How Stanford Teaches AI-Native Engineering — Mihail Eric, Monaco · 斯坦福怎么教 AI 原生工程

00:00

Okay. This is a pretty special episode because we're celebrating one million subscribers on the AIU code AI engineer YouTube channel

好，这是一期非常特别的节目，因为我们在庆祝AIU code AI engineer YouTube频道达到一百万订阅者。

00:06

So go over there and subscribe if you haven't heard about it or the guest is Mehl Eric who is not only teaching the monosophage developer at Stanford

所以如果你还没听说过，去那边订阅吧，或者我们的嘉宾是Mehl Eric，他不仅是Stanford的monosophage开发者课程的老师，

00:14

Which is going viral right now for basically covering all the AI native development practices

这门课现在非常火，基本上涵盖了所有AI原生开发的实践，

00:19

But also he is head of AI and Monaco. So he's not only a teacher, but he's a practitioner and those of you coming to AIU code summit next week

而且他还是Monaco的AI负责人。所以他不仅是老师，也是实践者。那些下周来参加AIU code summit的人，

00:27

It's going to be the celebration of all the extreme progress that you see in encoding in the last year

这将是对过去一年你在编码领域看到的所有巨大进步的庆祝。

00:33

Or I'm excited about all the themes of fighting slop of system one models and of increasing your ambition for

我很期待所有的主题，包括对抗slop、system one模型，以及提升你的雄心，

00:39

Major rewrites of basically everything that you've seen in code

去对你在代码中看到的一切进行大规模重写。

00:43

There's never been in a more exciting time to be an AI engineer and with one million AI engineers

做AI工程师从来没有像现在这样令人兴奋，而且有一百万AI工程师。

00:48

I think it's just getting started. So let's get into the interview. Okay. We're in a studio with Mehl Eric

我觉得这才刚刚开始。那我们就进入采访吧。好的。我们现在在演播室里，和Mehl Eric在一起

00:54

junk professor Stanford and head of AI at Monaco. Welcome. Thank you. I'm glad to be here

junk，斯坦福大学教授，也是Monaco的AI负责人。欢迎。谢谢。很高兴来到这里

00:57

You caught our attention because of your course that you are refreshing for the third time

你引起了我们的注意，是因为你的课程，你正在第三次更新它

01:04

Second time yeah, the CS146 one for success the modern software developer at Stanford this particular

第二次，对，CS146那门课，面向现代软件开发者的成功之路，在斯坦福，特别是这一届

01:10

Generation of junior developers of junior engineers the people that are now entering the workforce

初级开发者、初级工程师，那些现在正在进入职场的人

01:14

Well, I think be the first kind of generation of that new shift

嗯，我觉得会是这种新转变的第一代人

01:17

And I looked at the syllabus and I was like is everything we cover

我看了教学大纲，我就在想，我们涵盖的内容是不是都是

01:21

So every now and then I'd like to invite people like yourself who like you cover broad range of topics

所以时不时我会想邀请像你这样的人，因为你涵盖了很广泛的话题

01:26

And then we can sort of go topic by topic and

然后我们可以大致按主题一个一个来

01:29

Give people the state of the art on each of these things, which I assume you're doing with all your guests anyway

给大家讲讲每个方向的最前沿进展，我猜你跟所有嘉宾也都在做这件事

01:34

Now let's the history of the course. What made you want to spring out CS146?

现在聊聊这门课的由来。是什么让你想开出CS146这门课的？

01:40

Yes, this is the second time it's being taught and I taught it for the first time last fall

是的，这是第二次开这门课，我第一次教是去年秋天

01:44

It's a fall 2025. This was pre like opus 4 5 4 6 days

那是2025年秋季。那还是在opus 4 5 4 6之前的日子

01:48

But the even before fall 2025 the course had been about like a year in the making

但甚至在2025年秋季之前，这门课就已经酝酿了大约一年

01:52

It's the genesis of the classified November of 2024 and I had sort of had this idea that we're seeing all this stuff happening and coding agents

它的起源是2024年11月的那次classified，我当时就有个想法，我们看到coding agents领域发生了这么多事情

01:59

And yes, they were an earlier version of the coding agents

是的，那时候还是coding agents的早期版本

02:01

But already it felt like there was something drastically changing in terms of how developers are doing their work

但当时就已经感觉开发者的工作方式正在发生剧烈的变化

02:05

How quickly we were getting work done all that fun stuff

我们完成工作的速度有多快，所有这些有意思的事情

02:08

And so at the time as it was all just scatter blog posts scattered like things you could find online

所以当时这些都只是零散的博客文章，散落在网上能找到的各种东西

02:14

If you were trying to really get into the methodology of how to do these things

如果你真的想深入了解做这些事情的方法论

02:17

And so I had this idea like hey, well, is there a way to kind of structure a curriculum around this?

所以我有了这个想法，嘿，有没有办法围绕这个来构建一个课程体系？

02:22

It's a little bit more formal had the idea and you know

稍微更正式一些，有了这个想法，你知道

02:24

I my own matter is Stanford and I always had this dream to go back and teach something and kind of pay it forward

我自己嘛，是Stanford的，我一直有个梦想就是回去教点东西，把这份东西传递下去

02:29

I've always been very deep in education. I took a chip who also taught a class

我一直对教育非常投入。我拉上了Chip，他也教过一门课

02:33

What are your friends? Yeah, chip she exactly spoke to her which she taught her I guess designing machine learning systems

你的朋友们怎么样？对，Chip，她确实跟她聊过，我猜是她教她的，关于设计机器学习系统

02:39

And so like ask her what that process was like and sort of you know

然后问问她那个过程是什么样的，你知道

02:42

Took a lot of inspiration from how she designed the curriculum and thought about you know getting somebody through this through the school

从她如何设计课程以及如何考虑让某个人通过这个课程中学到了很多灵感

02:47

Put it together spend about nine to ten months

把它整合起来花了大约九到十个月

02:50

This was now fall of 2024 through to like you know mid to late of 2025

这时候是2024年秋季到大概2025年中后期

02:55

Got it through the committee made a case for the class to exist

让它通过了委员会，为这门课的存在做了论证

02:59

You know had to put together proposed on everything and that took me to fall 2025

你知道得把所有东西的提案都整理好，那把我带到了2025年秋季

03:03

I wanted out of the first time which was very well received you know students were really interested in it

我第一轮就想开出来，反响非常好，你知道学生们真的很感兴趣

03:07

There was even I guess like a business inside article that was written about the class last year and

甚至我记得去年有一篇关于这门课的商业内幕文章，然后

03:13

Yeah, this year a lot of things changed. I would say really coding agents autonomous coding agents really

对，今年很多事都变了。我会说真正是coding agents，自主coding agents真的

03:20

If picked up in a way that that we had not seen before

以一种我们之前没见过的方式崛起了

03:23

So it felt like it was time to do this again

所以感觉是时候再做一次了

03:26

refresh things

刷新一下

03:28

And then here we are

然后我们就到了这里

03:30

Meanwhile, you're also hit of AI at Monaco. Yeah, which is also a very busy startup. Yes. Yes

同时，你也是Monaco的AI负责人。对，那也是一家非常忙碌的创业公司。是的。是的

03:35

Yeah, we are we're we're in the airways right now and and having a fun time doing it

对，我们现在正在筹备中，做得很开心

03:39

Yeah, so like I do like that

是的，所以我还挺喜欢那样的

03:41

You also I guess a practitioner you guys are obviously implementing all these stuff at Monaco. Yeah, any I guess

你也是，我猜你算是个实践者，你们显然在Monaco把这些东西都落地了。是的，任何我猜

03:47

Principles or observations about just what you have changed from last year to this year at a high level

