本集简介
双语字幕
仅展示文本字幕，不包含中文音频；想边听边看，请使用 Bayt 播客 App。
我们现场直播参加AI工程师峰会，嘉宾是传奇人物史蒂夫·耶吉，他是Stevie's Tech Talks、Stevie's Platforms品牌的创始人，最近还涉足了Source Graph和AMP。
We are here live at AI Engineer Summit with Steve Yeghi, the legendary Steve Yeghi of Stevie's Tech Talks, Stevie's Platforms brands, and most recently, Source Graph and AMP.
欢迎。
Welcome.
最近，还有Coding，
And most recently, Coding,
我应该说，没错。
I should That's right.
《Vibe Coding》这本书。
The Vibe Coding book.
所以这是一场关于Vibe Coding的重磅讨论。
So this is the big Vibe Coding discussion.
在会前聊天中，我们讨论了Vibe Coding与AI工程的交汇点。
In the pre chat, we were discussing the intersection of Vibe Coding and AI engineering.
所以我们今天请来了这两个领域的领军人物。
So we got the kind of movement leaders of both sides here.
你怎么看？
How do you see it?
这绝对是一个运动。
It's absolutely a movement.
对吧？
Right?
你得让更多人支持它。
You gotta get people behind it.
我的意思是，我在今天的演讲结尾提到，现在出现了巨大的反弹，而这种反弹才刚刚开始酝酿。
I mean, there's a I said at the end of my talk today that there's a huge backlash, and the backlash is only just brewing now.
所以你和我正在推动这些浪潮，你知道的，AI工程是关于构建AI赋能的应用程序并投身于AI领域。
So you and I are pushing forward, right, on on these waves of, you know, AI engineering is about is about building AI enabled applications and being being in AI.
而Vibe Coding则是关于抛弃旧的软件开发方式，拥抱新的方式。
And vibe coding is about abandoning the old ways of producing software and embracing the new ways.
对吧？
Right?
这两者都让很多人感到愤怒。
And both of these are making people pretty mad.
对吧？
Right?
我认为，如果他们的身份与今天的工作方式紧密绑定，且不容任何改变，他们才会生气。
I don't I think they're mad if their identity is tied to the way that they work today with no changes no room for changes.
是的。
Yeah.
那我先说说我的第一个大胆观点。
So I'll start with my first hot take.
好的。
Okay.
来吧。
Let's go.
有一个群体受此影响最大。
There is a demographic that is the most affected by that.
他们的身份与工作方式紧密相连。
Their identity is the most tied up with the way that they work.
好的。
Okay.
不是初级工程师。
It's not junior engineers.
也不是非工程师。
It's not non engineers.
他们都在搞氛围编程。
They're all vibe coding.
是资深工程师、资深领导者，基本上可以将范围缩小到拥有十二到十五年经验的人。
It's senior engineers, senior leaders, people who have so, basically, you can you can narrow it down to twelve to fifteen years of experience.
他们讨厌氛围编程，也讨厌人工智能，还在网上说：我十五年的经验比那AI强。
They hate vibe coding, and they hate AI, and they're online going, my fifteen years is better than that AI.
明白吗？
K?
你有没有看到 NVIDIA 的乔丹·哈伯德发的帖子，他详细阐述了一些关于如何在编码时最大化利用智能代理的绝佳建议。
That you saw I don't know if you saw Jordan Hubbard's post from NVIDIA where he just laid out some really nice advice on how to get the most out of agents as you're coding.
然后有个人回复说：不，不是这样的。
And this guy posted and he's like, yeah, you know, no.
你还是继续做你的总监工作吧，编程就交给程序员吧。
You you stick with your doing your director stuff and leave the programming to programmers.
对吧？
Right?
当你像我一样拥有十五年经验时，你才有资格发表意见。
When you have fifteen years of experience like me, then Then you're qualified to talk.
对吧？
Right?
没错。
Right.
所以我跟他说：我觉得你得学会看钟。
So I said something to him like, I think you need to learn to read a clock.
他这么说，直到你有十五年的经验，而我则说
And he's like, until you have fifteen years of experience, and I'm like
嗯，你的经验比他多。
Well, you got more experience than him.
或者我有四十五年。
Or I have forty five.
所以我要等到六十岁才能跟你说话吗？还是该砍掉三十年经验，好跟你一样无知？
So should I like go to 60 before I can talk to you, or should I like cut out thirty years of experience so I can be as dumb as you?
对吧？
Right?
这就是我的选择。
Those are my options.
所以我不确定。
And so I don't know.
我想我十五年后再见他吧。
I guess I'll see him in fifteen years.
我，我，好吧。
I I okay.
我想有一个元素是我正在试图弄清楚的，就是当这些人必须共存时，对吧？
I think there's one element that I'm trying to figure out of while these people have to coexist, Right?
大多数公司都会有一个混合团队。
And most companies are gonna have a mix.
顺便说一下，OpenAI也是如此，我们昨晚吃饭时还聊过这个。
Even OpenAI, by the way, like, we talked about this last night at dinner.
伙计们，OpenAI里有不使用AI来编程的人。
Guys, OpenAI has people who don't use AI to code.
他们有一些人不使用Codex。
They have people who don't use codex.
他们可能在用Cursor之类的东西。
They probably are using cursor or something.
好吧。
Okay.
但他们没有使用智能代理循环。
But they're not using the agentic loops.
对吧？
Right?
是的。
Yeah.
对。
Yeah.
而且确实如此。
And yeah.
我们还跟那边的开发产品总监安德鲁·格洛弗聊过，你知道的。
It's so, you know, we talked to, you know, Andrew Glover there, you know, the director of DevProd.
据他所说，他们计划在积累更多相关数据后公开这一做法。
And from what he was saying, they've been planning on going public with this once they have more data about it.
是的。
Yeah.
据 anecdotal 消息，他们表示性能提升了。
Anecdotally, they're sharing that The performance.
是的。
Yeah.
无论用哪种方式衡量，差距都是十倍。
Difference is like 10 x by any way that you measure it.
无论是代码行数、提交次数，还是业务影响，都如此。
So lines of code, commits, business impact, whatever.
这种差异如此明显，以至于那些没有采用它的人，在绩效评估时生产力直接降低了十倍。
And it's it's so stark and pronounced that the people who aren't adopting it are now 10 times less productive at performance review time.
两个人，同样的职位，同样的工作，突然间，其中一个的生产力是另一个的十倍。
Two people, same title, same job, and all of sudden, one of them is 10 times as productive than the other one.
你该怎么办？
What do you do?
答案是，你慌了。
And the answer is you panic.
你实际上会去找人力资源和法务部门，问他们：我们在这里有什么选择？
You actually go to HR and you go to legal, and you're like, what are our options here?
因为时间快到了。
Because the time is coming.
行吗？
K?
再来说一个热门观点。
Here's another hot take.
明白吗？
Alright?
如果你到明年1月1日还在用IDE写代码，那你就是个糟糕的工程师。
If you're still using an IDE to develop code by January 1, you're a bad engineer.
嗯。
Mhmm.
这够劲爆了吧。
There's a there's a hot take for you.
对吧？
Right?
现在你还有大约五到六周的时间，可以继续用IDE做个还不错的工程师，但你现在必须放下它，学习如何让智能体编写代码。
Now you still have a, what, five, six weeks to to still be an okay engineer while you're using your IDE, but this is the time that you need to drop it and learn how agents code.
明白吗？
Okay?
因为这是一种技能。
Because it's a skill set.
我的意思是，这太复杂了。
I mean, it's so complicated.
我和金珍写了这本书来讲解这个。
We wrote this book about it, me and Jean Kim.
因为去年我们自己在玩这个，一直在写博客、讨论它。
Because we were, you know, we were playing with it ourselves last year, and we're blogging about it and talking about it.
每篇博客文章都有30页。
And every blog post was 30 pages.
那你能拿这30页的博客文章怎么办？
It's like, are gonna do with 30 page blog posts?
这长度对我来说都太长了。
That's long even for me.
对吧？
Right?
是的。
Yeah.
在某个时刻，我只是想，要让AI完成大家抱怨它在做的那些事情，你得掌握多少技能啊。
And at some point, I was just, man, like the skills that you gotta learn in order to get the AI to do the things that everyone's mad because it's doing them.
对吧？
Right?
因为每个人都说：嗯，我试过了。
Because everybody's everybody's like, well, I tried it.
我花了两个小时，结果它产出的全是垃圾。
I spent two hours with it, and all it produced was garbage.
答案其实是，你得花两百个小时和它打交道。
And the answer is, actually, you have to spend two hundred hours with it.
你得花两千个小时和它打交道。
You have to spend two thousand hours with it.
这其实并不是夸张。
And that's not actually an exaggeration.
