# **How do you hire engineers when agents write 99% of the code?** # **当智能体（agents）编写了 99% 的代码时，你该如何招聘工程师？** We found ourselves asking that question after realizing our hiring process was built for a different world. Software engineers have always done more than just write code, but the ability to write code very well was the first pre-requisite to getting hired. 在意识到我们的招聘流程是为过去的时代而设计之后，我们不禁开始思考这个问题。软件工程师的工作从来都不只是写代码，但“代码写得好”一直是被录用的首要先决条件。 But, as agents get better at implementation, the engineers who create the most leverage are the ones with product taste, architectural judgment, and the ability to direct systems of humans and agents toward the right outcome. 但是，随着智能体在代码实现方面变得越来越强大，那些能创造最大杠杆效应的工程师，是具备产品品味（product taste）、架构判断力，以及能够引导人类与智能体系统达成正确结果的人。 At Augment, we've been writing openly about[going AI-native](https://www.augmentcode.com/blog/hardest-part-about-going-ai-native) and what it actually takes. This is the next chapter of that story. 在 Augment，我们一直公开撰文探讨[走向 AI 原生（going AI-native）](https://www.augmentcode.com/blog/hardest-part-about-going-ai-native)以及它真正需要的条件。本文便是这一系列故事的下一篇章。 # The skills that matter now # 现在真正重要的技能 As engineers begin working alongside increasingly capable AI agents, the nature of the job is shifting. Less time goes into writing code itself, and more time goes into deciding what should be built, designing systems that will hold up in production, orchestrating agents, and aligning teams around clear outcomes. 随着工程师开始与能力日益增强的 AI 智能体协同工作，这份工作的本质正在发生转变。花在编写代码本身上的时间变少了，而更多的时间被用于：决定该构建什么、设计能在生产环境中稳定运行的系统、编排智能体（orchestrating agents），以及让团队围绕明确的结果达成共识。 Coding still matters. But increasingly, it’s the part machines can help with. 写代码依然重要。但这越来越成为机器可以代劳的部分。 More important now is **judgment**: the ability to choose the right problems, make sound architectural decisions, and direct both humans and agents toward meaningful outcomes. 现在更重要的是**判断力（judgment）**：选择正确问题的能力、做出合理架构决策的能力，以及引导人类和智能体共同达成有意义结果的能力。 If you zoom out, the shift looks something like this: 宏观来看，这种转变大致如下： | Traditional Engineering
传统软件工程 | AI-Native Engineering
AI 原生软件工程 | | :--- | :--- | | **Writing code**
编写代码 | Specifying intent and evaluating tradeoffs
明确意图与评估权衡 | | **Implementing solutions**
实现解决方案 | Orchestrating agents
编排智能体 | | **Solving problems**
解决问题 | Choosing the right problems
选择正确的问题 | | **Individual output**
个人产出 | System-level outcomes
系统级成果 | Recognizing that the craft of engineering is moving up the abstraction stack led us to a simple question: 认识到软件工程这门手艺正在向更高的抽象层级移动，这让我们想到了一个简单的问题： > In an AI-native environment, what capabilities actually separate exceptional engineers from good ones? > 在 AI 原生环境中，究竟是什么能力将杰出的工程师与普通优秀的工程师区分开来？ # 6 dimensions of AI-native engineering # AI 原生工程的 6 个维度 It’s helpful to ground in[Augment’s definition of AI-native engineering](https://www.augmentcode.com/blog/hardest-part-about-going-ai-native): 回顾一下 [Augment 对 AI 原生工程的定义](https://www.augmentcode.com/blog/hardest-part-about-going-ai-native)会很有帮助： > ***The human role is shifting from author to architect and editor. You're defining the intent, making design and trade-off decisions, setting guardrails, obsessing over user experience, and being the last line of quality.*** > ***人类的角色正在从“作者（author）”向“架构师和编辑（architect and editor）”转变。你需要定义意图、做出设计和权衡决策、设定护栏、执着于用户体验，并成为质量的最后一道防线。*** With that in mind, we took half a day with a cross functional team of EM’s, IC’s and recruiters to go deep on this topic and think about this from first principles. From this discussion, six capabilities consistently surfaced. These are the dimensions we believe will matter most as engineering becomes increasingly AI-native. 带着这个概念，我们花了大半天时间，与由研发经理（EM）、独立贡献者（IC）和招聘人员组成的跨职能团队一起深入探讨了这个话题，并从第一性原理出发进行了思考。通过讨论，有六项能力反复出现。我们认为，随着软件工程日益走向 AI 原生，这些将是最重要的维度。 | Dimension
维度 | Core Question
核心问题 | | :--- | :--- | | **Product & Outcome Taste**
产品与结果品味 | Are we building the right thing?
我们在构建正确的东西吗？ | | **System & Architectural Judgment**
系统与架构判断力 | Will this survive production?
这能在生产环境中存活下来吗？ | | **Agent Leverage**
智能体杠杆 | Can you turn AI into real engineering throughput?