关于你从去年到今年在高层次上改变了什么的原则或观察

03:53

Yeah, yeah, I mean a ton. I think that

是的，是的，我是说，太多了。我觉得

03:56

When I reflect on what last year's version of the class was I mean coding agents had been around

当我回顾去年那个版本的课程时，我是说coding agent已经存在了

04:01

But we were still in kind of that phase where companies were emerging that we're tackling every single part of like the IS DLC

但我们还处在那种阶段，就是公司在涌现，各自在攻克IS DLC的每一个环节

04:08

There was like a coding agent for a code review. There was a coding agent for you know spec design

有专门做code review的coding agent。有专门做spec design的coding agent

04:12

There was a coding agent for SRE and you know some of those of course some of those companies still exist

有专门做SRE的coding agent，你知道其中一些当然，其中一些公司现在还在

04:16

But it felt like last year

但感觉就像去年

04:17

It was a lot more about a broad tour of here's everywhere you can use coding agents and

更多的是广泛地带你浏览一遍，这里是你可以在哪些地方使用coding agents，以及

04:23

In a way a lot of the curriculum last year was was geared toward

某种程度上去年很多课程内容都是围绕

04:27

Exposure and this this kind of survey this kind of like Odyssey of sorts through a hey everything that you knew about the IS DLC

接触和这种概览式的、有点像奥德赛之旅一样的，带你经历一遍——嘿，你之前所知道的关于IS DLC的一切

04:33

Is now changing because of AI

现在都因为AI而改变了

04:35

This year I think there's some principles that are solidifying right the idea of like very careful context engineering

今年我觉得有一些原则正在固化下来，对吧，比如非常精细的context engineering这个理念

04:40

The idea of like really thinking through so like design and architecture as a way of handling really complex tasks

比如真正深入思考，像设计和架构作为处理非常复杂任务的方式

04:46

And so the nature of my classes

所以我的课程的性质

04:47

It's not intended for the like beginner programmer

这不是给那种初学者程序员准备的

04:51

It is not a it's not a replacement for like traditional computer science in the sense of like algorithms

它并不是传统计算机科学那种意义上的替代品，比如算法之类的

04:56

And if in fact a lot of those classes are pre-rex to what I'm doing here

事实上很多那些课程是我这里所做内容的先修课

04:59

But I do know that if you have a good foundation going now toward

但我确实知道，如果你现在有良好的基础，朝着

05:03

Applying that and and sort of using those principles in in production environments in like real world

应用它、把那些原则用在生产环境里、用在真实世界中

05:07

That's something that I think is is now a place where you know

这是我认为现在你知道的一个地方

05:10

At Monaco we do with massive code bases. You know

在Monaco我们处理的是大规模代码库。你知道

05:13

I help a lot of companies also with thinking through their massive code base designs and things like that

我也帮助很多公司梳理他们的大规模代码库设计和类似的事情

05:16

And so really it's more of a practitioner's course for once the students have graduated or now entering the workforce

所以实际上这更像是一门实践者课程，针对的是学生毕业之后或者即将进入职场的时候

05:22

How to be the most productive members of the workforce

如何成为职场中最有生产力的成员

05:25

Yeah, you said you're passionate about education. I feel like students in that

对，你说你对教育很有热情。我觉得处于那个阶段的学生

05:32

Phase of being in college are in a pretty interesting spot

在大学那个阶段处于一个挺有意思的位置

05:36

Any advice and any takes should people study computer science right now

有什么建议和看法，人们现在应该学计算机科学吗

05:40

Is cloud code and codex all we need like I want to you know hear your thoughts. Yeah, I

是不是有cloud code和codex就够了，我想听听你的想法。对，我

05:46

You know, there's a lot of anxiety from students as you can probably imagine and and I I think a lot of what I'm trying to

你知道，学生们有很多焦虑，你能想象得到，我觉得我这次想传达的很多内容是

05:53

Communicate this time around is that things I do a lot of it still is here and there was anxiety last year as well

我做的很多事情其实还在，去年也有焦虑

05:58

But I want to make the case that

但我想提出一个观点

06:01

a well-trained

一个训练有素的

06:03

AI native engineer and I know that's like a busy word, but let's

AI native engineer，我知道这听起来像个很热门的词，但让我们

06:06

Well, well, well, well, we'll dive into maybe what asks are the that are most interesting a like well-trained and native engineer can be some of the most

好吧好吧好吧，我们会深入探讨一下也许最有趣的那些要求是什么，一个训练有素的native engineer可以成为一些最

06:13

You know the most useful assets like any company around and so

你知道的，任何公司周围最有用的资产之一，所以

06:17

And you know, this is the first week of class actually and like week one, you know first lecture

而且你知道，这实际上是开学的第一周，就像第一周，你知道第一堂课

06:21

I mean, hey, I really believe that every company needs you

我的意思是，嘿，我真的相信每家公司都需要你

06:24

But the floor is much higher, right? So you're entering as like your first job at a school

但门槛要高得多，对吧？所以你是作为第一份工作进入学校的

06:29

But it's not sufficient to just be like a you know an L3 and like a Google term if you're gonna use like that the kind of leveling system there

但仅仅达到L3或者谷歌那种级别是不够的，如果你要用那种等级体系的话

06:35

It's like you have to generally be better than you know almost at the senior level going to this

你得普遍比几乎senior级别的人还要强才能走到这一步

06:39

So like what what can we how do we get there more quickly, right? And so

所以我们怎么才能更快地达到那里，对吧？然后

06:44

I'm trying to say that there is still a core bit of like

我想说的是，仍然有一部分核心的

06:47

Foundational software engineering that is important to have

基础软件工程是很重要的

06:49

Some of these core values of sort of how soft rhythms are built and the trade-offs of those systems

一些关于软件系统如何构建的核心价值观以及这些系统的权衡

06:54

But then beyond that we go and

但除此之外我们就去

06:56

You know, that's where I think you as the humans are still be valuable rather than and the rest of the implementation details

你知道，这就是我认为作为人类仍然有价值的地方，而不是剩下的那些实现细节

07:00

You can kind of defer more often

你可以更经常地推一推

07:03

Yeah, there's something I've said before which is almost like the controversial

对，我之前说过一个几乎算是有争议的话

07:07

I it's like the kind of thing you say to be spicy, but I almost like I do generally kind of believe it

就是那种你为了显得犀利才说的话，但我其实大体上是真的相信

07:13

Which is most students are cooked some of them are gonna be super valuable. I don't think it's a very popular thing

那就是大多数学生是完蛋了，其中有些会变得超级有价值。我不觉得这是个很受欢迎的说法

07:20

San Francisco some of the kids are gonna be fine, right? There's different Stanford

在San Francisco有些孩子会没事的，对吧？有各种不同的Stanford

07:23

But like what happens to like the rest of America

但美国其他地方的人会怎样呢

07:27

Right, right? That is an awful question

对对，这问题太糟糕了

07:29

Yeah, I feel like a lot of education is slow to move like I'm even curious in how

是啊，我觉得很多教育领域行动都很慢，我甚至很好奇

07:35

How do you determine where do you put the end date to curriculum in a course on oh my god

你怎么确定一门课的课程内容该在哪里截止啊，我的天

07:40

You're teaching a Stanford modern software like that's changing every you cheat

你在教斯坦福现代软件之类的课，那玩意儿每时每刻都在变，你作弊

07:43

You just get the people from the

你就找那些

07:46

The coding agents to come and speak right? Oh, yeah, like yeah, haven't either yeah, that's a funny

coding agent的人来讲课对吧？哦对，是啊，也没请过，是吧，那挺好笑的

07:52

Well, yeah, I mean, you know, that was okay, so so on on the guest lectures point