吉恩刚展示了一项研究，表明你实际上需要花一年或两千个小时与AI相处，才能信任它。
Gene just pulled up a study that showed that you actually have to spend a year or two thousand hours with AI before you trust it.
那么，信任在这里意味着什么？
And what does trust mean?
在这种情况下，信任具体指的是，作为用户，你能预测它会做什么。
Trust in this case specifically means before you, as a user, can predict what it's going to do.
如果它不可预测，当然你会生气。
And if it's unpredictable, of course, you're gonna be mad.
但一旦你花了一整年的时间与它磨合，彻底理解它的能力和局限——这些局限其实并没有根本性改变，它只是变得更强大了，但边缘情况始终如一。
But as soon as you've worked at it with it for a full year to where you fully understand its capabilities and its drawbacks, which haven't really fundamentally changed, it's gotten more capable, but the edges are always the same.
它会幻觉。
It hallucinates.
它会迷失方向。
It gets lost.
它会健忘，甚至痴呆。
It gets amnesia, dementia.
它会骗你，不管怎样。
It lies to you, whatever.
对吧？
Right?
这些能力，我们已经培养了多年。
Those skills, we've been building them for years now.
所有尝试用AI写代码的人，我们都在尝试。
Everybody who's been trying to write code with AI, we've been trying.
它其实没怎么成功，但一直在变得越来越好，越来越好，越来越好。
It hasn't really worked, but it's been working better and better and better and better.
现在它已经达到了比其他所有选项都好得多的水平。
And now it's reached the point where it's working a lot better than all of the other options.
是吗？
Yeah?
是的。
Yeah.
如果你两个月没试过，那就严重落后了。
And if you haven't tried it in two months, you're way out of date.
这些模型比两个月前好太多了。
The models are much better than two months ago.
如果你一年没试过，那你就是恐龙了。
If you haven't tried it in a year, you're a dinosaur.
你有多糟糕，真是难以置信。
It's just unbelievable how bad you are.
而且，你知道，你可能在看。
And and, you know, you may be look.
我有一些朋友，他们的工程能力比我强得多。
I have friends who are much better engineers than I am.
明白吗？
Okay?
我的意思是，那些世界级的、可能是全球最顶尖的工程师，他们开发过你听说过的技术，但他们至今还没使用AI，顶多偶尔会像查维基百科那样问问Cursor聊天问题。
I mean, world class, maybe some of the best in the whole world, okay, have built technologies that you've heard of, and they're not using AI yet, except the occasional I'll ask cursor a chat question like Wikipedia or whatever.
明白吗？
Okay?
这些人一年后就会变成实习生。
Those people are gonna be the interns in a year.
你真的这么认为吗？
You really think so?
是的。
Yeah.
尽管他们有你所说的那些经验。
With all their experience that you have.
我有一个假设，之前完全没有得到任何轶事证据的支持，直到今天在你们的会议上遇到一个人，他告诉我他曾经处于这样的境地。
I've had this hypothesis that has not been really confirmed with any any anecdotal evidence at all until today when I met somebody at your conference that told me about how he had been in this position.
有十二年经验，却完全不想接触人工智能。
Twelve years of experience, didn't want anything to do with AI.
他遇到了两位来自欧洲某地的博士生。
And he met these two PhD students from somewhere in Europe.
我忘了是哪里。
I forget where.
他们俩都是超级硬核的极客程序员，用着各种智能代理。
And they were both just super hardcore vibe coders, you know, with with the the agents.
对吧？
Right?
他看着他们工作，发现他们资历非常浅。
And he was watching them work, and they were super junior.
他们其实不太清楚自己在做什么，但毫无畏惧，充满抱负。
And they kinda didn't know what they're doing, but they just had no fear and all the ambition.
他们只是不断反复地敲打这个东西，说：好吧。
And all they did was they just kept hammering on the thing going, okay.
那你为什么这么做呢？
Well, why did you do it that way?
给我解释一下。
Explain it to me.
好吧。
Okay.
让我们来看看其他选择。
Well, let's let's let's look at other options.
他们就像是完全没有背景的完美工程师。
And they would just be kind of the perfect engineer with no context.
完全没有背景的完美工程师，他们会问什么问题？
The perfect no context engineer, what questions are they gonna ask?
你有考虑过扩展性吗？
Have you thought about scaling?
你考虑过安全性吗？
Have you thought about security?
你的测试覆盖率如何？
How is your test coverage?
对吧？
Right?
我的意思是，工程师们都会问同样的问题。
I mean, engineers are gonna all ask the same questions.
对吧？
Right?
他意识到，这个被局限的工程师与知道该向大语言模型问什么问题相差不远。
And he realized that engineer in a box is not too far off from knowing the right questions to ask an LLM.
而这两位学生用它时效率极高。
And that these two students were so productive with it.
他震惊了，心想：天啊，不是吧。
He was blown away that he was like, oh, no.
就是那一刻，灵感突然闪现。
Like, that's when the ray the light bulb went on.
他说，我必须学会这个。
He said, I have to learn this.
从那以后，他就一直这么做。
And now he's been doing it ever since.
对吧？
Right?
但这并不容易。
But it ain't easy.
你不可能随便拿个Claude来，试一下就觉得它能直接奏效，也许你运气好能成功一次，但别指望这样。
I I you're not gonna pick up Claude Coe, and you're not you're not gonna just try it and be like, it's just gonna work for it might you might get lucky.
但如果你没有正确的思维模式，没有正确的态度，即使有了正确的态度，过去两天里你有多少次对着你的代理破口大骂，用了真正的脏话？
But eventually, if you don't have the right mindset, if you don't have the right attitude going in now even with the right attitude, how often have you swear sworn at your agents in the last two days with the actual f word
或者说是吧？
or like right?
我很有礼貌。
I'm pretty polite.
我会说谢谢和请。
I say thank you and please.
我会说谢谢和请。
I say thank you and please.
然后你就说，什么？
And then you go, what
你干嘛这么做？
the did you do that?
对吧？
Right?
而且这是因为吉恩和我在出版这本书之后才意识到的。
And it's it's because it's because Gene and I realized this after we published the book.
你有个助手。
You had this helper.
它们非常像人。
They're very human like.
它们会进来。
They come in.
你得告诉它们很多东西，它们需要大量的指导。
You have to tell them a lot of stuff, and they they they need a lot of guidance.
但随着时间推移，它们需要的指导越来越少。
But over time, they need less guidance.
你的提示变得越来越简短。
Your prompts get shorter.
事情变得更加顺畅。
Things get streamlined.
它们似乎明白了。
They seem to get it.
它们在发挥作用。
They're working.
如果这是一个真人，你会得出结论：他们理解你、懂你，终于成为团队的一员了。
Now if this were a human being, you would draw the conclusion it's because they understand you and they get you and they're finally part of the freaking team.
不要对大语言模型犯这种错误。
Do not make that mistake with LLMs.
永远不要像拉里·埃里森那样，把大语言模型拟人化。
Never make the mistake of anthropomorphizing an LLM like Larry Ellison.
对吧？
Right?
大语言模型随时可能背后捅你一刀。
The LLM at any moment can stab you in the back.
明白吗？
Okay?
它可能会突然说：‘嗯，是的。’
It can just be like, yeah.
我们已经解决了那个棘手的问题。
We took care of that really hard problem.
现在我要删除你的数据库。
Now I'm gonna delete your database.
而你只是说：不。
And you're just like, no.
对吧？
Right?
正因如此，它才很脆弱。
And and it's because of that, it's weak.
我们称之为热手效应。
We call it the hot hand.
你有点觉得：这东西真在运转啊，老兄。
You you sorta like you're like, it's going, man.
我感觉不错。
I'm feeling good.
我觉得这玩意儿懂我。
I'm feel this thing gets me.
我要让它执行一个生产环境的变更。
I'm gonna make it do a production change.
这就是我发现这件事的方式。
And that's that's how I found out about this.
我当时想，我的脚本不应该能访问生产环境。
And it's like I was like, my script can't access prod.
所以它选择了以最糟糕的方式去做。
And so it chose to do it in the worst imaginable way.
它所做的就是锁定了整个宇宙的其余部分，包括我的在线游戏和其他所有内容，只允许我的脚本访问生产环境。
What it did was lock out the entire rest of the universe and including my live game and everything else and only allowed my scripts to access prod.
它在更改密码。
And it was changing password.
它更改了密码。
It changed the password.
我当时想，你为什么要改我的密码？
And I was like, why did you change my password?
对吧？
Right?
嗯。
Yeah.
然后我就说，哦，真抱歉。
And it's like, oh, I'm so sorry.
我确实不该这么做。
I I definitely shouldn't have done that.
记在哪儿了？
What what Noted in?
好的。
K.
我就只是对吗？
And I'm just right?
