Fiona Fung shares lessons on adapting team norms as engineering bottlenecks shift in the age of AI coding.
要点 · TL;DR
扩展 Claude Code 将瓶颈从工程转向验证与协调，需要重写团队规范。 Scaling Claude Code shifts bottlenecks from engineering to verification and coordination, requiring rewritten norms.
即时规划与 AI 辅助的 PR 取代了长期路线图和白板争论。 Just-in-time planning and AI-assisted PRs replace long roadmaps and whiteboard debates.
扁平组织、经理作为个体贡献者以及持续流程审计，驱动敏捷性和吃狗粮。 Flat organizations with IC managers and continuous process auditing drive agility and dogfooding.
核心观点 · Key points
工程带宽不再是瓶颈；验证、审查和跨职能协调成为新的瓶颈。 Engineering bandwidth is no longer the bottleneck; verification, review, and cross-functional coordination are the new bottlenecks.
团队规范必须重写：规划更加即时化，设计文档被PR取代，代码审查大量由AI辅助。 Team norms must be rewritten: planning is more JIT, design docs are replaced by PRs, and code reviews are heavily assisted by AI.
代码是事实来源，文档应与代码保持同步，通常借助AI。 Code is the source of truth, and documentation should be kept in sync with code, often with AI help.
招聘应关注具有产品感的创造性建设者和深度系统专家，而非原始吞吐量。 Hiring should focus on creative builders with product sense and deep systems expertise, not raw throughput.
组织应尽可能扁平化，管理者以个人贡献者身份开始，以深度使用产品。 Organizations should stay as flat as possible, with managers starting as individual contributors to dogfood the product.
持续审计并淘汰不再起作用的旧流程；尽可能用AI自动化工作流。 Continuously audit and kill old processes that no longer serve their purpose; automate workflows with AI where possible.
反共识 · Contrarian takes
六个月的路线图太长；规划应像即时编译一样即时进行。 Six-month roadmaps are too long; planning should be just-in-time, like JIT compilation.
技术争论应通过AI生成多个PR来解决，而非白板争论。 Technical debates should be settled by generating multiple PRs with AI, not by whiteboard arguments.
代码所有权变得模糊；与其问谁做了更改，不如深入挖掘根本问题并自动化回答。 Code ownership is fuzzy; instead of asking who made a change, double-click into the underlying question and automate answers.
管理者应以个人贡献者身份开始，招聘人员认为没有管理者会接受，但这能建立信誉和敏捷性。 Managers should start as ICs, which recruiters thought no manager would accept, but it builds street cred and agility.
传统的iOS和Android团队划分可能不再合理，因为工程师借助AI跨平台灵活工作。 Traditional iOS and Android team splits may no longer make sense as engineers flex across platforms with AI.
如果每个人都在用笔记本电脑，就取消昂贵的每周评审；问会议为何存在，若无意义就取消。 Cancel expensive weekly reviews if everyone is on laptops; ask why the meeting exists and kill it if not serving purpose.
本期章节 · Chapters（共 16）
开场与简介Opening and Introduction
重写规范简介Introduction to Rewriting Norms
代码审查与入职变更Code Review and Onboarding Changes
规划与技术辩论Planning and Technical Debates
精简设计文档，强化验证Reducing Design Docs and Doubling Down on Verification