Revenge of the junior developer
打开互动全文版（逐段中英对照 + 图/公式 + 论文问答）→本文探讨了初级开发者如何借助 AI 工具和新兴平台，在科技行业中重新获得影响力和话语权，挑战了传统上由资深开发者主导的格局。
Opens in a new window Opens an external website Opens an external website in a new window This website utilizes technologies such as cookies to enable essential site functionality, as well as for analytics, personalization, and targeted advertising. Privacy PolicyCookie PolicyGoogle Privacy Policy Accept Deny Non-Essential Manage Preferences
在新窗口中打开 打开外部网站 在新窗口中打开外部网站
Opens in a new window Opens an external website Opens an external website in a new window
本网站使用 cookie 等技术来实现基本网站功能，以及用于分析、个性化和定向广告。隐私政策 Cookie 政策 Google 隐私政策
This website utilizes technologies such as cookies to enable essential site functionality, as well as for analytics, personalization, and targeted advertising. Privacy PolicyCookie PolicyGoogle Privacy Policy
接受 拒绝 非必要 管理偏好
Accept Deny Non-Essential Manage Preferences
提交即表示我同意 Sourcegraph 的服务条款和隐私政策。
By submitting, I agree to Sourcegraph's Terms of Service and Privacy Policy.
Steve Yegge 关于 viiiiibe 编码及其对开发者工作意义的最新文章。
The latest installment from Steve Yegge on viiiiibe coding and what that means for developer jobs.
你好，你好，你好！很高兴再次见到大家。
Hello, hello, hello! Good to see everyone again.
我现在真的得开始注意自己说的话了。有太多人在看着我。
I've really gotta start being careful what I say these days. I've got so many people watching.
总之，前几天我放了个屁，听起来像_viiiibecooode_，然后立刻有 3 个投资者来找我。我不得不告诉他们，不，那只是个屁，好让他们离我远点。
So anyway, I ripped a fart the other day that sounded like _viiiibecooode,_ and I was immediately approached by 3 investors. I had to tell them no sorry that was just a fart, just to get them off me.
发生了太多事情，我几次试图写这篇文章，但每次尝试都变得庞大而狂野，我不得不像对待老黄狗一样把它们都处理掉。这次我就在它还小的时候把它发出去。（编辑：该死。至少它从头到尾都充满动作场面。）
There's so much going on that once again I tried several times to write this post, but each attempt grew huge and rabid, and I had to put 'em all down like Old Yeller. This time I'll just ship it while it's still a pup. _(Edit: Damn. At least it's action-packed to the end.)_
_关于“vibe 编码”含义的简要说明：_ 在这篇文章中，我假设 vibe 编码会发展成熟，人们会将其用于真正的工程，而“关闭大脑”的版本则仅用于原型设计和有趣的项目。对我来说，vibe 编码只是意味着让 AI 完成工作。你选择如何密切_关注_AI 的工作完全取决于手头的问题。对于生产环境，你要关注；对于原型，你可以放松。无论哪种方式，只要你不是手动编写，就是 vibe 编码。
_Brief note about the meaning of "vibe coding":_ In this post, I assume that vibe coding will grow up and people will use it for real engineering, with the "turn your brain off" version of it sticking around just for prototyping and fun projects. For me, vibe coding just means letting the AI do the work. How closely you choose to _pay attention_ to the AI's work depends solely on the problem at hand. For production, you pay attention; for prototypes, you chill. Either way, it's vibe coding if you didn't write it by hand.
_再说明一点：_ 复仇部分发生在最后，就像电影里一样。
_One more note:_ The revenge part happens at the very end, just like in the movies.
好了！把这些行政事务处理完后，我们开始吧！
OK! With those administrative items out of the way, let's goooooo!
氛围编码是基于聊天的编码的一个异想天开的名字，你让大语言模型写代码，然后把结果反馈给它，再要求更多，如此循环往复。这与传统编码，甚至与带有代码补全的编码截然不同。
Vibe coding is a whimsical name for chat-based coding, where you ask the LLM to write code, and then you feed it the results and ask it for more, in a continuous loop. It's very different from traditional coding, or even coding with code completions.
聊天编码在编码助手中已经存在了一段时间，但一直缺乏一个响亮的号召。它终于在二月初得到了一个，当时著名的安德烈·卡帕西博士——因共同创立 OpenAI 等事迹而闻名——给聊天编码起了一个好听的名字。他称之为“氛围编码”，它几乎在一夜之间就成了一个蓝金裙式的现象。
Chat coding has been around a while in coding assistants, but without a rallying cry. It finally got one, when in early February the illustrious Dr. Andrej Karpathy, famed among other things for co-founding OpenAI, put a pretty name to the face of chat. He called it "vibe coding" and it became an instant blue/gold dress situation pretty much overnight.
今天，截至——等等，让我看看我的手表——_现在_，当你读到这段话时，氛围编码已经进入了一个奇怪、前所未有的、量子般的三重态：
Today as of, wait lemme look at my watch, _right_ _now_ as you're reading this, vibe coding has entered a strange, unprecedented, quantum-like triplet-state:
* 氛围编码对于硅谷以外 80%的行业来说仍然完全不可见，他们几乎不知道我们在说什么。许多人甚至还没听说过“氛围编码”这个词，更不用说“编码智能体”了。我猜他们从不看新闻？不幸的是，他们都有被 AI 突然袭击的风险，不，是被 AI 侧面撞击。
* Vibe coding is still completely invisible to 80%of the industry outside Silicon Valley, who will have little clue as to what we're talking about here. Many haven't even heard the phrase "vibe coding" yet, let alone "coding agent". I guess they never open the news? Unfortunately they all risk getting blindsided, nay, T-boned by AI.
* 氛围编码目前正在疯狂传播，以戏剧性的指数曲线快速增长，登上《纽约时报》等主要媒体，充斥社交媒体，有人欢呼，有人谴责。就在谷歌非正式地采用它时，许多公司却忙着禁止它。每个人仍在争论“氛围编码”到底意味着什么。但很多人——而且每天都在增加——认为它现在就是未来。
* Vibe coding is currently going batshit viral, growing like crazy on a dramatic exponential curve, hitting major media outlets like the NYT, flooding social media, celebrated by some, decried by others. A bunch of companies were busy banishing it just as Google was unofficially adopting it. Everyone's still arguing about what "vibe coding" even means. But a ton of people, more every day, think it's the future right now.