嗯对，我是说，你知道，那还行，所以关于客座讲座那点

07:56

I think that was I mean that was also last year

我觉得那个，我是说那个也是去年的事

07:58

That was also the case so so that was by design because I

去年也是这种情况，所以那是刻意安排的，因为我

08:02

Look, I came in right and I'm saying I'm a practitioner as you point out the way it's like I'm a practitioner

你看，我来了之后，我说我是个实践者，正如你指出的那样，就像我是个实践者

08:07

I do this thing on my you know my day-to-day design these systems

我平时做这件事，你知道的，我日常就是设计这些系统

08:10

You know we build these things in Monaco and have seen that

你知道我们在Monaco造这些东西，而且已经看到了

08:13

many times before and so

很多次之前，所以

08:15

That that being said, I don't want to claim that I have all the answers today and that would be that would be like

话虽如此，我不想声称我今天就有所有的答案，那样的话就太

08:20

Very you know be a lot of hubris to make that claim. So

说这话确实有点太自大了。所以

08:23

the structure of the class was always

这门课的结构一直都是

08:26

Here's one practitioner's sort of aggregated set of insights

这是一位从业者汇总的一些见解

08:30

From working in the field from my talking a lot of people in the field

08:33

But then there's people that are like literally on the front here, you know, same things even beyond that and invite them to sort of put their

但还有那些真正站在前沿的人，你知道，甚至超越这些的事情，邀请他们把他们的

08:41

Their proposals forward their view of the world with the world forward and that's why the guest

他们的提案提出来，他们对世界的看法，推动世界前进，这就是为什么嘉宾

08:45

I think it's really impactful for students to see

我觉得这对学生来说真的很有影响力，能看到

08:48

big names

大名字

08:50

People actually doing it like here from them directly be able to like ask them of course, you know

真正在做这件事的人，直接从他们那里，能够去问他们，当然，你知道

08:54

They want to intern they want to get jobs. It's it's good to have guests like yours. No, I think what he's saying so

他们想实习，他们想找工作。有像你这样的嘉宾是很好的。不，我觉得他在说的是

09:00

Okay, I'll sort of rephrase and

好吧，我换个说法

09:02

No, so okay there, you know, we've also interviewed Anj

不，所以好吧，在那里，你知道，我们也采访了Anj

09:05

Who teaches the Coachella course, right? Which is like here's all the names like it's only the names right

Coachella课程是谁教的，对吧？就像这里列的全是名字，就只有名字对吧

09:11

Whereas like he's actually also producing you know, here's called code into 100 lines of code

而实际上他也在产出，你知道，这里叫把代码变成100行代码

09:15

Right, which is no one's telling you to do that no one's going to come speak to do do do do do do that

对吧，没人让你那么做，没人会过来跟你说做做做做做做那个

09:19

But like yeah that when you want to teach fundamentals in a framework that you can use in working understanding

但就像，是的，当你想在一个框架里教基础知识，让你能在工作中理解运用

09:24

This is the kind of thing you do that's right. That's right. Yeah. Yeah, exactly

这就是你会做的事，对。对。是的。是的，没错

09:27

And and I but you think that there's a balance for for this kind of material to to your point is that there is a you can't

而且我，但你觉得这类材料是有个平衡的，说到你的点就是，有一个你不能

09:32

It's not

它不是

09:33

100% solidified. I mean by by design I mean by design

100%固化的。我是说从设计上，我是说从设计上

09:36

It's a feature and maybe a bug of some some kind but

这是特性，也许某种程度上的bug，但

09:39

A big portion of the curriculum did change from last year, right?

课程内容很大一部分确实和去年不一样了，对吧？

09:42

And so that there's like a dynamism to the whole development of the curriculum

所以整个课程的开发有一种动态性

09:45

But I'm trying to at least put forward that there are some things that I believe are kind of sticking around and yes

但我想至少提出一点，有些东西我认为是会留下来的，是的

09:49

Prompting coding agents. They've gotten smarter

Prompting coding agents。它们变得更聪明了

09:51

What does that mean in terms of the new ways of dealing with them? But then still

那在应对它们的新方式上这意味着什么？但还是

09:55

Boris is also going to know something that I don't know, right? So let's let's have let's have him also come to the table and share his learnings

Boris也会知道一些我不知道的东西，对吧？所以让让他也来桌上分享一下他的心得

10:01

One of the fun things I noticed this time around is a good bit of the portion is on open source contributions

这次我注意到的一个有趣的点是，相当一部分内容是关于open source贡献的

10:09

Yeah, what's the thinking behind that you have people?

对，你们是怎么考虑的？

10:12

Yeah, yeah, this is um this is one of the things that I'm honestly kind of most proud of and most excited about

对对，这是，嗯，这是我老实说最自豪也最兴奋的事情之一

10:18

I mean this

我是说这个

10:20

About to be week two of the class

马上就是课程的第二周了

10:21

Yeah, so this is this is the very gracious open source partners that I spent the greater part of this summer

对，所以这是非常慷慨的开源合作伙伴，我花了这个夏天大部分时间

10:26

Just kind of talking to and getting them kind of excited about and and interested in sort of this offering

就是跟他们聊，让他们对这个课程感到兴奋和感兴趣

10:32

Basically the idea was so last year when I taught at the class

基本上想法是这样的，去年我教这门课的时候

10:34

We had like nine assignments and then we had like a capstone open-ended project, right?

我们有大概九个作业，然后有一个开放式的 capstone 项目，对吧？

10:38

The nine assignments were basically me sort of like coming up with these pedagogical exercises like oh

那九个作业基本上就是我想出的一些教学练习，比如说

10:43

Here's a repo like use an AI code review tool to you know

这有个repo，用AI代码审查工具去你知道

10:46

Post-to code review and then see what why is it good? Why is it not good? So

提交代码审查，然后看看为什么好？为什么不好？所以

10:50

relatively pedagogical honestly pretty easy

说实话挺教学性质的，挺简单的

10:52

I think a lot of folks they could just like one shot it and I thought okay

我觉得很多人他们直接一次就能搞定，然后我想好吧

10:55

This is not that helpful at this point agents are round

到这个时候这已经不太有帮助了，agents已经到处都是了

10:58

I mean they they have every students that I was surveyed, you know on Tuesday

我是说他们，我周二调查的每一个学生

11:03

Not a single person really types by by hand anymore. They're like everyone's using coding agents

没有一个人还真的手打代码了。他们都是，每个人都在用coding agents

11:06

There's no there's and this is you know like all sophomores juniors at Stanford. Do you do survey them? What's the market share?

