AI didn’t kill Software Engineering. It made it more important than ever.

AI 没有杀死软件工程，它反而让软件工程变得前所未有的重要。

For years, we asked the wrong question: “Will AI replace developers?” Today, we know that wasn’t the right question. The right question is: What becomes valuable when writing code is no longer expensive? This completely changes software engineering. 多年来，我们一直问错了问题：“人工智能会取代开发人员吗？”今天我们知道，那不是正确的问题。正确的问题是：当编写代码不再昂贵时，什么才具有价值？这彻底改变了软件工程。

For decades, methodologies like Waterfall, Scrum, XP, DDD, and Clean Architecture were born in a world where writing code was expensive. Documentation aged quickly because rewriting it was costly. Specifications were abandoned because implementation took weeks. Then came AI. For the first time in history, producing code became almost free. 几十年来，瀑布模型、Scrum、XP、DDD 和整洁架构等方法论诞生于一个编写代码非常昂贵的世界。文档很快就会过时，因为重写它的成本很高。规范被废弃，因为实现它们需要数周时间。然后，人工智能出现了。历史上第一次，生产代码变得几乎免费。

Value migrated. Code became cheap. Judgment did not. Today, any LLM produces hundreds of lines of code in seconds. But it doesn’t decide: which problem to solve; what business rules exist; which exceptions matter; which architectural trade-offs should remain for the next five years. These remain human responsibilities. 价值发生了转移。代码变得廉价，但判断力没有。今天，任何大语言模型（LLM）都能在几秒钟内生成数百行代码。但它无法决定：要解决什么问题；存在哪些业务规则；哪些异常情况重要；哪些架构权衡需要在未来五年内保持不变。这些仍然是人类的责任。

The bottleneck has shifted. Before, we wrote code. Now, we write decisions. Specifications. Architectures. Acceptance criteria. Reviews. The competitive advantage is no longer typing speed. It has become clarity of thought. 瓶颈已经转移。以前，我们编写代码。现在，我们编写决策、规范、架构、验收标准和评审。竞争优势不再是打字速度，而是思维的清晰度。

Why “vibe coding” doesn’t scale: Conversations are a terrible source of truth. Each prompt increases the context. Each correction adds more tokens. Each interaction forces the AI to reconstruct your intent. At some point, it stops reasoning about the system and starts reasoning about the conversation. This is the true hidden cost of “vibe coding.” 为什么“Vibe Coding”（凭感觉编程）无法扩展：对话是极其糟糕的真理来源。每个提示词都会增加上下文，每次修正都会增加更多的 Token，每次交互都迫使人工智能重新构建你的意图。在某个时刻，它不再对系统进行推理，而是开始对对话进行推理。这就是“Vibe Coding”真正隐藏的成本。

Specifications have become the center of the project. This realization gave rise to Spec Driven Development. The idea is simple: The conversation is no longer the project’s memory. The specification is. Each feature is born from a spec, evolves into a plan, transforms into tasks, and only then is it implemented. The result is predictability. Less rework. Fewer tokens. Fewer ambiguities. 规范已成为项目的核心。这一认识催生了“规范驱动开发”（Spec Driven Development）。这个想法很简单：对话不再是项目的记忆，规范才是。每个功能都源于一份规范，演变为计划，转化为任务，最后才被实现。结果是可预测性。更少的返工，更少的 Token，更少的歧义。

But specification doesn’t solve architecture. Even good specifications fail when the project is chaotic. Architecture exists to answer only one question: Where does this belong? This question gave rise to the second book: FOCUS Architecture. Instead of dozens of layers, the proposal reduces responsibilities to four well-defined pieces: View, Orchestrator, Use Cases, Repository. 但规范并不能解决架构问题。当项目混乱时，即使是好的规范也会失败。架构的存在只为回答一个问题：这属于哪里？这个问题催生了第二本书：《FOCUS Architecture》。该方案没有使用数十个层，而是将职责简化为四个定义明确的部分：视图（View）、编排器（Orchestrator）、用例（Use Cases）和仓库（Repository）。

The goal was never to create a new architecture. It was to reduce the cost of change. In the AI era, this also means reducing context, token consumption, and the risk of unexpected changes. 其目标从来不是创造一种新的架构，而是降低变更成本。在人工智能时代，这也意味着减少上下文、Token 消耗以及意外变更的风险。

Standing on the shoulders of giants: These books do not reinvent engineering. They synthesize decades of ideas from Martin Fowler, Kent Beck, Eric Evans, Bertrand Meyer, Robert C. Martin, Andrew Hunt, David Thomas, and many others, reinterpreted for a reality where AI agents write a good portion of the code. 站在巨人的肩膀上：这些书并没有重塑工程学。它们综合了 Martin Fowler、Kent Beck、Eric Evans、Bertrand Meyer、Robert C. Martin、Andrew Hunt、David Thomas 等人的数十年思想，并针对人工智能代理编写大部分代码的现实进行了重新诠释。

Two books. One goal. Spec Driven Development organizes thought. FOCUS Architecture organizes code. One answers what to build. The other answers where to build. Together, they help human teams and AI agents produce predictable, readable, and sustainable software. 两本书，一个目标。《规范驱动开发》组织思维，《FOCUS 架构》组织代码。一个回答“构建什么”，另一个回答“在哪里构建”。它们共同帮助人类团队和人工智能代理生产出可预测、可读且可持续的软件。

Final thoughts: AI made code cheap. But it made engineering even more valuable. The future does not belong to the teams that have the best AI. It belongs to the teams that think best. 结语：人工智能让代码变得廉价，但它让工程学变得更有价值。未来不属于拥有最强人工智能的团队，而属于思维最清晰的团队。