* 对于一群增长_更快_的开发者来说，_聊天_编码总体上已经彻底过时了，他们现在甚至不愿意过马路去给着火的聊天编码上浇尿。他们仍然在氛围编码，而且确实比任何人都获得更好的氛围。然而，他们再也不会关心_你的_氛围编码——那些冗长的来回聊天对话——了，再见，先生，我说，再见！
* _Chat_ coding in general is already utterly passé to an exponentially-_faster_ growing group of developers who now wouldn't walk across the street to piss on chat-based coding if it was on fire. They are still vibe coding and indeed getting better vibes than anyone. And yet they could not possibly care any less about _your_ vibe coding – those long back-and-forth chat conversations – ever again, good day to you Sir, I said Good. Day!
在夸张中心这里，我们很难编造出比这更疯狂的事情。这是真实的，但发展如此之快，以至于感觉真的超现实。
Here at Exaggeration Central, we are finding it hard to make anything up that's crazier than this. It's real, but unfolding so fast that it feels genuinely surreal.
氛围编码正在急剧上升，而基于聊天的编码——_你_认为的氛围编码，以及我曾经称之为 CHOP 的东西——确实也仍在上升……暂时如此。但智能体式编码——本文的主题——将很快像火箭一样超越聊天编码，仿佛它静止不动。
Vibe coding is in steep ascent, and chat-based coding – what _you_ think of as vibe coding, and what I used to call CHOP – is indeed also still on the rise… for now. But agentic coding – the subject of this post – will soon rocket by chat coding as if it's standing still.
到第三季度，今天的聊天编码对许多人来说将成为最后不得已的可怕备选，当你负担不起用智能体以超快方式完成时才会使用。而在此过程中，随着聊天编码被超越，氛围编码将继续存在。
By Q3, today's chat coding will for many have become a dire fallback of last resort, reserved for when you can't afford to do it the superfast way with agents. And through it all, as chat coding is eclipsed, vibe coding will live on.
我尽力在图 1 中展示了我个人对此的看法。
I've done my best to represent how I personally think about this stuff in Figure 1.
图 1 中的图表描绘了编程的六波重叠浪潮：传统（2022 年）、基于补全（2023 年）、基于聊天（2024 年）、编码智能体（2025 年上半年）、智能体集群（2025 年下半年）和智能体舰队（2026 年）。
The chart in Figure 1 depicts six overlapping waves of programming: traditional (2022), completions-based (2023), chat-based (2024), coding agents (2025 H1), agent clusters (2025 H2), and agent fleets (2026).
在图中，传统和基于补全的编码——这两种手动模式——正在下降，而其他模式正在指数级上升。从聊天开始，每一波新浪潮的上升速度都比前一波快得多。最后，该图将氛围编码也描绘为指数级增长，但与其他浪潮并列的虚线，因为稍后我们将看到，氛围编码不是一种模式。
In the figure, traditional and completions-based coding – the two manual modalities – are on the decline, and the others are rising exponentially. Beginning with chat, each new wave rises much faster than previous waves. Finally the figure depicts vibe coding as also increasing exponentially, but on a dotted line alongside the others, because as we'll see in a bit, vibe coding is not a modality.
作为我们讨论的预告，“智能体集群”是我使用的占位术语，指开发者能够并行运行并有效管理许多编码智能体，甚至可能超过开发者本地机器能容纳的数量。而“智能体舰队”则是指当我们为叶节点配备 AI 监督者时的情况，如图 2“FY26 组织架构图”所示。
As a sneak preview for our discussion, "agent clusters" is the placeholder term I'm using for devs being able to run and fruitfully manage many coding agents in parallel, potentially even more than can fit on a dev's local machine. And "agent fleets" are what happens when we get AI supervisors for the leaf nodes, as shown in Figure 2, "FY26 Org Chart".
该图显示组织中所有个人贡献者（叶节点）开发者像二级直线经理一样运作，运行 AI“管理智能体”，而这些管理智能体本身又监督着编码智能体组。例如，在单个 IC 开发者的指导下，一个被管理的智能体组可能正在处理缺陷积压清理，另一个正在开发新的业务功能，第三个组则在进行一项长期运行的架构迁移。一个名副其实的智能体农场！
This figure shows all individual contributor (leaf-node) devs in the org acting like second-level line managers, running AI "manager agents," which themselves are supervising groups of coding agents. For instance, under the guidance of a single IC developer, one managed agent group might be doing bug backlog grooming, another working on new business features, and a third group working on a long-running architectural migration. A veritable agent farm!
这当然只是对事情如何展开的一个粗略近似，但我认为它足够接近。我们都预测了“第四波”编码智能体的到来，它们比我们大多数人预期的来得更快。而且现在已经可以手动跳上第五波，尽管需要付出努力。我一直在并行运行两个智能体。当你这样做时，很明显很多工作都可以借助智能体式帮助来推进。
This is of course just a crude approximation of how it will unfold, but I think it's close enough. We all predicted the "wave four" coding agents were coming, and they arrived faster than most of us expected. And it's already possible to jump on wave five manually, albeit with effort. I've been running two agents in parallel. When you do this, it becomes evident that a lot of the work can be facilitated with agentic help.
智能体如何帮助智能体？今天，你必须注意到一个智能体工作线程何时卡住、完成或偏离方向，并适当地推动它。监督智能体很快就能并且将开始为我们做大部分这些工作。结果：第六波。开发者将能够使大型编码智能体舰队中的工作队列保持满负荷，在庞大的企业遗留代码山脉中艰难前行。这将是辉煌的。
How do agents help with agents? Today, you have to notice when an agent worker is stuck, done, or gone astray, and nudge it appropriately. Supervisor agents can and will start doing most of that for us very soon. The result: wave six. Developers will be empowered to keep work queues full in large fleets of coding agents, grinding their way through vast mountain ranges of enterprise legacy code. It will be glorious.
这些神奇的智能体舰队最迟将在 2026 年初到来。因为构建它们实际上并不难——我们已经非常擅长并行化工作。
And these magical agent fleets will be here by early 2026 at the _latest_. Because building them isn't actually that hard – we're already very good at parallelizing work.
这就是我们的快速介绍。更多内容即将到来。如果这一切对你来说是一个巨大的意外，那么你在未来几个月将面临一些麻烦的水域。
That's our lightning intro. Lots more to come. If all this is a big fat surprise to you, then you're in for some troubled waters in the months ahead.
如果你仍然认为基于 AI 的代码自动补全建议是程序员使用 AI 的主要方式，并且/或者你仍在衡量补全接受率（CAR），那么你正坐在图 1 中代表传统编程的、形似恐龙的曲线上。这条曲线在 2027 年左右将急剧滑向过时。
If you still think of AI-based code-autocompletion suggestions as the primary way programmers use AI, and/or you are still measuring Completion Acceptance Rate (CAR), then you are sitting on the vaguely dinosaur-shaped curve representing Traditional Programming in Figure 1. This curve super-slides into obsolescence around 2027.
