The rise of one-pizza engineering teams - by Jampa Uchoa
【原文摘要】
- AI coding tools like Claude Code have reduced coding time, shifting engineering teams' bottleneck from coding to the delivery speed of product specs and wireframes, as LLMs are less effective for product managers (PMs) and designers.
- For designers, LLMs generate middle-of-the-curve ideas that avoid bad design but limit innovation, leading to generic AI product interfaces. For PMs, LLMs assist with data gathering but can't automate client communication, the most time-consuming PM task. This creates an imbalance, as teams typically have more engineers than PMs and designers.
- To address this, product engineers—software engineers taking on some PM and design responsibilities (owning roadmaps, engaging users, assembling design system building blocks without fully replacing PMs/designers)—are growing in relevance. Not all engineers will be product engineers; specialists are still needed to manage platform code, as AI-generated code risks careless changes, bug-fixing limitations, and replicating destructive patterns, requiring human review. Full-stack engineer roles may decline, with more demand for specialized back-end/front-end experts.
- Amazon's two-pizza-team rule (5-8 people) is being phased out. New ideal team size is 2-3 engineers per project, as smaller groups reduce communication overhead and align with AI's need for wider problem context. Large projects should avoid single-person assignments to prevent frustration and support skill growth.
- AI performance-tracking tools for managers fail due to lack of full context, as they only measure quantitative metrics. With smaller teams, engineering managers will spend less time on people management and more on coding, though team-critical tasks will still take priority over coding work.
- This is just the first wave of changes; progress depends more on how AI is used than better models, with AI complementing (not centering) new PM and designer tools. Unknowns remain, such as AI's impact on QA roles.
【单词表】
- bottleneck /ˈbɒtlnek/ 瓶颈，阻碍
- spec /spek/ 规格，说明书（此处为specifications缩写，四级拓展词汇）
- wireframe /ˈwaɪəfreɪm/ 线框图（专业四级拓展词汇）
- LLM /ˌel el ˈem/ 大语言模型（Large Language Model缩写，专业领域四级以上词汇）
- innovation /ˌɪnəˈveɪʃn/ 创新，革新
- generic /dʒəˈnerɪk/ 通用的，无特色的
- automate /ˈɔːtəmeɪt/ 使自动化，自动操作
- imbalance /ɪmˈbæləns/ 不平衡，失调
- relevance /ˈreləvəns/ 相关性，重要性
- roadmap /ˈrəʊdmæp/ 路线图，发展规划
- assemble /əˈsembl/ 组装，集合
- replicate /ˈreplɪkeɪt/ 复制，重复
- destructive /dɪˈstrʌktɪv/ 破坏性的，毁灭性的
- phase out /feɪz aʊt/ 逐步淘汰，逐步取消
- overhead /ˈəʊvəhed/ 开销，管理费用（此处指沟通成本）
- align /əˈlaɪn/ 使一致，使对齐
- quantitative /ˈkwɒntɪtətɪv/ 定量的，数量的
- metric /ˈmetrɪk/ 衡量标准，指标
- priority /praɪˈɒrəti/ 优先事项，优先权
- complement /ˈkɒmplɪment/ 补充，补足
- QA /ˌkjuː ˈeɪ/ 质量保证（Quality Assurance缩写，专业领域四级以上词汇）
【句子翻译】
- AI coding tools like Claude Code have reduced coding time, shifting engineering teams' bottleneck from coding to the delivery speed of product specs and wireframes, as LLMs are less effective for product managers (PMs) and designers. 像Claude Code这类AI编码工具缩短了编码时间，将工程团队的瓶颈从编码环节转移到了产品规格和线框图的交付速度上，因为大语言模型（LLM）对产品经理（PM）和设计师的作用相对有限。