没有，这个，你知道，就像斯坦福所有大二大三的学生。你会去调查他们吗？市场份额是多少？

11:12

Was that like what what was it used? Yeah, okay. I haven't yet, but that's a good cause. Yeah. Yeah

那是像什么，用来做什么的？对，好的。我还没做，但这是个好主意。对对

11:18

I won't give you the spicy Stanford prefers

我不会给你斯坦福偏好的那个劲爆版本

11:21

No, no, what I think through like the university. I think they have some the edge like most of the the providers have like an educational program

不不，我仔细想了一下，通过大学。我觉得他们有一些优势，就像大多数供应商都有教育项目

11:27

Right, so yeah, they're they're very generous on

对，所以他们对学生套餐非常慷慨

11:31

plans for students right right, so I think by default even before the class. There's like you know

对对，所以我觉得默认情况下甚至在开课之前。就有像你知道的

11:35

Clive license. I imagine cursor license. There's like probably a cognition license as well

Clive license。我猜有cursor license。大概也有cognition license

11:40

I don't know you might know better than I do

我不知道你可能比我更了解

11:42

But what I was able to also just to crew on top of that was hey like if I don't want people to be born on like by like the $20 a month

但我还能在此基础上做的是，嘿，如果我不想让人们一出生就背负每月20美元的费用

11:48

Subscription like I want you guys to go if you want to like token max something just to see what's possible

就像订阅一样，我希望你们去，如果你们想token max某个东西，就只是为了看看什么是可能的

11:53

Then then you should and so and I'll come back to the the partners

然后然后你应该，所以然后我会回到合作伙伴那里

11:55

But the the additional thing I was able to to crew through a set of targeted folks was like additional credits for students

但我额外能做到的事情是，通过一组定向联系的人，为学生争取到了额外的学分

12:00

So like factories also getting involved the cognition folks are getting involved. You open it as well

所以像工厂也参与进来了，认知那边的人也在参与。你也把它开放了

12:04

But anyway, so so for the open source partners. I sort of like was like okay

但不管怎样，所以对于开源合作伙伴。我有点像是说好吧

12:08

One of the things I'm most concerned about when it comes to educate the next kind of generation of students is that

我最担心的一件事，就是在教育下一代学生的时候，

12:14

If everyone is saying kind of anticipates point like they're cooked like you know what's gonna happen to them

如果大家都在说，某种程度上已经预判到了这一点，就像他们完蛋了，你知道他们会遭遇什么

12:19

There there's almost like this sense of like them being left behind and I think that mentorship is like a really important kind of peace

就有一种他们被落下的感觉，我觉得 mentorship 其实是一种很重要的平和

12:25

Of just up leveling everyone and making sure that there isn't this like massive gap that folds that that occurs between

就是让所有人都提升上来，确保不会出现那种巨大的鸿沟，那种

12:30

You know the senior people that have already been in the industry before coding agents and know what's up

你知道，那些在 coding agents 出现之前就已经在行业里的 senior 的人，他们知道怎么回事

12:34

And they adopt coding agents and they become amazing and then the juniors that are just like okay now. What do we do?

他们采用了 coding agents，变得很厉害，然后那些 juniors 就只是，好吧，现在我们该怎么办？

12:39

So in a sense one of the the kind of goals was hey coding agents are really good

所以某种意义上，其中一个目标就是，嘿，coding agents 真的很厉害

12:44

I want to put you guys in situations where you're dealing with complex code bases large code bases at scale

我想把你们放到那些你们要处理复杂代码库、大规模大型代码库的情境里

12:49

We're gonna work with someone like what I think are among the best like a native repos

我们会和一些人合作，比如我认为是最好的一批，像 native repos

12:53

You know that I'll do AI focus things or either coding agents themselves or building tools in the service of agents

你知道，我会做 AI 相关的事情，要么是 coding agents 本身，要么是构建服务于 agents 的工具

12:58

You know in the modern world. I want you guys to get involved with these projects

你知道在现代世界里，我希望你们参与到这些项目中去

13:01

I want part of your assignments to be actually contributing to these projects in the course of being all those projects

我希望你们的一部分作业实际上就是在参与这些项目的过程中对这些项目做出贡献

13:06

You actually be talking to the maintainers. They will be reviewing your code

你们实际上会跟maintainers交流，他们会review你们的代码

13:10

They'll be helping you build mental models of like good software in that way

他们会以这种方式帮助你们建立对好的软件的mental model

13:13

There's effectively this like passing on of like the guard of true good software design in a way that you know

这实际上就像是一种传承，把真正好的软件设计的guard传递下去

13:18

I don't think most students get the opportunity to do and then there's the benefit of course

我觉得大多数学生没有机会做到这一点，然后当然还有好处

13:22

They like I'm sure some of them will get you know poached and picked up as part of the

我相信他们中有些人会因为作业做得好而被挖走

13:26

Doing well knows assignments. Yeah, one of the interesting things you know

你知道，有件有意思的事

13:30

For what it's worth if I'm gonna get a bunch of hate if I don't say it is

顺便说一句，如果我不说的话肯定会招来一堆骂声

13:33

I think you say cooked and versus cracked for dramatic effect

我觉得你说cooked和cracked是为了制造戏剧效果

13:38

But you know like it's just nice to state extremes

但你知道，把极端情况说出来就是挺好的

13:42

Uh sure. Yeah, so so maybe we should go into the sort of detail of the curriculum

呃，当然。对，所以也许我们应该深入讲讲课程的具体细节

13:46

Is there anything else you want to explore at the top level? God, I mean like that. That's uh very healthy

在顶层还有什么你想探讨的吗？天哪，我是说那样。那呃非常健康

13:51

Yeah, although that's there. Okay. I have high level questions about students

对，虽然那是在那儿。好吧。我有一些高层面的问题关于学生

13:55

using

使用

13:57

Tools best practices are you pro it like

工具的最佳实践，你是支持它像

14:00

Individual tutors um, I'm assuming this will also be picked up by students

一对一的导师，嗯，我猜这也会被学生们用上

14:05

So you know before we dive straight into the course just anything you have guidance on there for

所以在我们直接进入课程之前，你有什么相关的建议吗

14:10

Like in terms of what you said you're passionate about education right like yeah, there's also a stigma of

比如就你刚才说的你对教育很有热情对吧，是的，还有一种污名化的看法

14:16

Once I did how do teachers professors use it? Yeah, yeah, and then how can students best use these tools

一旦我做了，老师和教授们怎么用它？对对，然后学生们怎样才能最好地使用这些工具

14:21

That's not just throw your assignments in

不是简单地把你的作业扔进去

14:24

It can kind of one shot it how do you still learn like if you have anything there? Yeah, yeah

它一下子就能搞定，那你还怎么学习，如果你有什么建议的话？对对

14:29

And I think that there's I still have some assignments thrown in there

而且我觉得还有一些作业被扔进去了

14:32

Which I think are not intended to be the hardest things in the world

我觉得那些本来也不是要设计成世界上最难的东西

14:34

But are meant to be more about concepts like more conceptual things like you know

但更多是关于概念性的东西，你知道的

14:38

What does a skill mean sort of like the maybe more of the the tactical of some of these tools skills like and you know

一个skill到底意味着什么，可能更多是这些工具skills中偏战术层面的东西，你知道的

14:45

Agent-ready code basis like agents MDs things like this

Agent-ready code basis，比如agents MDs之类的

14:47

So there's a certain amount of just hey if you haven't heard these terms kind of get up to speed on them

所以有一定程度上就是，嘿，如果你还没听过这些术语，那就先跟上一下

14:51

I do think most of the learning though in terms of the kind of putting the tools to the test will happen

不过我确实认为大部分学习，就是把工具付诸实践的那种，会发生在

14:56

Both in the OSS and then also in the like the final project where it's like very open-ended

既在OSS里，也在最后的project里，那个project非常开放式

15:00

And they'll have to just kind of explore something interesting

他们就得去探索一些有趣的东西

15:03

Cool, yeah

酷，好的

15:05

So so maybe oh yeah, I just want to talk that vote. Let's go through one two three four four two ten

所以所以可能哦对我只是想聊聊那个投票我们来过一遍一二三四四二十

15:09

Yeah, and then we'll dive in deeper. So like breath first basically sure sure. Yeah

对然后我们会深入进去所以就像先呼吸一下基本上当然当然对

15:15

So we start with the internals of coding agents

所以我们从coding agent的内部机制开始

15:17

You teach people how to write their own clock clone. It's very helpful. That is not open source