我有坏消息：代码补全在一年前非常流行，那时感觉像是遥远的序章。但它们现在已是 AI 界的“行尸走肉”。
I have bad news: Code completions were very popular a year ago, a time that now feels like a distant prequel. But they are now the AI equivalent of "dead man walking."
如果你更前卫一些，你可能会认为基于聊天的编程是今年发展的方向——即 IDE 内的编码助手聊天界面，如 Copilot、Cursor、Sourcegraph 和 Windsurf。如果你属于这一群体，那么你做得并不差。处于中游水平，值得表扬。你至少采用了一种有用的模式——与代码补全相比极其有用——而且聊天仍在日益流行。
If you're a bit more avant-garde, then you might think that chat-based programming is how things are going to unfold this year – meaning, in-IDE coding assistant chat interfaces like Copilot, Cursor, Sourcegraph, and Windsurf. If you're in this group, then you're not doing too badly at all. Middle-of-the-pack, pat on the back. You've at least adopted a modality that's useful _– extremely_ so compared to code completions – and chat is still rising in popularity.
但突然之间，我们迎来了最新一波浪潮，新的编码智能体如 Aider.chat 和 Claude Code——很快，类似且更漂亮的智能体将出现在你所有喜爱的 IDE 中，_眨眼暗示咳嗽_。
But suddenly we have this latest wave, the new coding agents like Aider.chat and Claude Code – and soon, similar and prettier agents in all your favorite IDEs, _wink nudge cough cough_.
一旦你尝试过编码智能体，并学会了如何高效使用它们，你将再也不想回头。它们将碾压聊天编码。而且妙处在于，使用智能体时你仍然在“氛围编码”。这就是为什么它不是一种模式：你可以通过任何非手动的 AI 模式进行氛围编码：聊天、智能体、集群。只要 AI 在做工作，你就在氛围中！与智能体的唯一区别是你不需要频繁地与它们会合。
Once you have tried coding agents, and figured out how to be effective with them, you will never want to go back. They are going to stomp chat coding. And the great thing is, with agents you are still vibe coding. That's why it's not a modality: You can vibe code with any non-manual AI modality: chat, agents, clusters. As long as AI is doing the work, you're vibing! The only difference with agents is that you don't rendezvous with them as often.
既然智能体已经出现，我们可以开始看到模式。从聊天开始的每一波连续模式，保守估计都比前一波高效约 5 倍。聊天可能比手动编码高效 5 倍，智能体可能比聊天高效 5 倍，依此类推。请注意，如果不受挑战并给予时间成熟，每一波可能发展到比前一代高效 10 倍。但它们不断被新的、更快的模式所碾压。
Now that agents have emerged, we can start to see patterns. Each successive modality wave, beginning with chat, is conservatively about 5x as productive as the previous wave. Chat can be 5x as productive as manual coding, agents can be 5x as productive as chat, and so on. Note that each wave would probably grow to be 10x as productive as its predecessor, if left unchallenged and given time to mature. But they keep getting flattened by new, even faster modalities.
这就是我今天看到的局面。我们发现自己身处 AI 海洋中的一场大竞赛，被越来越猛烈的浪潮拍打。成功者将驾驭这些浪潮。每家公司都落在图 1 中一条或多条采用曲线上。你身处何方？
That's the situation as I see it today. We find ourselves in a big race in the AI ocean, beslapped by increasingly violent waves. The ones who make it will ride those waves. Every company falls somewhere on one or more of the adoption curves in Figure 1. Where are you?
亲爱的朋友们，这就是我对未来的迷人、迪士尼式的心理模型。我断言即将到来的集群和舰队浪潮不仅不可避免，而且几乎就在眼前。氛围编码仍然是那个景观中持久且永恒的特征——但并非大多数人认为的方式。氛围编码仅仅意味着再也不写代码。
And that, dear folks, is my charming, Disneyfied mental model of what lies ahead. I've asserted that the upcoming waves of clusters and fleets are not only inevitable, but practically right around the corner. Vibe coding remains a durable and lasting feature of that landscape – but not in the way most people think. Vibe coding simply means never writing code again.
如果你到目前为止同意我的观点，那么让我们看看财务影响。首先，我会快速让你了解编码智能体的工作原理——这并不复杂，你只需开始烧钱，烟雾会让它们更聪明。如果你到目前为止不同意我的观点，我鼓励你去玩玩这些新的编码智能体。说真的。或者看看懂行的人怎么用。
If you're with me so far, then let's take a look at the financial impacts. First I'll quickly get you up to speed on how coding agents work – it's not complicated, you just start burning money and the smoke makes them smarter. And if you're not with me so far, I encourage you to go play with these new coding agents. Seriously. Or watch someone who knows how.
无论你是信服还是怀疑，至少让我们看看这些新的编码智能体实际上是如何工作的。因为其中没有魔法。
Whether you're convinced or skeptical, let's at least look at how these new coding agents actually work. Because there's no magic.
让我们看看为什么这个仅几周前的发展可能很快让你的公司陷入真正的困境。你可能会说，一个棘手的困境。一团糟。
Let's see why this mere weeks-old development could quickly put your company in a real bind. A right pickle, you might say. A fine kettle of fish.
我们以前听说过软件编码智能体的说法。但这次不同。这些“真正的”编码智能体仍然非常非常新，最多只有几周大，而且它们只在基于文本的 1970 年代 Unix 风格终端中运行。得到一个这样的智能体，很像你一辈子都在走路，然后有人给了你一头免费的骆驼。事实上，他们说，你想要多少骆驼就拿多少。拥有一头骆驼是很棒的。一头。与到处走路相比，它们很棒，但它们会朝你吐口水、咬你，并且需要大量的绿叶食物，主要是 50 美元和 100 美元的钞票。
We've heard claims of software coding agents before. But this is different. These "true" coding agents are still _very_ _very_ new, weeks old at best, and they only run in text-based 1970s Unix-style terminals. Getting one of these is a lot like you've been walking all your life and someone gives you a free camel. In fact, they say, take all the damn camels you want. And it's amazing having one. _One_. They're great compared to walking everywhere, but they'll spit on you and bite you and they require large amounts of leafy green food, primarily fifties and hundreds.
你们中的很多人，我确切知道，一直对聊天编码持怀疑态度。我甚至听说有些开发者已经明确无误地向他们的经理表示，他们想继续编写代码。他们说，这就是他们在这里的目的。编写。代码。他们说得很慢，好像他们认为你聋了，这样会有帮助。他们声称永远不会把编码工作委托给 AI。嘿！我看到你了。
A lot of you, I know this for a fact, have been super skeptical about chat coding. I've even heard that some developers have expressed to their managers, clearly and unambiguously, that they want to keep writing code. That's what they're here for, they say. Writing. Code. They say it slowly, like they think you're deaf and it would help. They claim they're never going to delegate their coding work to an AI. Hey there! I see you.
