Design engineer Jeffrey Lit discusses how LLMs are shifting digital product design towards personalized, user-controlled software, moving beyond traditional SaaS models.
要点 · TL;DR
可塑软件将控制权从开发者转移到用户，支持本地定制。 Malleable software shifts control from developers to users, enabling local customization.
版本控制是人机协作创意工作的基础。 Version control is essential for human-AI collaboration in creative work.
AI 应教导并培养批判性思维，而不仅仅是自动化任务。 AI should teach and foster critical thinking, not just automate tasks.
核心观点 · Key points
可塑软件让用户能根据本地情境自定义工具，将控制权从开发者转移到用户。 Malleable software empowers users to customize tools for their local context, shifting control from developers to users.
在设计中暴露系统结构（如自行车上的数字齿轮）能让用户更深入理解并发挥创造力。 Exposing systemic structure in design, like numbered gears on a bike, enables deeper user understanding and creativity.
版本控制是人机协作的基础，允许用户对AI输出进行分支、审查和合并。 Version control is fundamental for human-AI collaboration, allowing users to branch, review, and merge AI outputs.
AI聊天是高级用户界面，其优势在于灵活性，但对新手缺乏引导。 AI chat is a power user interface; its strength lies in flexibility, but it lacks onboarding for novices.
设计师的角色正从创造成品转向设计模式和工具包供用户使用。 The role of designers is shifting from creating finished products to designing patterns and toolkits for users.
AI应被用于学习和保持对细节的掌控，而不仅仅是自动化和委派任务。 AI should be used to learn and stay connected to details, not just to automate and delegate tasks.
反共识 · Contrarian takes
可塑软件反而带来更稳定、可预测的环境，因为用户能控制变化。 Malleable software leads to more stable, predictable environments because users have control over changes.
好的默认设置只解决了70%的问题，其余需要AI教会用户如何自定义。 Good defaults are only 70% of the solution; the rest requires AI to teach users how to customize.
AI工具应优先考虑教学和学习，而不仅仅是速度；摩擦反而能推动更好的创意成果。 AI tools should prioritize teaching and learning, not just speed; friction can drive better creative outcomes.
当前应用商店模式是分享自定义工具的瓶颈；分享应像发送链接一样简单。 The current app store model is a bottleneck for sharing custom tools; sharing should be as easy as a link.
像外科医生一样编码意味着在AI辅助下保持亲自动手，而不是成为智能体的管理者。 Coding like a surgeon means staying hands-on with AI assistance, not becoming a manager of agents.
AI可以通过解释其工作来帮助用户成为更批判性的思考者，而不仅仅是替他们完成工作。 AI can help users become more critical thinkers by explaining its work, not just doing it for them.
本期章节 · Chapters（共 19）
引言与SaaS传统问题Introduction and the problem with traditional SaaS
赞助商信息Sponsor messages
Incan Switch实验：EmbarkExperiments at Incan Switch: Embark
大型旅行应用问题与Embark概念The Problem with Mega Travel Apps and the Embark Concept
入职情境的价值Value of context in onboarding
设计工程师角色Role as design engineer
暴露结构：噩梦自行车Exposing structure: The Nightmare Bicycle
通过结构实现简洁Simplicity through structure
AI聊天作为高级用户界面AI Chat as a Power User Interface
创意工作的版本控制Version Control for Creative Work
人类意图转代码的核心挑战The core challenge of translating human intent to code
可塑软件与一次性软件Malleable vs. disposable software
工具构建中的熟悉性与互操作性Familiarity and Interoperability in Tool Building
瓶颈与软件分发变革Bottlenecks and the Shift in Software Distribution
设计师在流动界面中的角色变化The Changing Role of Designers in a Fluid Interface World