你教人们如何写自己的clock clone 这很有帮助那不是开源的

15:23

And then yeah, yeah, so week one. It's all internals of harnesses, you know kind of build a cone agent from scratch

然后对对所以第一周全是harness的内部机制你知道从零构建一个cone agent

15:30

Week two is more

第二周更多是

15:31

Okay, now assume that there's how do we start thinking about like context engineering at scale

好现在假设有我们怎么开始思考大规模context engineering

15:36

So beyond just vanilla prompting and then I think coupled with that is just

所以不只是vanilla prompting 然后我觉得与之配套的就是

15:41

In in doing that thinking about like tool design MCP you know people I think for a while and see who's getting a lot of hate

在做这件事的时候，思考一下像工具设计、MCP这些，你知道，人们我觉得有一段时间了，看看谁被骂得很惨

15:46

I'm still pro MCP. I see it all the time

我仍然支持MCP。我到处都能看到它

15:48

And so I think that there's at this point it's a standard that's existed

所以我觉得在这一点上，它已经是一个存在的标准

15:51

And then I think it will continue to exist

然后我觉得它会继续存在下去

15:52

So you know talking about that effective tool design

所以你知道，谈到那个有效的工具设计

15:55

It's week two

这是第二周

15:56

We've three I would say is more about just like I

我们第三周我会说更多是关于就像我

15:59

Skills and like what does it mean to like extend beyond simple

技能，以及像超越简单的东西意味着什么

16:02

Single-purpose tools to more like procedures and that's where we'll talk about skills and you know that kind of starts leading into like the automation of agents

从单一用途的工具，更多转向类似流程的东西，这就是我们会讲 skills 的地方，你知道那大概就开始引向 agent 的自动化了

16:10

Week four is is a lot of just more of like the foundational pieces of how can you

第四周主要是更多基础性的内容，讲你怎么去

16:16

Customize an agent?

定制一个 agent？

16:17

This is things like agent config files a la agents mb cloud md hooks

这包括 agent config files，比如 agents md、cloud md、hooks

16:22

You know kind of sub agents like there's a lot of the mechanisms that exist in modern harnesses

你知道，还有 sub agents 之类的，现代 harness 里存在的很多机制

16:26

Which you can use for different purposes. I want to expose students to that

你可以用于不同的目的。我想让学生接触到这些

16:30

Agent ready code bases. This is probably one of my favorite topics. It's like

Agent ready code bases。这大概是我最喜欢的话题之一。就是

16:34

And this is especially relevant for folks that are going into the industry and thinking about code at scale is

这对那些要进入行业、思考规模化代码的人来说尤其相关

16:39

You know essentially what does it mean to have a repo that's ready to be just to consumed effectively by an agent

你知道基本上，拥有一个准备好被agent高效使用的repo意味着什么

16:45

And so thinking a lot about you know the trade-offs of designing of kind of service of the boundaries, but also docs and and verifiability and things of that nature

所以想了很多关于设计服务边界时的权衡，还有文档和可验证性之类的东西

16:53

So that's week five

所以那是第五周

16:55

Yeah, week six sort of agente code review and here I think what is interesting is thinking also kind of extending from week five

对，第六周大概是agent代码审查，这里我觉得有意思的是，也想着从第五周延伸出去

17:02

Week five around code ace like agent ready code bases kind of week six is more about like

第五周围绕代码ace，就是agent就绪的代码库，第六周更多是关于

17:07

Okay

好

17:08

In order to my code review is now becoming a bottleneck which I don't know that students unless they've you know have done a lot of

为了我的代码审查现在变成了一个瓶颈，我不确定学生们除非他们已经做了很多

17:14

Some of them are master's students they've been in the industry, but a lot of them have not seen this

有些人是硕士生，他们在行业里待过，但很多人没见过这个

17:17

But for most people that are using coagions today

但对于今天大多数使用coagions的人来说

17:20

We're all experiencing something of a new bottleneck in code reviews

我们都在经历代码审查中的一个新瓶颈

17:22

So how do we unblock that or sort of unbottle the bottleneck a little bit

那么我们如何解除这个瓶颈，或者说稍微疏通一下这个瓶颈

17:27

What mechanisms are there there and so verifiability and you know things of that nature and how we can extend that into the code review process

有哪些机制，比如可验证性以及诸如此类的东西，以及我们如何将其扩展到代码审查流程中

17:33

Week seven security, I mean obviously I think there was a point

第七周安全，我的意思是显然我认为有一个点

17:38

So this was in last year's class as well and this this I wanted to reinforce

所以这在去年的课程中也有，这个我想强调一下

17:41

I mean with any autonomous system taking a lot of actions. We're seeing a lot of this year, especially a lot of things getting hacked

我的意思是，对于任何采取大量行动的自主系统。我们今年看到了很多这种情况，尤其是很多事物被黑客攻击

17:48

You know in not least of which are just some of the the core pieces of software from from for sale to you know

你知道，尤其是从for sale到你知道的一些核心软件

17:54

Gateways all in gateways entirely. So I want to spend some time just like talking about what new tack vectors exist

网关全都在网关里。所以我想花点时间聊聊有哪些新的攻击向量存在

17:59

Exist in security and and you know how we can mitigate some of them

存在于安全领域，以及你知道我们如何能缓解其中一些

18:05

Week eight is

第八周是

18:07

Background agents. So this is what I want to start touching on here is

后台agents。所以这是我想要开始涉及的内容

18:11

you know, how do you like?

你知道，你如何

18:14

Basically run agents in many different like product service areas

基本上在很多不同的产品服务领域运行agents

18:17

Not just in what was traditionally an ID but also having things that are like based on triggers in your code base and in your repo at

不仅仅是在传统意义上的ID里，还要有一些基于你代码库和repo中触发器的东西

18:23

You know during CICD

你知道，在CICD期间

18:25

Things that get triggered even when you know based on events things like SRE style bots

那些即使你基于事件知道也会被触发的东西，比如SRE风格的机器人

18:31

How do you even build a system like I think this past year was when a lot of companies started releasing their

你怎么才能构建这样一个系统呢？我觉得过去这一年正是很多公司开始发布他们的

18:37

Hosts like ramp inspect. There was Spotify has their own version of this

像ramp inspect这样的主持人。Spotify有他们自己的版本

18:41

You know, I think striped minions so all these folks were doing really cool write-ups of like their internal info

你知道吧，我觉得striped minions，所以这些人都在写很酷的东西，比如他们的内部信息

18:45

They were building to do this. I want to touch on that. I'll give people exposed to what that is

他们当时就是在朝着这个方向构建的。我想谈谈这一点。我会让人们对那是什么有所了解

18:50

Yeah, this is um

是的，这个是嗯

18:52

Yeah, this one is a more on like what happens when you're in in a team in in the modern world and sort of like the core pieces of of team level

对，这一期更多是讲在现代世界里，当你在一个团队中时会发生什么，以及团队层面的核心要素之类的。

19:00

Infrastructure so everything from

基础设施，所以从

19:02

LMGateways to you know MCP portals which is like MCP at scale and then just also how to think about the evolving

LMGateways，还有你们知道的MCP portals，就像是MCP的规模化版本，然后还有怎么去思考那些不断演变的

19:10

Rolls I would say so this is maybe a little bit like a soft thing. It's like what what what is now?