所有怀疑论者都应该放下手头正在做或拿着的东西，直接扔在地上，然后拼命跑向最近的骆驼并跳上去。下载并尝试一个编码智能体，最好是 2025 年 3 月 1 日之后发布的。因为它们会把你所知道的或自认为知道的关于用 AI 编码的一切彻底颠覆。我自己在三周前几乎不敢相信我所看到的。
All of you skeptics should drop whatever you're doing or holding, just throw it on the ground, and run like mad towards the nearest camel and hop on. Download and try out a coding agent, ideally one launched after March 1st 2025. Because they turn whatever you know or thought you knew about coding with AI, _right_ on its head. I myself could scarcely believe what I was seeing, just three weeks ago.
编码智能体在原理上很简单。它们的工作方式就像典型的氛围编码聊天会话，LLM 完成大部分分析和繁重工作，而你主要戴着耳机。但有了智能体，你不需要做所有丑陋的双向复制粘贴和相关提示工作，这是缓慢的人类部分。相反，智能体接管并为你处理这些，只在完成、卡住或你钱花光时返回与你聊天。
Coding agents are simple enough in principle. They work just like a typical vibe-coding chat session, with the LLM doing most of the analysis and heavy lifting, and you mostly wearing headphones. But with agents, you don't have to do all the ugly toil of bidirectional copy/paste and associated prompting, which is the slow human-y part. Instead, the agent takes over and handles that for you, only returning to chat with you when it finishes or gets stuck or you run out of cash.
而且它们通常能在完全无人协助的情况下取得相当不错的进展。它们只是埋头苦干，直到把任务做对，根据需要抛出 token 来探索空间。人类被移除了 90-99%工作的瓶颈，但除此之外，它基本上就像聊天氛围编码的更快版本。
And they often get pretty darn far, entirely unassisted. They just grind away at their task until they get it right, throwing tokens at the problem to explore the space as needed. The human is removed as the bottleneck for 90-99% of the work, but otherwise it's pretty much just like a faster version of chat vibe coding.
除了成本之外，与聊天的唯一实际区别是，智能体可以一次执行更大的子任务，可能包含许多单独的步骤。在此期间，监督开发者可以腾出手来做重要的工作，比如吃完那袋奇多和浏览 HN。
The only practical difference from chat, aside from cost, is that agents can perform much larger subtasks at a time, potentially encompassing many individual steps. During this time, the supervising developer is freed up for important work like finishing off that bag of Cheetos and browsing HN.
为了具体说明，你可能会告诉一个编码智能体类似这样的话：“这是 JIRA 工单号；请去修复它。”这就是你需要说的全部。智能体会首先努力获取 JIRA 工单的访问权限：它可能会寻找 JIRA 命令行工具，甚至可能询问你是否可以下载它。它甚至可以为自己编写一个一次性程序，以编程方式获取工单字段。我们经常看到它们编写一次性程序。
Just to make it concrete, you might tell a coding agent something like, "Here is JIRA ticket #; please go fix it." That is all you would need to say. The agent would first try hard to get access to the JIRA ticket: it might look for the JIRA command-line tool, maybe even asking you if it can download it. It could even write a throwaway program for itself to fetch the ticket fields programmatically. We see them write throwaway programs pretty often.
一旦智能体能够读取工单，它就会使用你机器上的工具，像你一样检查你的代码，以追踪你的 bug。它会请求你允许每个工具——这是目前过程中最大的减速之一。一旦智能体找到 bug，它会提出修复方案，编写测试来验证修复，运行这些测试，并进行任何其他必要的更改以使测试通过——所有这些都在一个循环中完成，大部分时间不需要你。
Once the agent can read the ticket, it uses tools on your machine, examining your code just like you would, to track your bug down. It asks you for permissions for each tool – one of the biggest slowdowns in the process today. Once the agent finds the bug, it will propose a fix, write tests to verify the fix, run those tests, and make any other changes necessary to get the tests passing – all in a loop without needing you, for the most part.
这些新的编码智能体可以解决巨大的问题，制造更大的混乱，并且通常表现得像一个超自然速度的人类开发者，总是有点盲目和落后于计划。
These new coding agents can solve huge issues, create even bigger messes, and generally behave like a supernaturally fast human developer who's always flying a little blind and a little behind schedule.
这听起来像科幻小说，但你现在就可以使用它们。
It sounds like science fiction but you can use them right now.
重要的是要理解，这些新智能体目前仍然只能一次处理适度小到中等规模的任务。任务图分解，这是我们在过去的聊天时代（12 月）都学会的技能，在如今转向使用智能体进行氛围编码时同样重要。甚至更重要，因为使用智能体很容易过度和过于雄心勃勃。它们如此高效，以至于很容易变得贪婪并扼杀这只鹅。
It's important to understand that these new agents are still only capable of handling modestly small-ish to medium-esque tasks at a time. Task graph decomposition, a skill we've all learned during the chat days of yore (December), is just as important today as you switch to vibe coding with agents. Even more so, because it's so easy to overshoot and be over-ambitious with agents. They are so incredibly effective that it's easy to get greedy and smother the goose.
对你的鹅好一点。不要让它吃得太饱。你需要分解任务并小心地引导编码智能体。如果你给它的任务太大，比如“请修复我所有的 JIRA 工单”，它会拼命冲向问题，但几乎一无所获。它们今天需要仔细的监督和深思熟虑的任务选择。简而言之，它们是脾气暴躁的小动物。
Be nice to your goose. Don't overstuff it. You need to break things down and shepherd coding agents carefully. If you give one a task that's too big, like "Please fix all my JIRA tickets", it will hurl itself at the problem and get almost nowhere. They require careful supervision and thoughtful problem selection today. In short, they are ornery critters.
但这种情况将会改变。在你还没来得及挥动蝙蝠（说到小动物）之前，智能体就会潜入你的 IDE，不是作为骆驼，而是作为备好鞍的马：主要是人体工程学上的改进，当然，但也是受欢迎的。有一个不会向 36 米外的物体高精度喷射恶臭液体的工具，那将是很好的。
But that will change. Before you can so much as lash a bat, speaking of critters, agents will creep into your IDE, not as camels but as saddled horses: a mostly ergonomic improvement, sure, but a welcome one. It will be nice to have a tool that can't spit foul-smelling fluids with high accuracy at objects up to thirty-six meters away.