这就是你如果只是随便尝试自主编码会发生的事。
This is what will happen to you if you just you just try to do agentic coding.
嗯？
K?
会发生糟糕的事情。
Bad things will happen.
这其实正是我们这本书的主题。
This is what our book is about, really.
对吧？
Right?
嗯，我的意思是，这个广告并不好，那然后呢？
Well, I mean, that's not the best ad because then what?
比如，
Like,
你会学到东西，然后最终学会如何应对减速带、拐角以及一切。
you learn, and then eventually, you learn how the speed bumps and the corners and everything.
就像开车一样。
It's like driving.
对吧？
Right?
就像开车一样。
It's like driving.
就像，你想要成为一名纳斯卡赛车手一样。
Like, you you you you're you wanna become like a NASCAR driver.
这可是高性能的东西。
Like, this is high performance stuff.
你同时用12个代理在写代码，比以往任何时候都更有野心。
You're coding with 12 agents at a time, and you're you're more ambitious than you've ever been.
我今天跟一个家伙聊了，他手头的项目比我多得多。
I was talking to a guy today who's got got way more projects going than I'm I've got.
我不知道他哪来这么多时间，但他现在可能同时在进行十到十二个主要项目。
I don't know where he gets all the time from, but he's probably doing 10 or 12, like, major projects at the same time right now.
而他全都是用Agenta涂层在做这些。
And he's just doing it all with with Agenta coating.
你知道吧？
You know?
我的意思是，老兄，这里的广告说你会变成蝙蝠侠，但你不能只是抓起战衣穿上就说‘我是蝙蝠侠’。
So, I mean, like, man, the the the the the ad here is that you will turn into Batman, but you can't just grab the suit and put it on and be like, I'm Batman.
你只是个角色扮演者。
You're just a cosplayer.
你只是在扮演氛围编码的氛围。
You're cosplaying at vibe coating.
你得学会工具带怎么用，这会是一段痛苦、挫折、犯错和学习的过程。
You gotta learn how the tool belt works, and that's gonna be pain, suffering, and mistakes and learnings.
现在你可以通过阅读这本书以及所有其他氛围编码书籍来掌握很多内容。
Now you can get a lot of it by reading this and all of the other vibe coding books.
读一读奥莱利的书。
Read the O'Reilly.
看那个演讲。
Watch the talk.
我的意思是，认真的，你真的应该从各个角度去理解它，因为对不同的人来说，它的意义似乎也不一样。
I mean, seriously, like, you should, like, get all of the possible angles at it because it seems to land differently for different people.
总会有一个类比让你突然明白。
There'll be some analogy where you finally get it.
你知道吗，我懂了。
You know, I get it.
就像这样。
It's like this.
它就像一台3D打印机，别人都没觉得它像3D打印机，但不知怎么的，这个比喻就是对你起了关键作用。
And and and it's like a three d printer, and nobody else thought it was like a three d printer, but somehow that was the magic that made it for you.
对吧？
Right?
嗯。
Yeah.
我想说，昨天晚餐最大的意外之一，是发现有那么多人都有过这样的经历：他们再也不写单行代码了。
I would say one of the biggest surprises from the dinner yesterday was how many people all have the experience where they no longer write single lines of code.
他们其实就是在做提示，然后我去放
Like, they they're really just kind of prompting and and doing I going to put
那个。
that.
你意思是说，他们根本不会写任何代码？
Single lines of code, you mean they would never write any code at all?
他们可能会修改一下。
They might edit.
但我觉得，当他们编写全新代码时，总是从提示开始。
But, like, I think when they're writing net new, they always start with the prompts.
不修改。
No editing.
不碰。
No touch.
不修改。
No editing.
当你发现那个标识符拼错了，而且是个局部变量时，这代价可就太高了，你知道的。
It is it is very expensive when you're like, that that identifier is misspelled and it's a local, you know.
你可以直接加上它，但更好的做法是关闭你的IDE，甚至可能卸载它。
You could just add it, but it's better for you to close your IDE and probably uninstall it.
不。
No.
实际上，这并不对。
Actually, that's not true.
有人终于说服我，IDE真是太棒了。
Somebody finally convinced me that IDEs are fantastic.
特别是IntelliJ。
IntelliJ in particular.
保持它打开。
Keep it open.
如果是Gradle构建的话。
If it's a Gradle build.
是的。
Yeah.
实际上，这并不是为了LSP，尽管你可以用它来做这个。
And actually, not for the LSP, although you can use it for that.
实际上，如果你有一个MCP服务器，这也是使用LLM的另一个好方法。
Actually, that's another good way to use the LLM if you get an MCP server.
但不，是因为IntelliJ的自动索引速度要快得多，增量重建也比LSP快得多。
But no, it's that IntelliJ's auto indexing is so much faster and incremental rebuild is so much faster than LSP.
昨晚。
Last night.
是的。
Yeah.
是的。
Yeah.
所以你们要做的就是让IntelliJ一直运行，但你不应该去看它。
So you all all you do is leave IntelliJ running, but you shouldn't look in it.
现在它是一个AI工具。
It's a tool for the AI now.
对吧？
Right?
太棒了。
Amazing.
你提到的那些热门话题中，还有一件事很重要，你说Cloud Code不是。
One other thing that is a big part of some of the hot things you're saying, you say Cloud Code is not it.
Cloud Code不是。
Cloud Code ain't
不是。
it.
解释一下。
Explain yourself.
好吧。
Alright.
这里每个人都喜欢Cloud Code。
Everyone here loves Cloud Code.
这里每个人都喜欢Cloud Code，或者如果你使用我们的产品AMP，它最近又因为Gemini 3超越了Cloud Code。
Everyone here loves Cloud Code or or AMP if you use our product, which is is just recently leapfrogged Cloud Code again because of Gemini three.
AMP有一个很酷的功能，它可以调用另一个模型。
AMP has this cool feature where it goes to another model.
在开始之前，我也想谈谈谷歌整体情况，以及这场Gemini革命如何改变了谷歌的形象。
And just to prewarm you, I also wanna talk about just Google in general and how this Gemini revolution has kinda changed Google's image.
但让我们先谈谈Cloud Code。
But let's talk about Cloud Code.
好的。
Sure.
Cloud Code自三月以来就一直存在。
Cloud Code's been around since March.
Cloud Code已经被证明是有效的。
Cloud Code has been proven to work.
但即便如此，全球90%的程序员中，可能有80%并没有使用它或任何类似的东西。
And so but yet, probably 80% of the world's 90% of world's programmers are not using it or anything like it.
有些公司确实广泛应用了，但大多数公司并没有。
You you get certain companies where it's really taken off, you know, but but most aren't.
世界还停留在Cursor上。
The world is stuck on cursor.
整个世界还停留在2024年。
The word world is stuck in 2024.
去年，我们还在努力让人们用聊天方式写代码。
Last year, we were trying to get people to write with chat.
对吧？
Right?
我们告诉他们这么做，但他们却说：不行。
And we were like, we're telling and they were like, no.
代码补全。
Completions.
我们当时说：天啊。
And we were like, oh god.
不行。
No.
但它可以生成代码，你只需要粘贴进去。
But it can generate the code, you just gotta paste it in.
你只需要做所有这些事情。
You just gotta do all this stuff.
他们说：听起来有点难。
And they were like, that sounds kinda hard.
我们说：但这样更快。
And we're like, but it's faster.
但他们就是不愿意做。
And they they wouldn't do it.
九个月后，终于慢慢接受了，现在他们都觉得：我喜欢Cursor。
And then nine months later, it finally percolated in, and now they're all like, I like cursor.
这都已经是去年的事了，老兄。
And it's like, that's so last year, dude.
对吧？
Right?
醒醒吧。
Like, wake up.
但他们还没有采用它。
And yet they haven't adopted it.
所以，到了现在，你不得不想想，为什么他们还没采用呢？
And so you have to, at this point, look at it and say, why haven't they adopted it?
我们来看看原因吧。
Let's go look at the reasons.
答案是，太难了。
And the answer is, it's too hard.
太难了。
It's too hard.
你得会读啊，大多数工程师，说实话，对他们来说，五段话就是一篇论文。
You have to be able to read man, most engineers, honestly, like, to them, five paragraphs is an essay.
明白吗？
K?
使用 Cloud Code 时，你得阅读大量信息，不仅仅是文字，还有代码和差异内容。
And with Cloud Code, you've gotta read waterfalls of not just information, but also code and diffs.
对吧？
Right?
因为如果你要把 IDE 放一边，你就真的得去看差异内容。
Because if you're gonna put your IDE away, you actually do have to look at the diffs.
现在我要告诉你，一旦你掌握了这门技能，你其实可以通过差异的形状、颜色和长度来判断。
Now I'm gonna tell you that once you get some expertise at this, you can actually tell from the shape of the diffs and the color of the diffs and the length of the diffs.
