关于超个性化软件的逆向观点，本地计算机被低估，以及 Claude 需要访问你所有工具才能发挥真正作用。
A contrarian view on hyper-personalized software, the undervaluation of local computers, and how Claude needs access to all your tools to be truly useful.
要点 · TL;DR
Claude 在本地电脑上配合用户工具效果最佳，可实现超个性化自动化。 Claude works best on local computers with user tools, enabling hyper-personalized automation.
技能以 Markdown 文件形式存在，可轻松创建重复任务的个性化自动化。 Skills as markdown files allow easy creation of personalized automation for repetitive tasks.
AI 将冲击初级岗位；模拟经验可将数年学习压缩至数周。 AI will disrupt junior jobs; simulated experience can compress years of learning into weeks.
核心观点 · Key points
Claude 的有效性取决于能否访问用户使用的相同工具，最好是在本地计算机上。 Claude's effectiveness depends on access to the same tools as the user, ideally on the local computer.
技能可以轻松创建为 markdown 文件，为重复性任务实现高度个性化的自动化。 Skills are easy to create as markdown files, enabling hyper-personalized automation for repetitive tasks.
虚拟机沙箱提供安全保障，允许 Claude 安装软件而无需用户批准。 The VM sandbox provides safety and security, allowing Claude to install software without user approval.
随着模型改进，脚手架可能变得不那么重要；投资于能力和安全是关键。 Scaffolding may become less important as models improve; investing in capabilities and safety is key.
AI 工具将显著影响初级员工，社会应讨论劳动力市场的变化。 AI tools will significantly impact junior employees, and society should discuss labor market changes.
反共识 · Contrarian takes
未来不是超个性化软件；共享工具和平台对协作仍然重要。 The future is not hyper-personalized software; shared tools and platforms remain important for collaboration.
硅谷低估了本地计算机的价值；MacBook 比 iPad 更受欢迎是有原因的。 Silicon Valley undervalues the local computer; MacBooks persist over iPads for a reason.
快速构建所有候选原型比编写规格更好；执行现在成本很低。 Building all candidate prototypes quickly is better than writing specs; execution is now cheap.
AI 不应该预订航班；这不是好的展示，因为个人偏好难以编码。 AI should not book flights; it's a poor showcase because personal preferences are hard to encode.
初级工程师可以通过模拟项目加速学习，将数年经验压缩到几周内。 Junior engineers can accelerate learning via simulated projects, compressing years of experience into weeks.
本期章节 · Chapters（共 30）
0. 超个性化反论与引言Introduction and contrarian view on hyper-personalization
1. 欢迎与工作室设置Welcome and studio setup
2. 什么是Claude Cocalc？What is Claude Cocalc?
3. 用户友好 vs 高级用户感知Perception of user-friendly vs power user
4. 基于现有原语 vs 从零构建Building on existing primitives vs. starting from scratch
5. Claude Code与VM架构Architecture of Claude Code and the VM
6. 云端 vs 本地机器Cloud vs Local Machine
7. Claude Code vs Claude Co-workClaude Code vs Claude Co-work
8. 为Claude Work定义评估Defining evals for Claude Work
9. 模型智能 vs 工具使用Model intelligence vs. tool usage
10. 从N2P服务转向技能Shift from N2P services to skills
11. 用Claude Code Work处理Discord和YouTubeUsing Claude Code Work for Discord and YouTube
12. 技能作为抽象层Skills as an abstraction layer
13. 通过Cloud Co-work扩展范围Expanding scope with Cloud Co-work
14. 系统提示与建议System prompts and suggestions
15. Claude Code内置浏览器Built-in browser in Claude Code
16. 个人偏好与技能可移植性Personal preferences and skills portability
17. 对入门级工作与模拟经验的影响Impact on entry-level jobs and simulated experience
18. AGI时间线的不确定性及其重要性Uncertainty about AGI timeline and its importance
19. 用Claude进行文件组织和任务自动化Using Claude for file organization and task automation
20. Claude用于财务与报税季Claude for finance and tax season
21. 用Claude处理税务与视觉能力Using Claude for taxes and vision capabilities
22. 视觉与计算机使用改进Vision and computer use improvements
23. Claude在电脑上操作Claude acting on your computer
24. Windows和macOS上的虚拟化Virtualization on Windows and macOS
25. 开发者的风险承受能力Risk tolerance of developers
26. Claude工作的未来Future of Claude work
27. Electron vs Tauri与操作系统Web视图Electron vs. Tauri and OS Web Views
28. AI生成应用的性能与优化Performance and optimization of AI-generated apps
29. 多人模式与协同工作交互Multiplayer mode and co-work interaction