从现在开始，工具的每一次迭代都将帮助编码智能体变得更容易、更可并行化、更强大。而且我们将在今年更频繁地看到真正戏剧性的进步。
Every iteration of the tools from here on out will help make coding agents easier, more parallelizable, and more powerful. And we'll start seeing truly dramatic steps forward even more often this year.
本节面向首席信息官和财务人员。大家好。感谢你们读到这里。
This section is for CIOs and finance folks. Hi. Thanks for reading this far.
在你们几周前刚完成的 FY26 规划中，为开发者 LLM 支出预留了多少运营预算？可能不多？很多？有家公司告诉我，他们考虑给每位开发者每天 25 美元的慷慨预算。这看起来很大胆，像是一大笔钱，几乎有些鲁莽。
In your FY26 planning that you _just_ wrapped up a few weeks ago, how much opex budget did you put aside for developer LLM spend? Maybe a little? A lot? One company told me they were considering a generous budget of $25 per developer _per day_. That seems bold, like a lot of money. An almost reckless amount.
好吧，事实证明他们走对了路。编程智能体非常昂贵，非常非常贵。它们消耗大量 LLM 令牌，按当前费率每小时 10-12 美元。你们现在为编程助手支付的每席位许可费是多少？每月 30 美元？大概？可能更少？
Well, it turns out they were on the right track. Coding agents are très cher, muy caro, we're talking very, very expensive. They burn lots of LLM tokens, to the tune of $10-$12/hour at current rates. How much are your per-seat licenses for your coding assistant right now? Thirty a month? Ballpark? Maybe less?
为了计算方便，作为经验法则，你可以将每个编程智能体实例视为大约相当于增加一名初级软件开发人员的价值，在本财年摊销——前提是有人（人类或 AI）让它每天大部分时间（8-10 小时）保持忙碌。
For calculation purposes, as a rule of thumb, you can think of each coding agent instance as being approximately as valuable, amortized over this fiscal year, as having one additional junior level software developer on staff – provided that someone (human or AI) is keeping it mostly busy for 8-10 hours a day.
这真是个了不起的经验法则。我想你会同意，每小时 10 美元对于一位只需要一个好保姆的专业软件工程师来说，简直是捡便宜。
That's a heck of a rule of thumb. I think you'll agree that ten bucks an hour is a steal for a professional software engineer who just needs a good babysitter.
因此，值得你为每位开发者每天预算 80-100 美元的 LLM 支出。每天 30 美元只够三个小时的“骆驼骑行”，之后你的开发者午饭后就得回去步行。但如果你付全额的“本杰明·富兰克林”（100 美元），每位开发者将轻松实现产出翻倍，因为他们可以同时照看两个智能体并完成其他工作。这毫无疑问。
So it's going to be worth your while to budget more like $80-$100 of LLM spend per developer, per day. $30 a day is only going to be enough for three hours of camel rides, and then your devs go back to walking after lunch. But if you pay the full Ben Franklin, each of your devs will easily _double_ their output, since they can play nanny for two agents and get other work done on the side. It's a no-brainer.
即将到来的浪潮，我称之为“智能体集群”——上一节我暗示的战车——将在第三季度登陆。这一浪潮将使你的每位开发者能够同时并行运行多个智能体，每个智能体处理不同的任务：修复漏洞、优化问题、新功能、积压清理、部署、文档，以及开发者可能做的任何事。
The upcoming wave, which I'm calling "agent clusters" – the chariot I hinted at in the last section – should make landfall by Q3. This wave will enable _each_ of your developers to run many agents at once in parallel, every agent working on a different task: bug fixing, issue refinement, new features, backlog grooming, deployments, documentation, literally anything a developer might do.
你的每位开发者将突然变得像多个开发者。至少，那些擅长的人会如此。（伏笔：我嗅到了复仇的味道。）
Each of your devs will suddenly become like many devs. At least, the ones who are good at it will. (_Foreshadowing: I smell revenge.)_
智能体集群的副作用将是最终将软件开发迁移到云端。几十年来人们一直在预测基于云的 IDE！对吧？你们中一半人可能曾尝试构建过。它们看起来是如此明显的想法。
Agent clusters will have the side effect of finally moving software development into the cloud. People have been predicting cloud-based IDEs for decades! Right? Half of you have probably tried to build one at some point. They seem like such an obvious idea.
但本地运行 IDE 总是更方便，因此基于云的开发从未流行起来。2025 年下半年的智能体集群浪潮将改变这一点。你的开发桌面没有足够的能力同时运行数十个智能体，更不用说数百个。开发者的大部分工作将几乎在一夜之间转移到云端。
But it has always been more convenient to run IDEs locally, so cloud-based development never took off. The agent-clusters wave of H2 2025 will change that. Your dev desktop does not have enough power to run dozens of agents at once, let alone hundreds. The bulk of developer work will shift up to the cloud practically overnight.
所以你可能需要更多的云预算。
So you probably need some more cloud budget.
同时运行 N 个智能体将开发者日常的每小时 10 美元支出乘以 N，这还不包括云成本，仅仅是令牌消耗。如果你的每位开发者平均同时运行五个智能体——这是一个非常保守的数字，因为智能体将大多独立工作，让开发者自由做其他事情——那么每位开发者现在每小时花费 50 美元，或大约每年 10 万美元。
Running N agents at once multiplies your devs' innocuous daily spend of $10/hr by whatever N is, and that's not counting cloud costs, just token burn. If your developers are each on average running, say, five agents at once – a very conservative number, since agents will work mostly independently, leaving the dev free to do other stuff – then those devs are each now spending $50/hr, or roughly $100k/year.
这不再是捡便宜，而是抢劫。我们讨论的是每位开发者到 2025 年第四季度逐步将生产力提升约 5 倍（考虑爬坡时间），而第一年的额外摊销成本仅为每年 5 万美元左右。谁会拒绝这样的交易？
It's not really a steal anymore, so much as a heist. We're talking about each developer gradually boosting their productivity by a multiplier of ~5x by Q4 2025 (allowing for ramp-up time), for an additional amortized cost of only maybe $50k/year the first year. Who wouldn't go for that deal?
不幸的是，你几乎肯定没有在 2026 年运营预算中为每位开发者包含每年 5 万美元的 LLM 支出。这种情况将迅速将公司分为有预算者和无预算者，而有预算者将拥有优势。无预算者几乎连一半都得不到。明白我的意思吗？
Unfortunately, you almost certainly didn't include $50k/year _per developer_ of LLM spend in your 2026 operating budget. This situation will rapidly separate companies into the have-budgets and have-budget-nots, and the haves will, well, have it. The have-nots'll hardly have half. Follow me?