角色，我会这么说，所以这可能有点像是一个比较软性的东西。就像是现在什么是什么？

19:14

You know and the the vanilla maybe like not super nuanced thing to say is like everyone's a builder

你知道，那种比较vanilla的、可能不太细致的说法就是每个人都是builder

19:18

But I want to like go a little bit deeper beyond that and like start like what thing is actually getting subsumed

但我想稍微再深入一点，超越那个层面，开始想什么东西实际上正在被吸纳

19:22

What what rules are actually either evolving or merging that didn't once exist

哪些角色实际上要么在演变要么在合并，而这些以前是不存在的

19:26

What should students think about in terms of defensible skill set that's maybe wasn't true before

学生应该从什么角度去思考有防御性的技能组合，那些以前可能不成立的东西

19:31

And so a lot more like team dynamic type type considerations

然后还有很多更像是团队动态类型的考量

19:34

And then yeah, this is a they're all kind of builds up toward toward the factory which I think is top of mine for now

然后是的，这是一个它们都在朝着factory方向构建的东西，我觉得那是目前我最关注的

19:40

for a lot of folks today

对今天很多人来说

19:43

Once we're we're building factories where do humans still exist and all that what is that you know

一旦我们在建工厂，人类还存在于哪里，所有这些，那是什么，你知道

19:47

We're the new choke points that are that are intentionally there around around human involvement things like intent things like

我们是新的瓶颈，那些是故意设置的，围绕人类的参与，比如意图，比如

19:54

Setting up a verify bill systems

建立可验证的系统

19:56

And what happens in the limit when maybe a 89% of code is fully driven by just agents where we got we kind of fill into all that so

以及在极限情况下会发生什么，也许89%的代码完全由agents驱动，我们有点陷入所有这些之中，所以

20:04

Yeah, nice solid 10 weeks

是的，不错的扎实的10周

20:07

I was gonna say like you know some of these

我正想说，你知道其中一些

20:10

Are more established

是更成熟的

20:12

Patterns than others and I think people need to know that like not all of this is settled science true

有些模式比其他模式更成熟，我觉得人们需要知道，并不是所有这些都已经是定论，没错

20:19

You know, but it is everything I'll endorse everything as like things that people talk about and you should know

你知道，但这就是全部，我会把人们谈论的所有东西都认可为你应该了解的内容

20:24

You just we know the open problems. We may not know the answers

我们只是知道开放问题是什么。我们可能不知道答案

20:27

That's absolutely yeah, yeah, I mean think even um, I mean

这绝对是，对对，我是说想想甚至，我是说

20:30

You probably have your own version of this I'd be curious to hear what you think is is still not as defensible

你可能对这个问题有自己的版本，我很好奇想听听你觉得哪些仍然不那么站得住脚

20:34

But are not invincible, but not as a set in stone

但不是不可战胜的，只是没那么板上钉钉

20:36

But you know, even like the the software factory are we talking about like dark factories?

但你知道，甚至像软件工厂，我们说的是黑灯工厂吗？

20:38

Are we talking about like these sort of half lights on factories like I think that even there's a spectrum there

我们说的是那种半亮灯的工厂吗，我觉得甚至那里也有一个光谱

20:44

Yeah, in there, you know

对，就在那里面，你知道

20:45

There's the people that are on the on the very frontier like like the warps and the factories that are that are like building factories and selling them

有一些人处在最前沿，比如那些warp和工厂，他们在建工厂然后卖

20:51

Services and you know, even they will only claim like

服务，而且你知道，他们自己最多也就声称

20:54

30 some percent if even that of like fully you know, gently drew contributions

百分之三十几，甚至可能还不到，就是完全你知道，温和地做出的贡献

20:58

So yeah, what does that mean in the limit? It's it is hard to say and there's people who

所以是的，这在极限情况下意味着什么？很难说，而且有些人

21:02

Went very very hard dog factory and then came out saying I was wrong guys. Yeah, right? So like

非常非常激进地搞dog factory，然后出来说“我错了各位”。对吧？所以就像

21:07

We covered that at on the ii

我们在ii上覆盖过那个

21:10

Just be aware that you know, not everyone they they probably believe the things they sing at the time

只要意识到，你知道，不是每个人——他们当时可能真的相信他们唱的那些东西

21:15

Yeah, but they have the right to change their mind and they do and even Boris recently tweeted that yeah

是啊，但他们有权改变主意，而且他们确实会变，甚至Boris最近还发推说，是啊

21:20

I am often wrong. Yeah, like and like everyone is like because he created cloud code

我经常犯错。是啊，就像每个人都这样，因为他创造了cloud code

21:25

It's like treating his word as gospel and I'm like no one sometimes he disagrees in himself. Yeah

就像把他的话当成圣旨一样，而我说没人有时候他自己都会跟自己意见不一致。是啊

21:30

And that's okay. I mean that's like this isn't politics, right?

那没关系。我是说这又不是政治，对吧？

21:33

This is where we're still scientists for engineers were we're subject to

这是我们仍然是科学家、是工程师的领域，我们受制于

21:36

I don't claim to be dogmatic, but you know, I just want people to experiment

我不声称自己教条，但你知道，我只是希望大家去实验

21:39

I think that's like a principle as well. Yes to a principles earlier just like experimentation

我觉得这也是一条原则。是的，之前提到的原则就是实验

21:43

I mean like you just have to be ready to and oftentimes when I sort of like give sort of these lectures or talk to people out the stuff

我是说，你就是得准备好，而且很多时候当我做这些讲座或者跟人聊这些东西的时候

21:49

Like how do I there's so much going on? How do I stand top of this?

比如我该怎么应对，有那么多事情在发生？我怎么才能跟上这一切？

21:51

And it's like you know, there's something similar just like experiment for 10 minutes a day

就好像你知道的，有类似的方法，就像每天做10分钟的实验

21:54

Like just spending some 10 minutes doing something that you didn't do before if that's a traditional way that you were solving a problem

就是花10分钟做一些你以前没做过的事情，如果那是你以前解决问题的传统方式

21:59

There's now a new tool you want to try or a new like framework or something and just spending maybe even doing the exact same problem

现在有一个新工具你想试试，或者一个新的框架之类的，就花点时间，甚至可以做完全相同的问题

22:05

Just with that new kind of thing is is alone will cover most of the gap

只是用那个新的方式来做，光是这样就能弥补大部分差距

22:09

Yeah, I will call out so some of your

是的，我要提一下，比如你们的

22:12

For example, you have cognition on the week that you're talking about code review

举个例子，你们那周有cognition，你们聊的是code review

22:16

I actually did work with scientists on devin review. Yeah, yeah, pretty good

我实际上和科学家们一起做过devin review的工作。是的，是的，相当不错

22:19

I would say like you know

我想说你知道的

22:20

It's hard to benchmark these things. There is a review bench. The one uses it

这些东西很难做基准测试。有一个review bench。用的那个

22:26

But like qualitatively they're really pretty good. Well, we use you guys. That's also really yeah

但定性来说它们真的挺不错的。嗯，我们用你们家的。那也确实是啊

22:30

So so Monica does use devin. Oh, and that's that's I mean

所以Monica确实在用devin。哦，那那我是说

22:33

And so I was actually spoke last year and so I'm your size is great

所以我去年实际上讲过，所以你们的规模很棒

22:37

It's just super super smart

它就是超级超级聪明

22:38

But um, yeah, we we use you guys and so I was like okay among among the providers of kind of code review tools that we and we sampled a lot of them

但是嗯，是的，我们用你们家的，所以我就想好吧，在我们采样过的很多代码审查工具提供商中

22:44

We actually did kind of converge on devin for yeah, but silence actually more models side these days and then actually

我们实际上确实最终选择了devin，是的，但silence现在其实更多在模型这边，然后实际上

22:50

It's walled in that you want yeah, yeah, yeah, yeah, yeah, so yeah, I was talking walled in as well

就是封闭的，你想要，对对对对对，所以对，我也在说封闭这件事

22:53

I will so shout out you know from factory because you know talking about agent rating as they've been all about that