- For designers, LLMs generate middle-of-the-curve ideas that avoid bad design but limit innovation, leading to generic AI product interfaces. 对于设计师而言，大语言模型只会生成中规中矩的方案，这类方案虽能规避糟糕设计，却限制了创新，最终催生千篇一律的AI产品界面。
- For PMs, LLMs assist with data gathering but can't automate client communication, the most time-consuming PM task. 对于产品经理来说，大语言模型可协助收集数据，但无法将客户沟通这一最耗时的工作自动化。
- This creates an imbalance, as teams typically have more engineers than PMs and designers. 这就造成了一种失衡状态，因为团队里工程师的人数通常远多于产品经理和设计师。
- To address this, product engineers—software engineers taking on some PM and design responsibilities (owning roadmaps, engaging users, assembling design system building blocks without fully replacing PMs/designers)—are growing in relevance. 为解决这一问题，产品工程师的重要性日益凸显，这类软件工程师会承担部分产品经理和设计师的职责（比如制定产品路线图、对接用户、组装设计系统模块，但不会完全取代产品经理或设计师）。
- Not all engineers will be product engineers; specialists are still needed to manage platform code, as AI-generated code risks careless changes, bug-fixing limitations, and replicating destructive patterns, requiring human review. 并非所有工程师都能成为产品工程师；仍需要专业人员来管理平台代码，因为AI生成的代码存在随意修改、修复漏洞能力有限以及复制破坏性模式的风险，需要人工审核。
- Full-stack engineer roles may decline, with more demand for specialized back-end/front-end experts. 全栈工程师的岗位需求可能会下降，而对专业后端/前端专家的需求会有所增加。
- Amazon's two-pizza-team rule (5-8 people) is being phased out. 亚马逊的“双披萨团队”规则（5-8人）正逐步被淘汰。
- New ideal team size is 2-3 engineers per project, as smaller groups reduce communication overhead and align with AI's need for wider problem context. 新的理想团队规模是每个项目配备2-3名工程师，因为小团队能降低沟通成本，也符合AI对更广泛问题背景的需求。
- AI performance-tracking tools for managers fail due to lack of full context, as they only measure quantitative metrics. 面向管理者的AI绩效跟踪工具因缺乏完整背景信息而效果不佳，因为它们只能衡量定量指标。
- With smaller teams, engineering managers will spend less time on people management and more on coding, though team-critical tasks will still take priority over coding work. 团队规模缩小后，工程经理在人员管理上花费的时间会减少，投入编码工作的时间会增加，但团队核心任务仍会优先于编码工作。
- This is just the first wave of changes; progress depends more on how AI is used than better models, with AI complementing (not centering) new PM and designer tools. 这只是变革的第一波浪潮；技术发展的推进更多取决于AI的应用方式，而非更先进的模型，AI将成为产品经理和设计师新工具的补充（而非核心）。
- Unknowns remain, such as AI's impact on QA roles. 仍存在诸多未知因素，比如AI对质量保证（QA）岗位的影响。
【译文】
- 像Claude Code这类AI编码工具缩短了编码时间，将工程团队的瓶颈从编码环节转移到了产品规格和线框图的交付速度上，因为大语言模型（LLM）对产品经理（PM）和设计师的作用相对有限。
- 对于设计师而言，大语言模型只会生成中规中矩的方案，这类方案虽能规避糟糕设计，却限制了创新，最终催生千篇一律的AI产品界面。对于产品经理来说，大语言模型可协助收集数据，但无法将客户沟通这一最耗时的工作自动化。这就造成了一种失衡状态，因为团队里工程师的人数通常远多于产品经理和设计师。
- 为解决这一问题，产品工程师的重要性日益凸显，这类软件工程师会承担部分产品经理和设计师的职责（比如制定产品路线图、对接用户、组装设计系统模块，但不会完全取代产品经理或设计师）。并非所有工程师都能成为产品工程师；仍需要专业人员来管理平台代码，因为AI生成的代码存在随意修改、修复漏洞能力有限以及复制破坏性模式的风险，需要人工审核。全栈工程师的岗位需求可能会下降，而对专业后端/前端专家的需求会有所增加。
- 亚马逊的“双披萨团队”规则（5-8人）正逐步被淘汰。新的理想团队规模是每个项目配备2-3名工程师，因为小团队能降低沟通成本，也符合AI对更广泛问题背景的需求。大型项目应避免单人负责的安排，以防员工产生挫败感，同时助力技能提升。
- 面向管理者的AI绩效跟踪工具因缺乏完整背景信息而效果不佳，因为它们只能衡量定量指标。团队规模缩小后，工程经理在人员管理上花费的时间会减少，投入编码工作的时间会增加，但团队核心任务仍会优先于编码工作。
- 这只是变革的第一波浪潮；技术发展的推进更多取决于AI的应用方式，而非更先进的模型，AI将成为产品经理和设计师新工具的补充（而非核心）。仍存在诸多未知因素，比如AI对质量保证（QA）岗位的影响。
文章来源：https://www.jampa.dev/p/the-rise-of-one-pizza-engineering