更直白地说：软件开发现在是一辆付费才能乘坐的高速列车。如果你买不起票，你就有被甩在后面的风险。
To put it more bluntly: Software development is now a pay-to-play bullet train. If you can't afford a ticket, you risk getting red-shifted away from the pack.
这里开始会让人有点不舒服。如果你已经在冒汗，或者感到任何心悸，不妨休息一下，拿瓶汽水，掸掸简历上的灰，随便你。慢慢来。等你准备好了。我们等着。
Here's where it starts to get a little bit uncomfortable. If you're already sweating, or feeling any kind of palpitations, maybe take a little break, grab a soda, dust off that resume, whatever. Take your time. Whenever you're ready. We'll wait.
好了，从现在开始，你们都保证冷静，童子军荣誉。开始吧。
OK, from _here on_, you all promise you are chill, scout's honor. Let's do this.
集群之后的一波，或者用个更合适的词——智能体“舰队”，将让你的开发者能够同时运行 100 多个智能体……这还要借助更多智能体的帮助。监督智能体能够管理成组或成群的编码智能体，代表它们进行协调，只有在智能体真正卡住时才引入人类。
The wave after clusters, or agent "fleets", for lack of a better word, will allow your developers to run 100+ agents at a time… with the help of yet more agents. Supervisory agents will be able to run groups or pods of coding agents, mediating on their behalf and only bringing in the human when agents get really stuck.
软件开发者未来的新工作很快将变成管理编码智能体及其 AI 监督者的仪表盘，如图 2 所示：FY26 组织架构图。有些人可能会轻蔑地称这份工作为“保姆”，并指责 AI 是爱哭的机器人宝宝，需要大人把食物切成小块、换尿布、清理烂摊子、防止它们从围栏里跑出来。但我们更愿意称之为“软件开发”。这就是我们的宿命。
The new job of a software developer going forward will soon be managing dashboards of coding agents and their AI supervisors, as sketched in Figure 2: FY26 Org Chart. Some might derisively call this job _babysitting_, and accuse AIs of being little whiny baby robots that need a grownup to cut their food into little pieces, and change their diapers, and clean up their messes, and keep them from wandering out of their playpens. But we prefer to call it _software development_. This is our destiny.
对于 CIO 类型的人来说，智能体舰队将使你的开发者每天花费数千美元。即使推理成本暴跌，杰文斯悖论也会导致更高的使用量抵消这些成本。如果你不相信，去问问你的 bug 积压情况；它基本上是无限的。
For you CIO-types, fleets will enable your developers to spend thousands of dollars a day. Even if inference costs plummet, the Jevons Paradox will result in higher usage offsetting those costs. If you don't believe that, go ask to see your bug backlog; it's basically infinite.
每天数千美元！？但这钱花得非常值！你的工程组织将开始能够按照你希望的速度前进，就这一次。你能相信吗？这就像再次成为初创公司。你将能够像杰夫·贝佐斯喜欢说的那样，“让客户惊喜和愉悦”，达到你从未梦想过的精英水平。
Thousands a day!? But it's money incredibly well spent! Your engineering org will start to be able to go as fast as _you_ want them to go, for once. Can you believe it? It'll be like being a startup again. You'll be able to "surprise and delight your customers", as Jeff Bezos is fond of saying, at an elite level you never dreamed possible.
但你必须找到大量新的预算。也许你很幸运，你的公司财力雄厚。例如，我刚听说，就在发稿时，一个你们都知道的大品牌今年有一笔非常大的自由支配资金用于 LLM 实验。我想知道有多少公司这样做了，并因此可能无意中躲过了今年的预算规划黑天鹅事件？
But you're going to have to find a whole lot of new budget somewhere. Maybe you're lucky and your company has deep pockets. For instance I just heard, going to press here, that a big familiar brand you all know has a very large slush fund allocated for LLM experimentation this year. I wonder how many companies did that, and in doing so, perhaps unwittingly dodged this year's budget-planning black swan event?
如果在沙发缝里找硬币后，你仍然无法在年底前为每个开发者凑出额外的 5 万美元，也许你可以通过某种方式筹集资金。目前，这可能更有利于初创公司而非大公司。我认为这个智能体的事情已经拉平了很多竞争环境。
If after searching the couch for coins, you're not able to scrape together an extra $50k per developer by EOY, maybe you can raise the money somehow. This probably plays better into the hands of startups than big companies right now. I think this agent thing has leveled a lot of playing fields.
这个故事可怕的部分是，如果你找不到或筹不到钱，又想保持竞争力，那么你将不得不做出痛苦的削减以释放运营预算。如果你再算一遍账，只有一个部门削减起来是合理的。
The scary part of this story is that if you can't find or raise money, and you want to stay competitive, then you're going to have to make painful cuts in order to free up the opex budget. And if you run the numbers again, there's only one department where it makes sense to cut.
剩下的，恐怕就是留给读者的练习了。我不知道会发生什么。我只是个普通人。也许这一切都被夸大了，我的预测需要再加六个月才能成真。我和 Claude 争论了一会儿，Claude 让步说，如果我把所有估计都延长六个月，那是合理的。所以坏消息可能并不全是坏事！
The rest, I'm afraid, is an exercise for the reader. I have no idea what will happen. I'm just some dude. Maybe this is all overblown, and you'll need to add six months to my projections before they come true. I argued with Claude about it for a while, and Claude capitulated and said it was plausible if I stretched all my estimates by six months. So the bad news mayn't be all bad!
现在说好消息！你确实要求先听坏消息，对吧？不管怎样，坏消息已经过去了。从这里开始就轻松了，我们快完成了。剩下的只有甜蜜的复仇。
And now for the good news! You did ask for the bad news first, didn't you? Well regardless, it's out of the way. It's easy peasy from here, and we're almost done. All that's left is sweet, sweet revenge.
事实证明，前路并非一片黯淡。恰恰相反！软件行业将会有大量工作岗位。只是不再是那种像野蛮人一样手写代码的工作。
It turns out, it's not all doom and gloom ahead. Far from it! There will be a _bunch_ of jobs in the software industry. Just not the kind that involve writing code by hand like some sort of barbarian.
自从我发表《初级开发者的消亡》以来，过去一年中我观察到一个一致的模式：初级开发者实际上比高级开发者更热衷于采用 AI。这并非总是如此；少数人告诉我们，他们的初级员工害怕使用 AI，因为他们有点非理性地认为 AI 会抢走他们的工作。（参见：行为遗憾理论。感谢 Daniel Rock 博士的指点！）
One consistent pattern I've observed in the past year, since I published "The Death of the Junior Developer", is that junior developers have actually been _far_ more eager to adopt AI than senior devs. It's not always true; a few folks have told us that their juniors are scared to use it because they think, somewhat irrationally, that it will take their jobs. (See: Behavioral regret theory. Thanks for the pointer Dr. Daniel Rock!)