这个
The
感觉。
vibe.
你可以判断出是否需要代码审查，是否有人在做错事，或者他们是否为这个问题过度重定向了太多代码。
You can tell whether it needs a code review, whether they're doing the wrong thing, whether they seem to be rerouting suspiciously too much code for this problem.
对吧？
Right?
仅凭差异的形状，就能让你对发生了什么有大量了解，而无需实际阅读代码，但你应该关注它们。
The diffs alone, just the shape of the diffs can tell you a lot about what's going on without actually reading the code, but you should pay attention to them.
否则，你以后会遇到问题。
Otherwise, you'll have problems that will only crop up later.
对吧？
Right?
但没错。
But yeah.
我的意思是，把IDE关掉。
I mean, like, put the IDE away.
好的。
Okay.
展开剩余字幕（还有 480 条）
粘贴代码，然后把 Clog 代码拿出去，试着开始使用它。
Clog code, and then get Clog code out and try to start using it.
好吗？
Alright?
你会发现它看起来是这样。
And you're gonna find that it's look.
说实话，我每天整整十到十二个小时都在用 Clog 代码，连续几个月、几个月、几个月，但我还是经常骂它。
I've been using Clog code, honestly, ten to twelve hours a day, literally, for for months and months and months and months, and I still curse it out all the time.
我真的要疯了。
I just lose my mind.
我会想，你明明刚说过是对的，怎么还能这么做？
I'm like, how could you have done that when you just said right?
实际上，这已经被证实了。
And it's like, it's actually been shown.
越来越多的研究表明，有时候当你给他们一点压力时，他们的表现反而会更好。
It's it's starting to be shown that sometimes when you put a little pressure on them, they perform better.
你可以通过这种方式突破堵塞。
You can break through logjams that way.
但不管怎样，你看，你会遇到问题，但关键是，明年工具会更好。
But anyway, look, you're gonna run into problems and and but the thing is, next year, the tools will be better.
好吗？
K?
如果Cloud Code不是，那什么才是？
If Cloud Code's not it, what is it?
好吧，我们得回到类似IDE的东西上。
Well, we gotta get back to something like an IDE.
对吧？
Right?
我的意思是，这必须对人们来说是自然的。
I mean, that's just gonna be it's gotta be natural for people.
你必须能够一看就明白发生了什么，而不是非要读代码。
You gotta be able to look at it and see what's going on, not have to read.
它必须有视觉指示器。
It's gotta have visual indicators.
对吧？
Right?
但它又不会是IDE，因为IDE主要专注于帮助你编写代码，而你现在已经不这么做了。
And and yet it's not going to be an IDE because an IDE is very much focused on helping you write code, and that's not what you do anymore.
对吧？
Right?
所以它将会是你的代理编排仪表板。
So what it's going to be is it's gonna be your agent orchestration dashboard.
你早上走进来时会说：嘿。
It's you're gonna walk in in the morning and be like, yo.
那么怎么做才对呢？
So how does Right?
它那些东西，嗯。
It things Yeah.
哦，那个还在运行。
It's like, oh, that one's still running.
那个在运行一个工具。
That one's running a tool.
那个需要我的输入。
That one needs my input.
好的。
Okay.
对吧？
Right?
你就逐个查看列表。
You just go through the list.
所以我正在做一个。
And so I'm building one.
你可以去看看，本来应该是私有仓库，但它是公开的，所以有人叉了，事情就发生了。
You can go look supposed to be a private repo, but it's public, so I've got forks and shit happens.
但不管了。
But whatever.
你可以玩一玩。
You can play with it.
它叫VC，VibeCoder。
It's called VC, VibeCoder.
这是我开发的VibeCoder系统的v2版本。
It's my v two of the VibeCoder system.
它的功能是创建一组扫描后的工作流，为你运行代理。
And what it does is it creates a set of scanned workflows that run the agents for you.
嗯。
Yeah.
我不知道你有没有看到谷歌前几天发布的反重力技术。
I don't know if you saw anti gravity from Google the other day.
我们两天前拍的。
We shot two days ago.
所以它是
So It's
真有趣，这么多人在你这个年龄结婚时发明了这么多东西。
so fun how much stuff people are inventing that are all your age of marriage.
是的。
Yeah.
是的。
Yeah.
不。
No.
你看。
So look.
我给它起了个名字，我不知道。
I called this I don't know.
我在三月给它命名的。
I called it in March.
在《初级开发者的复仇》中，我做了那个图表和所有内容，达里奥还在他所有的客户咨询委员会中引用它。
With Revenge of the Junior developer, I did that chart and everything and, like, Dario quotes it in all his customer advisory boards and everything.
对吧？
Right?
真的吗？
Really?
是的。
Yeah.
是的。
Yeah.
不。
No.
它确实产生了相当大的影响。
It was it was really pretty impactful.
我还说过会发生什么，那就是这些代理——早在三月，我就知道它们太难了。
And and I called that what's gonna happen is the that agents I even back in March, I knew they were too hard.
我当时就想，接下来会发生什么？它们可以编程运行，而你用它们做的90%的琐碎工作，其实都可以由一个模型来处理，通常还是个更便宜的模型。
I was like, what's gonna happen is they're they're you can run them programmatically, and 90 of the crap that you do with them could be handled by a model, often a cheaper model.
对吧？
Right?
如果只是比如，它在问你，这两个同样重要的事情，接下来该做哪个？
If it's just like if it's asking you, which of these two things should I do next that are equally important?
那就让Haiku随便选一个就行。
Like, just have Haiku say either one.
对吧？
Right?
所以我当时说，编排器就要来了，结果真花了快到年底才实现，这和我预测的时间差不多。
So, like, I called the orchestrators are coming, and it's taken close until, like, the end of the year to get there, which is roughly where I predicted them coming.
Replit、Agent three，还有很多其他的。
Replit, Agent three, there's a bunch.
还有Conductor。
There's there's Conductor.
有一个叫 VMAD 的开源项目发布了。
There's a VMAD came out open source.
它们各自有不同的理解和实现方式，你知道的。
They're all different, you know, takes on it.
对吧？
Right?
但还会有更多项目涌现。
And, but there's there would be more coming.
我想，谷歌也有。
I guess, Google's as well.
对吧？
Right?
是的。
Yes.
我喜欢他们这个比喻。
I like this analogy that they have.
这还是相当新的。
It's still pretty new.
所以谁知道最终的愿景会是什么？
So who knows what the eventual vision is?
你是说，你会在代理工作时收到它们的通知吗？
Is that you just get notifications from your agents as they're working?
没错。
Exactly.
是的。
Yeah.
所以在我这里的VC中，有一个活动流，这是我最早添加的功能之一，就是当我打算去工作时，我想定期收到一些有趣内容的通知。
So in mine, in VC, there's an activity feed that was one of the first features I added, which is like, I want to go work, and I just wanna get notifications periodically of interesting stuff.
很有趣。
Interesting.
我不知道它会不会有类似代理的社交网络。
I wonder if it'll have, like, social networks of agents.
嗯，这些代理之间会相互关注。
Well, so the agents each other, following each other.
嗯，我刚和杰弗里·埃曼纽尔喝了一个小时的咖啡，他就是那个做了MCP代理邮件的人。
Well, so I just had three hour coffee with Jeffrey Emanuel, who's who's he did the MCP agent mail.
他是我这辈子见过的最聪明的人之一。
He's one of the smartest people I've ever met in my life.
就是他写了那篇导致Stark市场崩溃的文章，关于英伟达的。
He's the one that wrote the article that crashed the stark market about NVIDIA.
哦。
Oh.
就是那个杰弗里·埃曼纽尔，他写了一篇极其出色的论文，解释了为什么这是个泡沫，整个市场都崩了，而卡帕西也开始关注他。
That Jeffrey Emanuel, the one that an incredibly well written article that said this is why it's a bubble, the whole market went, and Karpathy started following.
现在又涨上来了。
It's back up.
他写的就是你刚才说的那些。
He wrote what you just said.
他说已经回升了。
He said it is back up.
但他写了代理邮件，因为他厌倦了在各个代理之间复制粘贴东西。
But he wrote agent mail, which is he was just tired of having to copy stuff between his agents.
比如，你告诉我该对这个代理说什么。
Like, you tell me what to tell this agent.
于是他搞了个小小的、我不知道该怎么形容的HTTP服务器，像个收件箱，让它们能互相通信。
And so he made a little little, like, I don't know, HTTP server that's like an inbox for them, a messaging, and they talk to each other now.
现在他只需要说：你们自己协调，完成我刚布置的这个艰巨任务，它们就能搞定。
And now he goes, coordinate amongst yourselves to paralyze this task, this epic that I just put together or whatever, and they'll do it.