我要 shout out 一下你知道的 from factory，因为你知道在讲 agent rating 的时候他们一直都很专注这个

22:59

They build signals tooling on that you do you guys use that? Yeah, yes

他们在上面构建了 signals 工具，你们用那个吗？对，是的

23:03

Yeah, so so we do use agent readiness of our own over our own kind

对对，所以我们确实用了 agent readiness，用在我们自己的 kind 上

23:06

I mean, you know is awesome as well like I think they were super early on this on this train

我是说，你知道 is awesome 也很棒，我觉得他们在这条路上超早的

23:10

And I think when he codified that in his talk

然后我觉得当他在演讲里把那件事 codify 出来的时候

23:12

I guess at AIE was it like last I forget when it was but one of us yeah, yeah, yeah, you're up

我猜是在 AIE 吧，好像是上次，我忘了是什么时候了，但我们其中一个，对对对，你上

23:16

Um, just fantastic talk and I think it was kind of in a way ahead of its time

嗯，就是超棒的演讲，我觉得那在某种程度上是超前于它的时代的

23:20

And so I was like there's only one person I could see being like the most

然后我就想，只有一个人我觉得是最

23:23

Most appropriate person to talk about this where I like really want to give them the spotlight and there's a very side

最适合聊这个话题的，我真的很想把聚光灯打在他身上，而且有一个很巧的

23:27

Yeah, it was funny. I was like called we were talking about this stuff

对，挺有意思的。我当时就打电话，我们就在聊这些东西

23:30

And like we're like okay, we need to inform some opinions about it

然后我们就想，好吧，我们得先形成一些观点

23:32

And then like factually like shipped a full product on it that week that we're talking about it

然后实际上那一周我们就直接把它做成了一个完整的产品，就在我们聊这个话题的那周

23:36

Yeah, you won yeah, yeah, yeah

对，你赢了，对对对

23:38

I mean like everyone kind of agrees on the same thing

我是说大家基本上都认同同一件事

23:41

Sure, sure, and it's just a question of like do you want to turn it through a dashboard?

当然当然，问题只是你想不想把它做成一个仪表盘来看

23:44

Yeah, yeah, yeah, yeah, because it's a dashboard right right right right right right

对对对对，因为这是个dashboard嘛，对对对对对对

23:48

But you know, there's there's value and like I think they did well from mark and perspective jumping on like the

但你知道，这确实有价值，我觉得从mark的角度来说他们做得挺好的，抓住了那个

23:52

The ratings report and I've definitely spoken about you know

那个ratings report，我肯定也聊过你知道

23:54

I will regularly kind of pitch that to other people and say like yeah, you should run ratings reports of your own kind and

我会经常跟别人推销这个，说对啊，你应该给自己跑跑ratings report

23:59

Yeah, of course not people have open source versions of it

当然不是，大家有开源版本的

24:01

But I think the principles are super important. Yeah

但我觉得那些原则超级重要。对

24:05

Do you want to go piece by piece or do you have any other things?

你想一条一条过还是你还有别的什么？

24:08

I can go piece by piece

我可以一条一条来

24:10

Yeah, okay. So the first one right you focus on cloud code

好，行。那第一个呢，你重点讲的是cloud code

24:12

But then you'll say pie as an open source sponsor. Yeah, okay, so so so okay

但然后你会说pie作为开源赞助商。对对对，行行行

24:17

This one is maybe a little bit more marketing-y, but okay

这个可能稍微偏营销一点，但没关系

24:19

The one thing I do want to highlight and have made that I also made that clear last year

我唯一想强调的一点，去年我也说清楚过

24:23

But I'm not a I'm not endorsing any tool. Okay, like like I don't I I think that is the wrong way to approach something like this

但我不是在为任何工具背书。好吧，我觉得那样对待这种事情的方式是不对的

24:30

There are tools that come and go and when you know transparently when when the classes taught last year

工具有来有去，你知道，透明地说，去年上课的时候

24:37

Cloud code was not as popular. It wasn't as like really accepted

Cloud code还没那么流行，还没那么被真正接受

24:40

Boris was only at 4,000 Twitter followers and now he's that you know, whatever million

Boris当时只有4000个Twitter粉丝，现在他你知道的，多少百万了

24:44

I don't even know if he is, but you know like he's not a lot of followers

我甚至不知道他是不是，但你知道，他粉丝不多

24:47

So so cloud code is really like emerge on the scene. I think yeah

所以 cloud code 真的是横空出世。我觉得是的

24:50

I think we also underestimate the cloud code when we had Boris and cat on the podcast right was like two months old

我觉得我们当时也低估了 cloud code，Boris 和 cat 上播客的时候才两个月大

24:56

I hadn't really like I used it once a query cost me 50 cents. I was like who's gonna use this shit totally

我当时没怎么用过，用了一次一个 query 花了我 50 美分。我当时就想谁会他妈用这玩意，完全

25:00

So it was still in beta for a while, but I think it's good on them for how they built it before the model was ready

所以它还在 beta 阶段待了一段时间，但我觉得他们在模型还没准备好之前就把它做出来，这点很厉害

25:05

Yes, like you hear so much build for where the model will go totally super ambitious like models were not that good

是的，你听到那么多“为模型未来的方向而构建”，完全超级有野心，模型当时没那么好

25:11

You couldn't fill it and then we see that's where it came

你没法填满它，然后我们看到那就是它到来的地方

25:14

It's yeah, it's funny thing because when he gave his talk last year

是的，有意思的是因为他去年做演讲的时候

25:17

I remember um, you know

我记得嗯，你知道

25:18

It was like a very high level sort of set of slides

那大概是一组非常高层级的幻灯片

25:20

But I remember one of the sizes literally built for where the model is gonna be six months for now

但我记得其中一个尺寸 literally 就是按照模型六个月后要达到的位置来设计的

25:23

and this was like October last year and

而那大概是去年十月的事

25:27

I can see why he would have made that point if if you knew in six months where like opus was gonna be because that was also

我能理解他为什么会提出那个观点，如果你知道六个月后 opus 会到什么程度的话，因为那也是一个

25:32

What I think was a huge unlock where I would argue was one of the reasons why the curriculum changes much as it did right where there was starting

我认为是一个巨大的突破点，我甚至会说那是课程变化如此之大的原因之一，就是从那时候开始

25:37

This huge arms race around true like autonomy in these agents like long-running agent coding agent tasks

围绕这些 agent 的真正自主性展开了巨大的军备竞赛，比如长时间运行的 agent 编码任务

25:43

But yes, so like you know back to your points. It's like less on I'm not endorsed in cloud. I mean a cloud code

但是的，所以就像你知道的，回到你的观点。这不太是关于我并不是在给 cloud 背书。我是说 a cloud code

25:48

I of course I use it and use other things as well, but but here I was more about like the what is a modern coding harness look like

我当然用它，也用别的东西，但这里我更多是想说，一个现代 coding harness 到底长什么样

25:54

I wrote a blog post that actually

我写了一篇博客文章，其实

25:57

Yeah, this one where this was yeah earlier this year where I sort of broke down some of some

对，就是这篇，今年早些时候的，我在里面拆解了一些

26:02

Similar-ish principles the TLDR was that you know, you could sort of through like pretty much introductory algorithms level understanding

类似的原则，TLDR 就是，你其实可以靠差不多入门级算法水平的理解

26:09

You just find like essentially an agentic loop that does the bulk of what you would call a coding agent