但大多数情况下，初级开发者——包括（a）新晋开发者，（b）仍在学校的开发者，以及（c）还在考虑上学的开发者——都学得很快。他们拿起 O'Reilly 的《AI 工程》一书（现在所有开发者都需要从头到尾掌握），并将其视为职业培训。他们都在使用聊天编码、编码助手，而且我知道很多初级开发者已经在使用编码智能体了。
But for the most part, junior developers – including (a) newly-minted devs, (b) devs still in school, and (c) devs who are still thinkin' about school – are all picking this stuff up really fast. They grab the O'Reilly AI Engineering book, which all devs need to know cover to cover now, and they treat it as job training. They're all using chat coding, they all use coding assistants, and I _know_ a bunch of you junior developers out there are using coding agents already.
初级开发者们状态正佳。他们明白了。世界在变，你必须适应。所以他们适应了！
Junior devs are vibing. They get it. The world is changing, and you have to adapt. So they adapt!
而高级开发者呢，嗯……温和地说，他们在挣扎。我有很多好朋友，像我这样的老手，基本上从未接触过 LLM，甚至没见过它们“裸奔”。还有很多人只是浅尝辄止地使用过编码助手。我甚至从许多行业领袖那里听说，有些高级开发者群体公然反对 AI。
Whereas senior developers are, well… struggling, to put it gently. I have no shortage of good friends, old-timers like me, who have basically never touched an LLM or even seen one naked. Plenty of others have only barely dabbled with coding assistants. I even hear about senior developer cohorts, from many industry leaders, who take a flat-out stand against it.
举个例子：一家知名品牌的技术总监刚刚告诉我，他们的一位开发者发来一份 PDF，用彩色幻灯片和图表解释为什么他们都需要放弃 AI，回归常规编码。现在你明白我为什么说我们拥有技术史上最广泛的理解分布了吗？还有人认为这就像加密货币之类的。哎呀！
Example: a tech director at a well-known brand just told me that one of their devs sent them a PDF explaining, with color slides and charts, why they all needed to abandon AI and go back to regular coding. Now do you see what I meant when I said we have the widest distribution of understanding in tech history? There are people who still think this is like, crypto or something. Yikes!
听着，有些高级开发者确实在挣扎，因为他们太忙了。我理解。但我认为对大多数人来说，还有更深层次的原因。当我以前写关于编程语言的博客时，我发现仅仅说我喜欢某种编程语言——任何语言——就会让我陷入非常严重的麻烦。人们会在帖子里大喊大叫，唾沫横飞。我不明白发生了什么。就因为我喜欢一种语言？
Look, some senior devs are struggling, no doubt, because they're just busy. I get it. But I think for most of them, there's something deeper going on. When I used to blog about programming languages, I found that simply saying that I _liked_ some programming language, any language at all, would get me in surprisingly serious hot water. People would yell in threads, digital spittle flying everywhere. I didn't get what was going on. All this, just because I said I liked a language?
几年后我明白了，因为他们觉得如果人们听我的，大家都会改用那种语言，然后高级开发者也得学它。他们把学习新东西——真正的新东西，有点像从头开始——等同于失去工作、失去医疗保险、破产、死在医院外的台阶上。这只是人类在面对巨大不确定变化时的本性。
After several years I figured out that it's because they felt if people listened to me, then everyone would switch to that language, and then the senior devs would have to learn it too. They equated having to learn something new – and I mean really new, sort of like starting over – with losing their job and their health insurance and going bankrupt and dying outside a hospital on the steps. It's just human nature at work, in the face of big uncertain change.
我相信拒绝 AI 的人不幸地在现状上投入了很多，他们错误地认为这等同于工作保障。他们都告诉自己，AI 还没有证明自己在执行 X、Y 或 Z 方面比他们更好，因此它还没准备好。
I believe the AI-refusers regrettably have a lot invested in the status quo, which they think, with grievous mistakenness, equates to job security. They all tell themselves that the AI has yet to prove that it's better than they are at performing X, Y, or Z, and therefore, it's not ready yet.
但在我看来，他们才是没准备好的人。朋友们，我详细阐述这些，是为了让你们能帮助自己。
But from where I'm sitting, they're the ones who aren't ready. I lay this all out in detail, my friends, so you can help yourselves.
无论勒德分子不采用 AI 的原因是什么，他们已经输了。初级开发者占据了高地，战斗已经结束。不仅初级开发者平均采用 AI 更快，而且初级开发者——惊喜！——更便宜。如果公司要削减开支，以便让开发者用 token 取胜，你认为他们会留下哪些开发者？
Regardless of _why_ the luddites aren't adopting it, they have lost. Junior devs have the high ground, and the battle is now over. Not only are junior devs on average adopting AI faster, but junior devs are also – surprise! – cheaper. If companies are going to make cuts to pay for their devs to win with tokens, which devs do you think they're gonna keep?
AI 的抵制者还看不到这一切的到来，所以初级开发者们将不得不放下光剑，从山顶上喊出这个信息：
The AI holdouts don't see _any_ of this this coming yet, so the junior devs are going to have to lower their light sabers and shout this message down from the hilltop above:
否则，你就会进入……else 子句。你知道的。else 子句。岩浆。你掉进去了。为什么我还要明说？
Otherwise, you get… the else clause. You know. The else clause. The lava. You fall in. Why do I have to spell this out?
所以！现在我们到了电影的结尾。你成功了。击个掌吧，初级开发者们！✋去年我没想到这一点，但我非常印象深刻，你们竟然成功地把失败者投票淘汰出局。也向那些已经搞明白并继续前进的高级开发者们击个掌！你们的人数比你以为的要少，至少在湾区泡沫之外是这样。
So! Here we are, at the end of the movie. You made it. High five, junior devs! ✋I didn't see this coming last year, but I'm very impressed that you've managed to be the ones voting the losers off the island. And high five to those of you senior devs who _have_ figured this out and are truckin' along already. There aren't as many of you as you'd think, at least outside the Bay Area Bubble.
至于其他人……投入进去。像初级开发者一样。说真的。你是谁并不重要，甚至不管你是个人还是公司。投入进去。是时候了。AI 已经来了。
For the rest of you… _lean into it._ Be like a Junior. Seriously. It doesn't matter who you are, really, nor even if you're a person or a company. Lean in. It's time. AI is here.