有些人从上往下入手，试图构建能替你完成一切的编排器。
Some people are coming at it top down and trying to build orchestrators that do it all for you.
但有趣的是，Beads，就是我做的那个问题跟踪会话工具，
But interestingly, beads, right, which is the issue tracker session thing that I that I made,
再加上他纯靠直觉写出来的代码。
plus his Purely vibe coded, by will.
是的。
Yes.
完全是氛围编码。
Purely vibe coded.
是的。
Yes.
所以，我的意思是，我每天都会收到一些我引入的糟糕问题的拉取请求，但没人介意，因为我们现在有稳定版本了。
So, I mean, like, I get PRs every day for horrible problems that I introduced, but nobody seems to mind because we've got stable versions now.
因此，Beads 证明了只要你和别人提出正确的问题，并让 AI 去查看代码，你就根本不需要亲自看代码。
So Beads is like living proof that you never actually have to look at the code as long as you and other people are asking the right questions and having the AI look at the code.
我经常收到别人发来的拉取请求，很明显所有分析和编码都是 AI 做的，而我只需要看一下。
I get PRs from people all the time where it's obvious that the AI did all of the analysis and all of the coding, and I look at it.
有时候我会问我的 AI：你觉得他们的 AI 提交的 PR 怎么样？
And and sometimes I'll just be like, so my AI, what do you think of their AI's PR?
对吧？
Right?
你看到了摘要。
And you saw summarization.
我的意思是，这不好吗？
I mean, isn't that bad?
你不希望你的代码看起来是这样的吗？
Don't you want It's bad if your code if look.
关键是结果。
It's all about the outcome.
VEEDS正在运行，已经有数以万计的用户非常满意地使用它。
VEEDS is working, and it's got tens of thousands of very happy people use it.
所以，显然这并不坏。
So, obviously, it's not bad.
我可能想，如果你要
I I may wanna If you do
如果你把这用在公司生产网站上并导致网站崩溃，那确实不好。
this to your your company's production website and bring it down, then, yeah, it's bad.
但即便如此，Beads 本质上还是一个数据库，你知道的，而数据库是最难构建的东西之一。
But still, Beads is kind of a database, you know, and database is one of the harder things to make.
你知道，Beads 非常奇怪。
You know, Beads is really weird.
它的架构真的很奇怪。
The architecture is really weird.
它之所以能工作，唯一的原因是，在过去它根本不可能实现。
And the only reason it works is because it wouldn't have worked in the old days.
在过去，它会因为太难管理、无法编程化而根本行不通。
It would have been just too hard to manage and and and not programmatically.
但你所做的就是告诉 AI：去把它全部修复好。
But what you do is you tell the AI, go fix it all up.
无论何时出现损坏、合并冲突，或者只是需要修复，就让它去处理。
And whenever it's corrupted or there's a merge conflict or just just fix it.
有趣的是，做邮件系统的杰弗里·埃曼纽尔基本上也做了同样的事情。
And it's funny because Jeffrey Emanuel, who did the mail, basically did the same thing.
他让所有的代理都在同一个目录下运行，并且它们会进行文件预留。
He has all his agents running the same directory, and they do file reservations.
它们会说：我需要那个文件。
They're like, I need that file.
天哪，九十年代我在安达信就是这样做的。
Man, I used to do that Accenture in the nineties.
对吧？
Right?
我会冲到某人的隔间里说：我需要那个文件。
I'd, like, run over to a dude's cubicle and be like, I need that file.
他们的版本控制太差了。
Their revision control was so bad.
所以，他搞了个文件预留系统，但一旦实施，他的代理们就开始正常工作了。
So, like, he's got a file reservation system going, but but it all what happened was as soon as he put it in place, his his agents just started working.
现在他有了一个由代理组成的小村落。
And now he's got this little village of agents.
对吧？
Right?
而这就是我们正在走向的方向。
And that's that's where we're headed.
因此，协调器的作用不再是把单个代理固定在轨道上，而是让所有代理都走在轨道上并相互沟通。
So the orchestrators are gonna be about not keeping the agent on the rails, but keeping all of your agents on the rails and communicating with each other.
是的。
Yeah.
然后你就撞上了墙。
And then you hit the wall.
砰。
Boom.
有人知道，在你突破了这一切之后，所谓的‘墙’是什么吗？
Does anybody know what the wall is once you get past all this?
合并。
Merge.
合并就是现在每个人都在遭遇的瓶颈。
Merging is the it's the wall that everyone is hitting right now.
是的。
Yeah.
我认为最有可能解决这个问题的公司是Graphite。
I think the company that's best poised to solve it is Graphite.
我本来打算去和他谈谈这个问题。
I was gonna go talk to him about it.
代码方面，他们会很乐意和你交流。
The code They'd be happy to talk to you.
是的。
Yeah.
是的。
Yeah.
我认为每个人都需要解决这个问题。
I think everybody needs to solve it.
如果你在一家企业，就像我们听到的那样，因为吉恩·金和我经常交谈，我们与所有公司交流，我纯粹是Sourcegraph的SaaS销售。
And if you're at an enterprise, like, what we hear because Gene Kim and I talk we talk to companies all the I'm a SaaS seller solely in the Sourcegraph.
所以我们能听到这些大公司内部的真实情况。
So we get to hear the the inside story from all these big companies.
对吧？
Right?
他们说，是的。
And they're saying, yeah.
一旦你达到每个开发者的生产力都提升十倍的阶段，合并代码就会变成一个极其复杂的问题，因为我和你同时工作两三个小时。
As soon as you get to the point where, like, every developer is 10 times as productive, merging their code becomes this incredibly complicated problem because I you and I work at the same time for two or three hours.
我们各自做出了大约三万行的代码变更。
We make, you know, 30,000 line change each.
我的代码先提交并被合并了。
Mine makes it in first, and it and it gets merged.
然后你过来时，我已经彻底改变了我们的日志系统、架构以及你所使用的API。
And then you come along, and I have literally changed our logging system and our, like, you know, our architecture here and APIs that you are using.
是的。
Yeah.
所以这不会那么简单，不是简单地解决合并冲突就行了。
And so it's not gonna be as simple it's not as simple, let's let's fix the merge conflicts.
你需要重新构想、重新设计并在我的改动基础上重新实现你的改动。
It's like you're gonna have to re envision and reimagine and reimplement your change on my change.
或者把你的代码删掉。
Or rip yours out.
或者把我的代码删掉，让我重新来做。
Or rip mine out and make me do it.
但最终，这些都由AI来完成。
But ultimately, ours are just the AIs doing it.
对吧？
Right?
没错。
Right.
但重要的是，它们必须被序列化。
But the the important thing is that they have to be serialized.
这是一个队列。
It is a queue.
当它们进入时，实际上必须基于新内容重新执行它们原本要做的操作。
And that when they go in there, they have to actually, like, basically redo what they were doing on top of the new thing.
没有人解决过这个问题，这目前是一个巨大的障碍。
This is Nobody has solved this, and it is a huge obstacle right now.
你知道一家公司是怎么做的吗？
You know what one company did?
抱歉。
Sorry.
最后一点。
Last thing.
嗯。
Yeah.
一家公司说，这是我们的解决方案。
One company said, here's our solution.
每个仓库一位工程师。
One engineer per repo.
我不是瞎说的。
Not making that up.
这是一个解决方案。
It's a solution.
这目前是一个解决方案。
It's a solution for now.
解决这个问题的经典方法是堆栈差异。
The the classic solution for this is stack diffs.
对吧？
Right?
合并队列，堆栈差异
Merge queues, stack diffs that
我不太了解堆栈差异，所以我想我挺笨的。
I don't know about stack diffs, so I guess I'm dumb.
这就像Facebook提出的一个概念，他们正试图将其推广到更广泛的领域。
It's a it's like a Facebook concept that they're trying to bring into the wider world.
GitHub正在添加这个功能。
GitHub is working adding it.
我可不是只跟贾里德·帕尔默聊过。
I didn't just talk to Jared Palmer there.
基本上，我现在听到的还没有解决方案，但你应该了解它，并围绕它进行设计。
Basically, I I'm hearing no solution yet, but you should be aware of it and design around it.
对。
Yeah.
我的意思是，这其实就是老办法，就是拼命硬扛，
I mean, it's the old fashioned way of just hammering through it really hard and
另外，你也可以直接跟对方说：嘿，我正在做一项相当深入的架构性改动。
Well, also, you know, you could just talk to the other guy and say like, hey, I'm doing this, you know, pretty deep architectural change.
让我先说，咱们先统一一下整体模式。
Let me go first and let's let's agree on the overall pattern first.
是的。
So yeah.
我的意思是，我遇到过几次这种情况，我确实试图提前通知这个代理，说另一个代理正在做影响它的更改。
I mean, I've run into this situation a few times where I've actually tried to give this agent the heads up that this one's making a change that affects this one.
