How AI coding is shrinking engineering teams to 'one-pizza' size and rewriting tech role rules / AI 编码时代：工程团队缩至“一块披萨”规模，技术角色规则正在改写
If you’ve worked in tech for the past decade, you’ve almost certainly heard of Amazon’s famous “two-pizza team” rule—any engineering team should be small enough that two large pizzas can feed everyone, usually 5 to 8 people. But a recent analysis of how AI coding tools are shifting work bottlenecks makes a pretty convincing case that this long-standing guideline is already becoming obsolete, and the “one-pizza team” of 2 to 3 engineers is quickly becoming the new ideal.
我们对这个观点其实深有共鸣：过去几年，AI 编码工具如 Claude Code、GitHub Copilot 已经把工程师写代码、改 bug 的效率提升了数倍，原本占据工程师 70% 以上工作时间的编码环节，已经不再是项目推进的瓶颈。但效率提升的红利并没有让项目整体上线速度同比变快，反而暴露了新的矛盾——产品需求文档、交互原型的输出速度，已经完全跟不上工程师的开发节奏了。
What makes this bottleneck even trickier is that AI is far less useful for product managers and designers than it is for engineers. For designers, current generative AI tends to produce work that sits squarely in the middle of the “bell curve” of its training data: it avoids obviously bad choices, but almost never comes up with truly innovative, differentiated design concepts. For product managers, AI can aggregate user data and draft reports, but it can’t replace the hours spent talking to customers, mediating cross-team conflicts, or aligning stakeholder expectations that make up the bulk of their work.
很多公司给出的解法是开始大批量招聘“产品工程师”——这不是一个新岗位，但在 AI 时代突然变得前所未有的重要。这类工程师不是只等着接产品需求的纯执行者，而是会直接参与用户调研、需求梳理、原型搭建的全流程，甚至可以自己决定一部分产品功能的优先级和实现逻辑。产品经理和设计师依然存在，但不再是需求的唯一输出端，而是变成提供方向指导、搭建通用设计模块的协作者，刚好补上了产品和设计产能不足的缺口。
Another unexpected shift we’re seeing is that the era of the “jack of all trades” full-stack engineer is slowly winding down, replaced by a new demand for deep specialists. AI can write passable, functional code for most common use cases, but it still makes careless mistakes: it often deletes critical code without context, replicates existing anti-patterns in the codebase, and can’t debug complex cross-layer issues that don’t fit standard patterns. These specialists won’t stop using AI tools, but they’ll act as gatekeepers, reviewing AI-generated code to make sure it doesn’t degrade long-term codebase health.
最后一个很有意思的观察是，技术经理的角色也在变。现在很多工具号称能用 AI 自动评估工程师绩效、追踪工作效率，但实际上 AI 永远掌握不到团队里 40% 以上的上下文信息，这类工具几乎都很难落地。不过未来不懂写代码的技术经理会越来越少：团队规模缩小之后，经理花在人员管理上的时间变少了，AI 工具也降低了他们参与编码的门槛，完全可以一边做团队协调，一边参与代码评审甚至核心模块的开发，既不会脱离一线技术，也不会耽误管理职责。当然我们也很好奇，未来 QA 等其他技术岗位还会发生什么变化？欢迎在评论区聊聊你的观察。
来源：https://www.jampa.dev/p/the-rise-of-one-pizza-engineering