在 Sourcegraph，我们每天都在深入研究这个问题空间。我们正朝着一个世界努力，在那里，所有这些尽管极其昂贵，但将惊人地、很快无可否认地变得有价值。对每个选择使用它的人来说，都是如此。编码智能体大军正在渡过卢比孔河，将它们与企业 IP 资产和代码库连接起来是下一个大热门。这就是我们专注的方向。
Here at Sourcegraph, we study the hell out of this problem space every day. We are working towards a world where all this stuff, despite being incredibly expensive, is astonishingly, and soon undeniably, _valuable_. To everyone who chooses to use it, anyway. Coding-agent armies are marching across the Rubicon, and getting them wired up to enterprise IP assets and code bases is the next big game in town. That's where we're focused.
更广泛地说，我们都认为会有工作。很多工作。我们认为目前的招聘停滞只是公司在表明他们还不知道该怎么做。但就在本财年，各种规模和形态的公司都可以比以往任何时候都更有雄心，我这么说毫无夸张。如果历史有任何指示，从蒸汽到电力到计算，我们很快就会看到更多的人创建软件。由此产生的生产力浪潮可能会使国家 GDP 增长惊人，达到 100%或更多。
More broadly, we all think there will be jobs. Lots of jobs. We think the flat hiring right now is just companies signalling that they don't know what to do yet. But this very fiscal year, companies of all shapes and sizes can be more ambitious than ever before, and I say that without a hint of hyperbole. If history is any indicator, from steam to electricity to computing, we're going to see vastly more people creating software soon. The resulting productivity wave may boost national GDPs by astonishing amounts, 100% or more.
但要参与其中，你必须学习下一波浪潮。作为开发者，甚至如果你是 PM 或任何与技术相关的角色——你需要赶上编码智能体，并保持跟进。不再拖延和浅尝辄止。现在就学会如何使用自动编码智能体，在掌握它之前不要放弃。推动它直到它为你工作。
But to participate, you're going to have to learn the next wave. As a developer, heck, even if you're a PM or literally any tech-adjacent role – you need to catch up on coding agents, and stay caught up. No more dallying and dabbling. Figure out how to use an automatic coding agent right now, and don't give up until you know how to wield it. Push on it until it works for you.
不要自以为是，试图过度使用它。编码智能体就像一台大型隧道掘进机，而你之前用的是动力铲。它很强大，当然，非常强大。但它很昂贵，仍然可能严重卡住，你需要时刻小心引导。而且它没那么快——它不会在一天内挖通英吉利海峡。所以不要设定不切实际的期望。只需专注于这东西与两年前 ChatGPT 刚出来时有多么不同，然后惊叹于它与两个月前我们最好的聊天工具相比有多么不同。
And don't get cocky and try to push it too hard. A coding agent is like a big-ass tunnel borer machine when you've been using power shovels. It is strong, sure, hella strong. But it is expensive, it can still get stuck badly, and you need to guide it carefully at all times. And it's not _that_ fast – it's not going to bore through the English Channel in a day. So don't set unrealistic expectations going in. Just focus on how different this stuff is from 2 years ago when ChatGPT came out, and then marvel at how different it is from 2 _months_ ago when the best we had was chat.
享受它。它被称为“氛围编码”是有原因的。事实证明，不写代码相当容易。
Have fun with it. It's called vibe coding for a reason. It turns out not writing code is pretty easy.
不要落入诱人的工作拖延陷阱。说“六个月后会快得多，所以我把这项工作推迟六个月”就像说“我要等交通高峰过去”。你的车程当然会更短。但你会最后一个到达。
Don't fall prey to the tempting work-deferral trap. Saying "It'll be way faster in 6 months, so I'll just push this work out 6 months" is like saying, "I'm going to wait until traffic dies down." Your drive will be shorter, sure. But you will arrive last.
智能体正在到来。庞大的舰队。不仅仅是编码智能体。智能体正在各处涌现，贯穿整个业务和生产技术流程。今天早上我和一个大客户谈过，他们已经构建了数十到数百个“AI 任务机器”——定制的智能体，执行他们庞大工作流中的特定部分。未来已来。智能体就在这里。
Agents are coming. Vast fleets of them. Not just coding agents. Agents are arising _everywhere_, across entire businesses and production tech processes. I talked to a big customer this morning who has already built dozens to hundreds of "AI task machines" – custom-built agents that perform specific parts of their giant workflows. The future is now. Agents are here.
如果你在寻找行动号召，那么我给人类和公司同样的建议：切换到聊天。放弃补全。停止手写代码。学习验证和确认在新世界中如何运作。熟悉这个领域，并跟进最新技术。停止抱怨，把它变成一项工程练习。保持领先。你能做到。
If you're looking for a call to action, then I give the same advice to both humans and companies: Switch to chat. Ditch completions. Stop writing code by hand. Learn how validation and verification work in the new world. Familiarize yourself with the space, and follow the state of the art. Stop whinging and turn this into an engineering exercise. Stay on top of it. You can do it.
最重要的是，密切关注新的编码智能体。它们今天对大多数开发者来说可能几乎不可用，但不会持续太久。完全不会太久。它们是极其昂贵的生产力机器——而且与人类相比价格低廉。所有人面前都有艰难的选择。
Above all, pay close attention to the new coding agents. They may be nigh-unusable for most devs today, but not for long. Not long at all. They are incredibly expensive productivity machines – and at bargain-basement prices compared to humans. Tough choices ahead for all.
到今年年底，“软件工程师”的新工作将涉及很少的直接编码，而大量的智能体照看。你越早接受这一点，你的生活就越轻松。
The new job of "software engineer," by the end of this year, will involve little direct coding, and a _lot_ of agent babysitting. The sooner you get on board with that, the easier your life will be.
如果你读到这儿真的不知道该做什么，去找一个初级开发者帮忙。
If you're really not sure what to do after reading this far, go ask a junior developer for help.
差不多就是这样。我们即将迎来“作弊就是一切”的两周年纪念。从那以后发生了太多疯狂的变化。如果我能把这篇博文发回过去，两年前的我不会相信。
That's pretty much it. We are coming up to the 2-year anniversary of "Cheating is all you need". It's absolutely insane how much has changed since then. If I could send this blog post back in time, my 2-years-ago self wouldn't believe it.
感谢我的老板 Quinn Slack，一周前用这些想法中的大部分震撼了我们。希望这对你有所启发。我要去给投资者放屁了。再见！
Thanks to my boss Quinn Slack for blowing our minds with most of these ideas a week ago. Hope you found this useful in some way. I'm off to fart on some investors. Ciao!
借助 Sourcegraph——面向企业的代码理解平台。
With Sourcegraph, the code understanding platform for enterprise.