你能将 AI 转化为真正的工程吞吐量吗？ | | **Communication & Collaboration**
沟通与协作 | Can you communicate intent clearly and collaborate across perspectives?
你能清晰传达意图并跨越不同视角进行协作吗？ | | **Ownership & Leadership**
主人翁意识与领导力 | Do you drive outcomes, not just tasks?
你是在驱动结果，还是仅仅在完成任务？ | | **Learning Velocity & Experimental Mindset**
学习速度与实验心态 | Can you evolve as fast as the tools?
你的进化速度能赶上工具的发展速度吗？ | One thing you may notice missing from this list: raw coding ability as a standalone dimension. Coding still matters, but it’s no longer the primary differentiator of engineering talent. 你可能会注意到这个列表中缺少了一项内容：作为独立维度的纯编码能力。写代码依然重要，但它不再是区分工程人才的首要标准。 ### Product & outcome taste ### 产品与结果品味 **Are we building the right thing?** **我们在构建正确的东西吗？** As code becomes cheaper to produce, the most expensive mistake is building the wrong thing. Engineers increasingly need to investigate user problems, cut through ambiguity, and define clear outcomes before implementation begins. 随着代码的生产成本变得越来越低，最昂贵的错误就变成了构建错误的东西。在开始代码实现之前，工程师越来越需要去调查用户的问题，扫除模糊地带，并定义明确的结果。 > **The highest-leverage engineer isn’t the one who writes the most code. It’s the one who ensures the team is solving the right problem.** > **最具杠杆效应的工程师不再是写代码最多的人。而是那个确保团队在解决正确问题的人。** ### System & architectural judgment ### 系统与架构判断力 **Will this survive production?** **这能在生产环境中存活下来吗？** Agents can generate working code, but they are far less reliable at judging whether the system around it is sound. Architectural judgment still requires understanding long-term tradeoffs, operational realities, and the hidden risks that emerge at scale. 智能体可以生成能跑通的代码，但在判断其周围系统是否稳健方面，它们还远不够可靠。架构判断力仍然需要人类去理解长期的权衡取舍、实际的运维情况，以及系统规模扩大时出现的潜在隐患。 > **“It works” is easy. “It will keep working in production” is much harder.** > **“能跑通”很简单。但“能在生产环境中持续跑通”则要困难得多。** ### Agent leverage ### 智能体杠杆 **Can you turn AI into real engineering throughput?** **你能将 AI 转化为真正的工程吞吐量吗？** AI-native engineers don’t simply use agents for assistance. They structure problems so agents can execute effectively, guide them when they drift, and validate the results they produce. AI 原生工程师不仅仅使用智能体作为辅助。他们会去拆解和重构问题，以便智能体能够有效执行；在智能体偏离方向时进行引导，并验证它们产出的结果。 > **Think of it as delegation — except your reports are incredibly fast and occasionally confidently wrong.** > **你可以把它看作是任务委派（delegation）——只不过你的下属不仅速度奇快，而且偶尔还会“自信地犯错”。** ### Communication & collaboration ### 沟通与协作 **Can you communicate intent clearly and collaborate across perspectives?** **你能清晰传达意图并跨越不同视角进行协作吗？** As implementation speeds up, more of the work shifts toward clarifying problems, surfacing tradeoffs, and incorporating input from different parts of the team. Engineers increasingly need to communicate clearly, listen well, and build shared understanding quickly. 随着代码实现速度的加快，越来越多的工作重点转向了澄清问题、暴露权衡取舍，以及整合团队各方的意见。工程师越来越需要清晰地表达、善于倾听，并迅速建立共识。 > **The fastest teams aren’t the ones that code the fastest — they’re the ones that [reach clarity the fastest](https://x.com/a16z/status/2031783478023635213?ref_src=twsrc%5Egoogle%7Ctwcamp%5Eserp%7Ctwgr%5Etweet).** > **跑得最快的团队不是写代码最快的团队——而是那些[最快达成清晰共识](https://x.com/a16z/status/2031783478023635213?ref_src=twsrc%5Egoogle%7Ctwcamp%5Eserp%7Ctwgr%5Etweet)的团队。** ### Ownership & leadership ### 主人翁意识与领导力 **Do you drive outcomes, not just tasks?** **你是在驱动结果，还是仅仅在完成任务？** Great engineers own outcomes end-to-end, not just their piece of the code. When something blocks progress — whether it’s slow builds, unclear workflows, or gaps between systems — they step in and fix it, even if it’s outside their immediate scope. 优秀的工程师会对结果进行端到端的负责，而不仅仅是负责他们那部分代码。当进度受阻时——无论是构建缓慢、工作流不清晰，还是系统之间存在缝隙——他们都会挺身而出解决问题，即使这超出了他们直接的职责范围。 > **Ownership means removing whatever stands between the team and the outcome.** > **主人翁意识意味着消除阻碍团队达成结果的任何障碍。** ### Learning velocity & experimental mindset ### 学习速度与实验心态 **Can you evolve as fast as the tools?** **你的进化速度能赶上工具的发展速度吗？