是的。
Yeah.
关于杰弗里做的邮件功能，我想一旦我把它连上，因为他不用工作树，而我要
With the mail thing that Jeffrey did, I think once I get it wired up because he doesn't use WorkTrees, and I'm going to
是的。
Yeah.
这个。
This.
但一旦它们能真正互相通信，我认为只要记住那个代理正在处理影响你的事情，就会变得很简单。
But once once they can actually talk to each other, I think it's gonna be as simple as just keep in mind that that agent's working on something that affects you.
你可能想去和他们谈谈这件事。
You might wanna go talk to them about it.
嗯。
Yeah.
并且就整体的基本原则达成一致。
And agree on the overall, like, fundamental Ifrah.
他们在这方面非常擅长。
And they're quite good at it.
他们只是，嗯。
I they just Yeah.
什么？
What?
因为他们没有自我。
It's because they have no ego.
他们不会想，哦，必须得是我。
They're not like, oh, it's gotta be me.
对。
Right.
所以谁先来谁就当领导。
So just whoever's first gets to be a leader.
太好了。
Great.
你和他有什么分歧？
What do you and him disagree on?
我和谁？
Me and who?
杰弗里。
Jeffrey.
埃马纽埃尔，我刚认识的那个人？
Emmanuel, the guy that I just met?
嗯，我们在根本上不同意让12个代理在同一个代码库副本中工作是个好主意。
Well, we so we foundationally, fundamentally disagree that having 12 agents work in a single repo clone is a good idea.
所以你是支持这一方的？
So you're on the pro side?
我支持很多类似的做法，比如使用多个分支的 Git 工作树，或者独立的仓库克隆。
I'm on the pro lots of like, either Git work trees with lots of branches or separate repo clones.
我猜他他
I would imagine he He's
它们是被隔离的。
them sandboxed.
他支持这个做法。
He's in favor.
他把它们都放在同一个地方，它们都在用同一个 Git，同一个构建系统。
He's got them all in the same they're all Oh, I they're literally they're using the same Git, the same build.
所以其中一个会正在做构建，比如需要运行测试。
So one of them will be, like, doing a build believe, like, need to run a test.
是的。
Yeah.
这变化太多了。
That's much churn.
是的。
Yeah.
但他有一个文件预留系统。
But he has a file reservation system.
有趣的是，我当时觉得这太疯狂了。
So the funny thing is, k, I was like, this is insanity.
他让我至少承认，如果你是一个独立开发者，而且刚入门，使用的代理不超过十几二十个，这种方法实际上效果不错，因为他确实用得很好。
And he's talking me into at least acknowledging that it probably works pretty well if you're a solo dev and you're new using no no more than a dozen or 20 agents because it is actually working for him.
他使用的原理和Beads一样，那就是在以前这根本行不通。
And he uses the same principle that Beads does, which is it wouldn't have worked in the old days.
这对真正的工程师来说毫无意义。
It doesn't make any sense to a real engineer.
但你只要告诉AI，如果出了任何问题就修复它，它们真的会这么做。
And yet you tell the AI, if anything gets messed up, just fix it, and they will.
所以这是对的？
And so that's right?
这就是他的系统有效的原因，因为偶尔文件预留会出错，然后他们就会说：嘿。
That's why his thing works because every once in a while, the file reservation gets screwed up, and they're like, hey.
我们需要解决这个问题，然后他们就能搞定。
We need to resolve this, and they figure it out.
有意思。
Interesting.
是的。
Yeah.
有意思。
Interesting.
有些人提出，明年这个会议的主题可以是多智能体。
I I some people have proposed that the theme of this conference next year is on multi agents.
哦，是的。
Oh, yeah.
我的意思是，当然了。
I mean, yeah, of course.
对。
Yeah.
对。
Yeah.
我的意思是，人工智能将会围绕多智能体展开。
I mean, AI will be about multi agent.
你看，我们现在还处于用手镰刀收割玉米的阶段。
Look, we're in this phase still where we're we're cutting down corn with scythes with our hands.
这正是如今真正的程序员在做的事。
That's what a a real programmer does these days.
我们明年要搬家。
We're moving next year.
这非常清楚。
It's very clear.
我们正在转向这些机器，你知道的，那些像今天农场上看到的巨型机器，大规模养殖厂那样的机器。
We're moving to, you know, these these machines that churn, you know, these giant just like those ones that you see on the farms today, factory farms.
我们要进行代码的工厂化生产。
We're gonna be factory farming coat.
K？
K?
而这一点，确实有很多人在哲学上、道德上、伦理上，无论如何都坚决反对。
And that absolutely, like, a lot of people are just so dead set against that philosophically, morally, ethically, whatever.
他们只是觉得
They're just like
他们太习惯于自给自足的农业了，我们并不习惯那种大规模的，他们想要的是
They're so used to subsistence agriculture that we're not we're not used to, like, the big They wanna be
约翰迪尔。
John Deere.
但事实上，我们正进入编程的约翰迪尔时代。
But we are we are actually moving into the John Deere era of coding.
太棒了。
That's amazing.
是啊。
Yeah.
但有趣的是，这个类比。
But the funny thing analogy, actually.
我刚刚也想到了。
And I just thought of it too.
我们得重复利用它。
We'll we'll have to reuse it.
是啊。
Yeah.
但它一直在慢慢打动我。
But it's been it's been it's been growing on me.
这整个想法就是，Claude Code、AMP 和 Codex，还有 Klein，我们都一样喜欢。
It's the it's the whole it's this idea that Claude Code and AMP and Codex, you know, Klein, we love them all equally.
它们都同样糟糕。
They're all equally bad.
今天我的演讲中提到，它们就像电锯或电钻。
I said in my talk today, they're like they're like a a power saw or a power drill.
熟练的工匠可以用它们做很多好事，但你也会用它们割伤自己的脚。
A a skilled craftsman can do a lot of good with them, and then you can also cut your foot off with them.
Claude Code 也是如此。
The same thing's true with Claude Code.
但想象一下，有一个大型机器，能运行并清理 Cloud Code。
But imagine a big machine, a big farming machine that knows how to run Cloud Code and scrub it.
对吧？
Right?
差不多就像，好吧。
Almost it's like it's like, okay.
你做好规划。
You plan.
你去实施。
You implement.
你去审查。
You review.
你去测试。
You test.
对吧？
Right?
你把这一切拆分开，现在你就有了工厂化养殖。
And you split it all up, and now you got yourself factory farming.
对吧？
Right?
它有效果。
It works.
人们正在构建它。
People are building it.
这一定会发生。
It's gonna happen.
它将要做的就是已经开始让非程序员也能编程，这彻底颠覆了整个公司。
And what it's gonna do is it's gonna it's already started to unlock programming for nonprogrammers, and this is completely turning companies upside down.
他们开始意识到，理想的团队规模可能是两到三个人。
They're starting to realize that maybe the ideal time team size is, like, two or three.
没错。
Yep.
我的意思是，对吧？
And I mean, like, right?
公司运作的整个方式、整个治理结构都将改变，因为编程不再是瓶颈了。
The whole way that companies are run, the whole governance structure is gonna change because now coding is no longer the bottleneck.
业务部门必须立即参与进来。
The business needs to get immediately involved.
反馈循环会变得更快，这真是令人兴奋的时代。
This feedback loops get faster, and it's really exciting times.
但这对很多人来说太多了，他们要么干脆放弃，要么在网上反抗。
But it's too much for a lot of people, and they just they're they're they're, like, checking out or they're they're revolting online.
我预测，随着这些能力不断提升，当我们越来越接近代码的工业化生产时，我们会看到卢德分子的大规模反弹。
And I predict that as these capabilities improve and as we get closer and closer to the factory farming of code, we will see a massive backlash from the Luddites.
你
You
你是少数几个我可以问这个问题的人，因为我知道我们观众中的很多人对全面采用这种方式持批评态度。
are the one of the few people I can ask this as a I know a lot a lot of people in our audience are critical of going the full hog with this.
是的。
Yes.
所以很多人觉得，前端和应用代码可以，但别碰我的云基础设施。
So a lot like, they're like, fine for front end, fine for application code, but don't touch my cloud infra.
别碰我的后端，我的分布式微服务。
Don't touch my my back end, my distributed microservices.
绝对别碰任何生产环境的东西。
Definitely don't touch anything production.
只修改代码。
Only touch code.
一开始，只有在Git作为后盾时才使用这些工具。
Only use these things when Git is your backstop for starters.
好吗？
K?
所以别碰生产环境。
So keep prod out.
写代码会非常诱人，但别写。
It's gonna be real tempting to write, but don't.
如果你有Git作为后盾，为什么要担心呢？
If you have git as your backstop, why should you be worried?
没错。
True.
