33 Sensitising| ID | URL and video title | Date published | Duration | Research phases |
| :--- | :--- | :--- | :--- | :--- |
| YT33 | https: / /www.youtube.com/ watch?v=5qwucCaHpWY <br> BUILDING A GAME IN 7 DAYS | 21-Mar-25 | 00:12:16 | Sensitising |
| YT34 | https://www.youtube.com/ watch?v=78jina4V7j4 <br> One Shotting a Rank and Rent Site With Lovable and Going Live (Vibe Coding With Alex) | 02-Apr-25 | 00:26:16 | Sensitising |
| YT35 | https://www.youtube.com/ watch?v=X-xiJgkqnok <br> Episode #503 - Vibe Coding | 01-Apr-25 | 00:17:33 | Sensitising |