** The tools we use today will not be the ones we use three months from now. Engineers who thrive here experiment constantly, change their workflows quickly, and drop old approaches when better ones show up. 我们今天使用的工具，绝不会是我们三个月后使用的那一套。在这个环境中如鱼得水的工程师，会不断地进行实验，迅速改变他们的工作流，并在更好的方法出现时果断摒弃旧方法。 > **Experimentation isn't a phase. It's the job now.** > **实验不再是一个阶段。它就是现在的工作本身。** ## From ideals to criteria ## 从理念到评判标准 A framework only matters if it changes how you hire. The next step was translating these dimensions into **observable signals** — behaviors we can evaluate during interviews. 只有当一个框架能真正改变你的招聘方式时，它才是有意义的。下一步就是将这些维度转化为**可观察的信号**——即我们在面试中可以评估的行为。 For example: 例如： * Can the candidate quickly clarify an ambiguous problem? * 候选人能否迅速理清一个模糊的问题？ * Do they recognize architectural risks before they appear in production? * 他们能否在架构风险出现在生产环境之前就将其识别出来？ * Can they effectively direct and validate AI-generated work? * 他们能否有效地指导并验证 AI 生成的工作？ We focused on engineering roles first, where the shift to AI-native workflows is already the most obvious. And we’ll keep moving through other disciplines to narrow on their implementation of these criteria. 我们首先将目光聚焦在研发工程师角色上，因为在这里，向 AI 原生工作流的转变已经最为明显。接下来，我们还会继续推进其他职能岗位的研究，细化他们在这些标准上的具体落实。 ## The profiles we look for now ## 我们目前寻找的人才画像 We identified four profiles that will anchor our hiring in the near term. 我们确定了四种人才画像，这将作为我们短期内的招聘锚点。 **AI-Native Systems Engineer** **AI 原生系统工程师** Strong architectural judgment and deep infrastructure instincts. Responsible for keeping the foundations sound as agents build faster on top of them. 具备强大的架构判断力和敏锐的基础设施直觉。负责在智能体于其上更快速地构建业务时，保持地基的稳固。 **AI-Native Product Engineer** **AI 原生产品工程师** Strong product taste and user empathy. Focused on defining the right problems and iterating toward outcomes that matter. 具备极佳的产品品味和用户同理心。专注于定义正确的问题，并朝着真正有价值的结果进行迭代。 **AI-Native Applied AI Engineer** **AI 原生应用 AI 工程师** Deep understanding of models and how to build effectively on top of them. Responsible for improving the capabilities of our agents and workflows. 对大模型及如何在模型之上进行高效构建有着深刻的理解。负责提升我们智能体和工作流的能力。 **AI-Native Early Professional** **AI 原生职场新人** Learning velocity above all else. Engineers who are growing up agent-first and adapt quickly as the tools and workflows change. 学习速度高于一切。这些在“智能体优先（agent-first）”的环境中成长起来的工程师，能在工具和工作流发生变化时迅速适应。 Each profile weights the six dimensions differently, and each now has an interview loop built around the signals that matter most for that role. 每种画像对这六个维度的权重各有不同，且现在每种角色都有了一套围绕该角色最核心信号构建的专属面试流程。 ## Hiring reflects your values ## 招聘反映你的价值观 One useful side effect of rethinking hiring was that it forced us to make our engineering values explicit. 重新思考招聘带来的一个有益的副作用是，它迫使我们将我们的工程价值观变得显性化。 These six dimensions aren't just shaping recruiting. They're also influencing how we think about performance, growth, and career development. If judgment, leverage, and learning velocity matter most, those capabilities should show up everywhere, not just in interviews. 这六个维度不仅重塑了我们的招聘。它们还在影响我们对绩效、成长和职业发展的看法。如果判断力、杠杆效应和学习速度是最重要的，那么这些能力就应该体现在方方面面，而不仅仅是面试中。 ## What comes next ## 下一步是什么 We're sharing this early because we expect the framework to change. The tools are changing quickly, and our view of what great AI-native engineering looks like is changing with them. 我们之所以尽早分享这些，是因为我们知道这个框架还会发生改变。工具在飞速迭代，而我们对于“什么是卓越的 AI 原生工程”的认知也在随之演变。 If you're excited by a future where small teams of engineers work alongside large teams of agents, and where the craft centers on product taste, systems judgment, and orchestration, we'd love to hear from you. 如果你对这样一个未来感到兴奋：由少数工程师组成的小团队与大规模智能体团队协同工作，工程手艺的核心转向产品品味、系统判断力和编排能力，那么我们非常期待听到你的声音。 If you're an engineering leader wrestling with the same questions, we’d be interested in how you're thinking about them. 如果你是一位正与同样问题搏斗的技术管理者，我们也很有兴趣了解你的思考。 No one has this fully figured out yet. But hiring is already changing, whether interview loops have caught up or not. 还没有人能把这件事完全想透。但不管面试流程是否已经跟上步伐，招聘的变革已然发生。