不过，我想人们觉得它在后端代码上效果没那么好。
Except, I guess, people have the perception that it is less good at back end code.
哦，这就是每个人数学都不好的问题。
Oh, this is the problem where everybody's bad at math.
是的。
Yeah.
好吧。
Okay.
那么，Chad GPT 3.5 在系统代码方面表现如何？
So how good was Chad GPT 3.5 at systems code?
很糟糕。
Pretty bad.
那是多久以前的事了？
How long ago was that?
好。
K.
两年前。
Two years ago.
人们认为，老实说，我认为这里的误解源于一种根本性的信念，即模型已经不再变聪明了。
People people think the honestly, I believe that the misunderstanding here is rooted in a fundamental belief that the models are done getting smarter.
对。
Right.
有趣的是，它们确实可能已经不再变聪明了。
And the funny thing is they could be done getting smarter.
它们并没有，但它们本可以，而我们仍然处于已经发现电力、现在需要加以利用的阶段。
They're not, but they could be, and we would still be over the hump where we've discovered electricity, and now we need to harness it.
是的。
Yeah.
即使使用今天模型的能力，我们依然会实现代码的工厂化生产，而且会很快实现。
We will still get to factory farming code with today's models capabilities, and it will get there fast.
到夏天我们就做到了。
We'll get there by summer.
但模型变得越来越聪明了。
But the models are getting smarter so fast.
你知道，这里其实有一种有趣的张力，那就是你正在为模型最终会内化到自身能力的工具进行构建。
You know, it's really there's there's this interesting tension of, you know, like, you're building tools for capabilities that the models will eventually have built into their brains.
是的。
Yeah.
所以你就不需要在工具中再保留这种能力了。
And so you won't need that capability in the tool anymore.
因此，这是一场持续的军备竞赛和工具的衰败——工具不断填补模型的空白，直到模型自身足够强大来填补这些空白，然后你的工具又转向新的领域。
And so there's this constant arms race and decay of your tool filling gaps for the model until the model's good enough to fill it itself, and then your tool moves on.
是的。
Yeah.
这就是我的意思。
That's what I mean.
所有代码和所有工具都正在变得一次性可用。
Becoming all code and all tools are becoming throwaway.
是的。
Yeah.
就像是
Like Which is
很棒，因为它们也更容易构建。
great because they're easier to build too.
是的。
Yeah.
顺便说一下，没错。
By the way, yes.
还记得乔尔·斯波尔斯基吗？我们这一代中最伟大的作家和思想家之一。
So remember Joel Spolsky, one of the greatest, you know, of our of our our generation, our time, one of greatest writers and thinkers.
他做过我见过最棒的技术演讲，我想请他回来重新演绎一次。
He gave the best tech talk I've ever seen, and I wanna get him to come and revive it.
他二十年前在亚马逊做的这场演讲。
He gave it at Amazon twenty years ago.
它仍然很有意义。
It's still relevant.
他被邀请来了吗？
He's invited here?
太好了。
Great.
很久以前，乔尔·斯波尔斯基写了一篇直到今天依然历久弥新的文章。
So Joel Spolsky, a long time ago, wrote something that was timeless until today.
所以这篇内容二十年来一直具有时效性，那就是
So it was twenty years timeless, which was
永远不要重写你的代码。
Never rewrite your code.
永远不要重写你的代码。
Never rewrite your code.
而现在我们发现，对于越来越多的代码库来说，与其试图修复，不如直接从头开始重写。
And now we've discovered that it is for a larger and larger and larger class of piece of bodies of code, it is better to just start over and rewrite it from scratch than it is to try to fix it.
大语言模型会做得更好。
The LLM will do a better job.
我第一次注意到这一点，是在我试图将所有的单元测试从一种架构移植到另一种架构时。
I first noticed this when I was trying to port all of my unit tests from one architecture to another.
最终，只是因为他们在不断修复，迭代变得毫无意义。
And eventually, was just, oh, just the iteration because they're trying to fix.
所以虽然有很多东西要保留，但如果你直接扔掉所有测试，重新编写，反而会更快，很快就完成了。
So there's a lot to keep in but instead, if you say, throw all the tests out and make them again, it just goes, and you're done.
对吧？
Right?
那么，我那个必须重构的库怎么办？
And so it's like, well, what about this library I gotta refactor?
这种情况正在逐渐增多，但我们正进入一个新时代：最快的做法是直接编写新代码，用更好的方式实现旧代码原本想完成的功能。
And so it's creeping up, but we're moving into a world where the fastest thing to do is just build new code that does a better job of what the old code was trying to do.
没错。
Yeah.
我的意思是，我们正在抛弃一切旧有的认知。
I mean, it's like we're unlearning everything.
我觉得已经够颠三倒四了，但这就像是我们进入了量子力学领域，你必须拥抱这个新世界。
I feel like enough upside down land, but this is it's like we've entered quantum mechanics, but you have to you have to embrace this new world.
我喜欢你带来的这种热情和可信度，因为一个年轻人说你这样的话，可能就没那么有说服力。
I love the energy and the credibility that you bring because a young kid could say what you're saying and not be as believable.
但你是从这样一个角度出发的：你已经是一个巨大的
But you're coming from the perspective of you've been a huge
我做了很久了，你也
I've been doing this a long You've been
游戏程序员。
a game programmer.
你什么都做过。
You've been everything.
是的。
Yeah.
我做过五年汇编语言，你知道的？
I've done I've done assembly language for five years, you know?
是的。
Yeah.
用汇编语言编写操作系统。
Operating systems in assembly language.
那是8086，还不是80x86。
And it was eighty eighty eighty eighty eighty six, not even f 80 x 86.
天啊，我们只有8位寄存器。
God, we had eight bit registers.
我都干过。
I've done it all.
你知道吗，游戏编程能让你学到一切。
And, you know, the game program game programming teaches you everything.
是的。
Yeah.
然后，当然，我还做过平台、谷歌、广告这些各种东西。
And then, of course, I've done platforms and Google and ads and this and that.
你知道吗，智能体循环和游戏编程循环有很多共同之处。
You know, the agentic loop and the game programming loops share a lot in common.
确实如此。
They do.
资源共享。
Resource sharing.
确实如此。
They do.
操作系统循环也是如此。
Operating system loops as well.
我感觉我一直在一遍又一遍地构建同样的系统，是啊。
I feel like I'm building the same systems over and over again Yeah.
就是这样，我们注定要在每个新领域中重新发明同样的设计。
It is it's it's there's only, we're cursed to reinvent the same designs in every new domain.
这也是一种特权。
It's a privilege too.
你知道吗？
You know?
我有一件事想让你评论一下，就是谷歌。
One thing I wanted to get you to comment on is Google.
就谷歌。
Just Google.
其中一个
One of
我最难忘的记忆之一，就是你退休前不久，我们讨论谷歌仍然不理解这一点，尤其是谷歌云，他们如何关闭
my favorite memories, which is, like, just before you retired was talking about how Google still doesn't get it, Google Cloud in particular, how they shut
了弃用政策。
down The deprecation policy.
弃用政策。
The deprecation policy.
我对此非常生气。
I'm so mad about that.
你得让我气得够呛，我才会写博客。
You gotta get me pretty mad to write a blog.
他们看起来有扭转局面吗？
They seem have they turned it around?
没有。
No.
我跟那里的一些人聊过，他们很多人都说，是的。
I talked to some people there, and a lot of them were like, yeah.
这根本不是谷歌的做法。
That's not a thing for Google.
有趣的是，你知道，亚马逊不在平台上，不涉及弃用问题，也不涉及重要事项。
And it's funny because, you know, Amazon Amazon not on the platform, not on the deprecation stuff, not on the important stuff.
谷歌在执行方面已经扭转了局面。
Google has turned it around on on execution.
是的。
Yeah.
他们终于做了本该在十五年前就做的事，那就是对人负责，而不是让工程师们一直为所欲为——这在过去二十年里一直是常态。
They finally did the thing that they should have done, you know, fifteen years ago, which is hold people accountable, and it's not just engineers do whatever they want all the time, which is what it was for twenty years.
这其实效果还不错，因为他们垄断了广告业务，有能力补贴工程师们为所欲为。
And it actually worked pretty well because they had a monopoly on ads, and they could afford to subsidize Google engineers doing whatever they wanted for all.
但最终，他们还是不得不做正确的事，成长为一个更成熟的组织。
But, you know, ultimately, they had to do the right thing and grow up and mature as an organization.
这个过程很痛苦，他们失去了一些谷歌文化，也不再那么有趣了。
It was painful, and they lost some Google culture, and it's not as fun anymore.
但他们现在执行得非常好，为公司做了正确的事。
But they now execute well, and they did the right thing for the company.
现在有了Gemini，你可以看到他们逐渐将重心转向人工智能，现在终于开始见效了。
And now with Gemini, you can see now they've they've been shifting their focus gradually towards more AI, AI, and now it's starting to pay off for them.
是的。
Yeah.
也许他们会成为最大的赢家。
And maybe they're gonna be the big big winners.
你对其他所有实验室也有类似的观察吗？
Do you have observations of a similar kind with all the other labs?
你知道吗，我只是好奇你对我的一张最爱图表的看法，就是那张微软所有人互相举枪的旧图。
You know, I I'm just kinda curious in your takes on one of my favorite charts is is that old chart where you had Microsoft, like, all pointing guns at each other.
是的。
Yeah.
脸书，每个人都是一个独立的实体。
Facebook, everyone's a thing.
有人问我这个问题。
Person to ask me this.
我记得那张图。
I remember that chart.
那张图很有趣。
That was funny.
是的。
Yeah.
有人也可以为OpenAI这么做。
It's just someone could do that for OpenAI.
他们可以。
They could.
他们可以。
They could.
这是一个有趣的问题。
You know, it's an interesting question.
这三家公司——谷歌、Anthropic和OpenAI——目前内部都极其混乱。
All three of those companies, Google, Anthropic, and OpenAI are an unbelievably chaotic internally right now.
是的。
Yeah.
混乱。
Chaos.
行？
K?
Anthropic 把它隐藏得很好。
Anthropic hides it really well.
他们看起来
They seem
他们看起来像是有条不紊。
they seem They seem like they've got their ass together.
这意味着他们的产品经理在混乱周围筑起了一道墙。
So what what what that means is their product managers formed a wall around that chaos.
干得好，Anthropic 的产品经理们。
And bravo, Anthropic product managers.
但这并不是因为 Anthropic 做得不好。
But it is and it's not because Anthropic's screwing up.
而是因为快速增长带来的必然结果。
It's because it's an inevitable function of growing that fast.
他们正在招聘一百多名员工来负责 Cloud Code，就在接下来的几个月里。
They're hiring, like, a 100 plus people for Cloud Code in the next, I don't know, month.
我的意思是，他们简直疯了，这还只是Cloud Code。
I mean, like, they're they're going wild, and that's just Cloud Code.
你不可能指望，我的意思是，我在谷歌和亚马逊经历他们高速扩张的时期，你只能接受混乱。
You're not gonna I mean, I was at Google and Amazon when they were in the big, big, fast, phases, and you're just gonna have chaos.
你会有人员流动。
You're gonna have churn.
没人知道该找谁，一切都乱糟糟的。
Nobody knows who to talk to what, and everything's crazy.
最终，情况会慢慢稳定下来，他们会找到方向的。
Eventually, it starts to smooth out, settle out, and they'll get there.
对吧？
Right?
OpenAI的混乱更像是，他们有很多人离职。
OpenAI is chaotic more like in a well, they had a lot of exits.
对吧？
Right?
你知道吧？
You know?
我不知道OpenAI是否像GitHub那样混乱，GitHub失去了大部分资深领导层，多年来一直处于彻底的混乱状态，但OpenAI确实相当混乱。
I don't know if it was chaotic as, say, GitHub, lost most of their senior leadership and was just complete turmoil for years, but they're pretty chaotic at OpenAI.
对吧？
Right?
还有谷歌，你知道吗，我们今天刚和一个人聊过，他说直到现在要让Jules团队在各个部门之间达成共识还是太难了。
And then Google, you know, I we were just talking to somebody today that was saying it was just still too hard to, like, get consensus across groups to with the Jules Jules team.
嗯。
Yeah.
他们无法在内部推广，因为谷歌的部门壁垒太严重了。
They can't get it rolled out internally because Google is so siloed.
这简直就是上十亿个单体架构。
It's a a billion monoliths.
对吧？
Right?
一些彼此不互通的小应用，导致在谷歌内部很难推行任何跨团队的项目。
Little little little apps that don't talk to each other that it's hard to roll anything out across Google.
所以这三家公司目前都存在执行问题。
So all three of them have execution problems right now.
我认为Anthropic的执行可能比另外两家稍好一点，但竞争非常激烈。
I think Anthropic's probably executing a little bit better than the other two, but it's real close race.
而且，是的，看看甲骨文或Facebook或其他公司能否赶上来，将会很有趣。
And, yeah, it'll be interesting to see and and see see if Oracle or Facebook or any of the others can catch up.
对吧？
Right?
Meta？
Meta?
Facebook将会是最有趣的事情。
Facebook will be the most interesting thing.
我的意思是，明年他们必须做些巨大的事情。
I mean, they'll have to do something huge next year.
明年可能是开源模型的年份。
Next year could be the year of open source models.
是的。
Yeah.
如果好的话，你看。
If well So look.
一旦开源模型达到与CloudSonnet三七相当的水平，你就可以打开Klein之类的工具，获得一个和三月时Cloud Code一样好的东西，而那时的Cloud Code还不如现在的水平，也不够好。
As soon as open source models get to the point where they're as good as CloudSonnet three seven was, then you turn on Klein or something, and you've got something that as good as Cloud Code was in March, which wasn't as good as today, and it's not good.
但它已经足够好了，而且你免费运行它。
But it's good enough, And you're running it for free.
免费。
Free.
免费。
Free.
在你的本地m4或任何设备上免费运行。
Free on your local m four or whatever.
对吧？
Right?
所以，是的，我了解过，据我所知，它们落后了七个月，但这个差距正在逐渐缩小，与前沿模型相比，这意味着LSS模型到明年夏天将和Gemini 3一样好。
So, yeah, I have and and from what I've heard, they they're seven months behind, and that that gap is gradually narrowing the frontier models, which means LSS models will be as good as Gemini three next summer.
对。
Right.
所以，是的，明年很可能就是关键的一年。
So, yeah, next year could very much be the year.
这意味着工具必须大幅改进，更好地分解任务，并为成本优化分配合适的模型和模型规模。
That means the tools are gonna have to get much, much better at decomposing the task and assigning them to the right model, the right size of model for cost optimization.
我来代表批判性观点，即它们之所以趋同，是因为已经趋于饱和。
I'll represent the critical side, which is that the reason they're converging is because they're saturating.
对吧？
Right?
因为能达到的极限只有100，你越接近100，按比例来说，就越难再进一步。
There there's only there's you can only ever hit 100, and the closer you get to 100, proportionally, it'll just get harder and harder.
对吧？
Right?
所以，显然，当你处于较低水平时，变化速率比已经达到饱和时要高。
So, obviously, the rate of change when you're lower down is is higher as compared to when you're already saturating.
但这只是一个细微的技术问题。
But that's a minor technical point.
不，不是这样的。
Well, no.
我的意思是，这根本不是小问题。
I mean, it's not it's not minor at all.
这实际上是一个根本性的问题：人工智能的智能发展曲线是会呈直线、指数增长，还是会开始达到峰值？
It's actually a foundational question, which is, is the line of AI intelligence gonna go straight, or is it going exponentially, or is it actually starting to peak?
渐近的。
Asymptotic.
是的。
Yeah.
是的。
Yeah.
而且，据我们从那些非常接近研究前沿的人那里听到的信息，我们知道在过去大约三十年里，由于摩尔定律，人工智能的智能水平每18个月就提升四倍。
And, you know, from what we've heard from people who are very, very close to the research, we know that AI has been getting, what is it, four times smarter every 18 months for the last, I don't know, thirty years because of Moore's Law.
他们认为，目前还剩下足够的训练数据，足以支撑再完成两个这样的周期，但之后会发生什么就不知道了。
And they think that there's enough data left, training data, for two more cycles of that before they don't know what happens.
是的。
Yeah.
也许它会继续上升更多，或者可能不会，我们还不清楚。
Maybe it goes up more or maybe it goes We don't know.
人类历史就此终结。
Human history ends.
但再经历两个周期，意味着三年后它们的智能水平将提升16倍。
But two more cycles means they're gonna be 16 times smarter in three years.
对吧？
Right?
我能看出来。
I can see.
所以，好吧，我甚至不知道那是什么意思。
So well, I don't even know what that means.
我花了很长时间试图弄清楚它的含义。
Well, I've spent a long time trying to figure out what it means.
但它的意思是，它们将会变得非常、非常、非常聪明，并且可能会以很多好的方式和很多坏的方式改变世界。
But what it means is they're gonna be really, really, really smart, and it's gonna change the world probably in a lot of good ways and a lot of bad ways.
是的。
And yeah.
我不确定你
I don't know if you
有没有听过这个版本的对话。
have this version of this conversation.
人们问我，他们的孩子是否应该学习编程。
People ask me if their kids should learn a code.
关于 Bayt 播客
Bayt 提供中文+原文双语音频和字幕，帮助你打破语言障碍，轻松听懂全球优质播客。