你基本上就能找到一个 agentic loop，它做了你所谓的 coding agent 的大部分事情

26:15

And of course, there's like bells and whistles

当然，还有各种花里胡哨的东西

26:16

There's no sub-agents. There's an MCP none of that stuff

没有 sub-agents，没有 MCP，这些都没有

26:19

But if you want to see something that you talk to and then it literally will edit your file

但如果你想看到一个东西，你跟它说话，然后它真的会去编辑你的文件

26:22

And you can do pre-sophisticated things with it you can it doesn't require more than 200 lines

你可以用它做一些很精巧的事情，不需要超过200行代码

26:27

So I you know a lot of what I try to do in educations to try like

所以我你知道我在教育中尝试做的很多事情就是

26:30

Peel away the stuff that's like the secret sauce and show that like really at its core

把那些像是秘方的东西剥开，展示它真正的核心

26:34

I mean these are still students and you want each conceptual principles

我的意思是这些还是学生，你想教每个概念原理

26:37

There's nothing more to this than in Stanford. There's this class called one of six b

这没什么比斯坦福的课程更复杂的。有一门课叫CS106B

26:40

It's like the introductory doubt introge algorithms

就像是入门的算法导论课

26:42

You don't need more than one of six beat up to build something like this. So there you go

你不需要超过CS106B的水平就能构建这样的东西。所以就是这样

26:46

And that there is given that you already taught this class

而且既然你已经教过这门课了

26:49

Any pushback any questions any weirdness that you observe from the students on what do they not get what do they ask seem to cluster questions around

学生在哪些地方有抵触、有疑问，或者你观察到他们有什么奇怪的反应，他们哪里没搞懂，他们问的问题是不是都集中在某些方面

26:58

You mean like now that this like this particular concept or just in general

你是说现在这个特定的概念，还是一般来说

27:02

this

这个

27:03

Concept of of like of like building it from scratch and then like doing doing what you did because this is basically what you needed for pie

就是从零开始构建这个概念，然后做你做的事情，因为这基本上就是你需要为pie准备的东西

27:10

Right right right exactly right. Yeah, right. So yeah, it's it not really

对对对，完全正确。是的，对。所以，其实并不是

27:14

I think they're I haven't got received any like pushback on on like this idea at all

我觉得他们，我还没收到任何对这个想法的抵触

27:18

I think they're actually more just most of them are consumers right most of the most students do just like end of using the tools

我觉得他们其实更多只是，大多数人是消费者，大多数学生只是在使用这些工具

27:24

I don't think that they get a lot of exposure if any sterling in the course work around

我觉得他们在课程作业中并没有太多接触，如果有的话，关于

27:28

peeling that onion back right and I think

一层层剥开来看，我觉得

27:30

So this is the week of the harness so so so Tuesday first lecture

所以这周是harness周，周二第一节课

27:35

You know logistics blah blah all that stuff

你知道的，后勤什么的乱七八糟那些东西

27:37

But then first half is for you know the first true like bit of curriculum is

但然后前半部分是，你知道，第一块真正的课程内容是

27:41

The 200 lines of of like a harness and then yesterday when I had sort of the follow-up to that is like

那200行的harness，然后昨天我做了个后续的内容就是

27:47

We actually took clot code and then you know you have these like proxy things

我们实际上拿了clot code，然后你知道你有这些proxy的东西

27:51

You either like for proxy tools reaching intercept or proxy? Yes exactly exactly. So use that you use that tool

你是用proxy工具来做拦截还是proxy？对对对没错没错。所以用那个工具

27:57

intercept requests and responses

拦截请求和响应

27:59

Look out of the hood see what's happening there and then just like okay

往外看看引擎盖下面发生了什么，然后就像这样，好吧

28:02

I did say you have clot code in 200 lines, but really okay

我确实说过你有200行的clot code，但真的，好吧

28:06

Here's like actual clot code and so what I literally did is I took up is five five

这里有真正的clot code，所以我实际上做的是我拿了五个五

28:09

So this like you know model of two days old and was like this is I intercepted those requests

所以这个就像你知道的，两天前的模型，然后就像这样，我拦截了那些请求

28:13

Here's exactly what's going on here's the prompt here's the system prompt here's the plan prompt here's the compaction prompt

这里确切地说明了发生了什么，这是prompt，这是system prompt，这是plan prompt，这是compaction prompt

28:17

And then we just like went through all of that and sort of showed it to students

然后我们就那样过了一遍所有这些，算是展示给学生们看

28:20

So now they really really have like a state of the art harness that they can understand

所以现在他们真的真的有了一个他们能理解的最先进的harness

28:24

Yeah, it's nice to get those three levels like students are very much consumers

是的，能得到那三个层次挺好的，学生们很大程度上是消费者

28:28

Here's the baseline agent in a loop and then the real one. I like how you have a

这是循环中的基线 agent，然后是真正的那个。我喜欢你有一个

28:34

Video walk through a few blog posts as well. Do you see students are on like

视频走查几个博客文章。你觉得学生们是处于像

28:39

extremes of

极端的两端吗

28:41

Some people are power users they consume a lot for some people you could assume that this is kind of

有些人是高级用户，他们消费很多内容，对有些人来说你可以假设这是那种

28:47

Very straightforward if you know how agents work. Right. How do you kind of balance that in the classroom setting like

非常直接的，如果你知道 agents 怎么工作的话。对吧。你如何在课堂环境中平衡这一点，比如

28:53

Do Taylor towards the average?

你是向平均水平做 Taylor 吗？

28:56

Yeah, I think it's a bit self-selecting because the curriculum is all here, right? So the curriculum has been available since

是的，我觉得这有点自我选择，因为课程都在这里，对吧？所以课程从

29:02

I mean you know left least like a few weeks and and I guess I announced it a few weeks ago

我是说你知道至少像几周前就已经可用了，而且我猜我几周前宣布了它

29:07

But I'd kind of been in the works for like probably just one to two months

但这个事情大概也就筹备了一两个月吧

29:09

So if you looked for you could probably find it

所以如果你去找的话应该能找到

29:12

And I feel like the reality is if the students are not interested in this

而且我觉得现实情况是如果学生对这个不感兴趣

29:15

Then they won't like they'll they just won't show up and they will they will choose not to show up

那他们就不会来他们会选择不出现

29:19

What I'm finding that is interesting is there are still definitely students that show

我觉得有意思的是确实还是有学生会来

29:22

But then there's also just a lot of folks from you know broader stanford postdocs at other departments

但也有很多来自更广泛的斯坦福群体的人其他院系的博士后

29:27

There's a lot of like interdisciplinary sort of interests where folks that are in you know

有很多跨学科的兴趣比如那些在

29:31

Somewhere in pathology or like you know like medicinal sort of fields or even like

病理学或者类似医学领域的甚至像

29:36

Random other engineering fields that aren't pure CS that are coming in and also interested in kind of getting exposed to this

其他一些不是纯CS的工程领域也在进来，也有兴趣接触这些东西

29:41

Also a lot of GSB students MBAs

还有很多GSB的学生，MBA们

29:44

They're also one I guess learn how to not have a technical co-founder or whatever the

他们也是我觉得要学会怎么没有技术联合创始人还是什么的那个

29:50

motivation there is but

动机是有的，但是

29:52

Yeah, so I think by the way if anyone hasn't done the MITM proxy exercise. Yeah, you should do it. Yes, you'll be shocked

对，所以我觉得顺便说一下，如果有人还没做过 MITM proxy 那个练习的话。对，你应该去做一下。是的，你会被震惊到的。

30:00

Cloud code has the longest system problems of all the coding agents something like 2000 tokens. Yeah. Yeah, although

Cloud code在所有coding agents里系统提示最长，大概2000个tokens。是的。是的，虽然

30:06

Yeah, that is true. That is true. Yeah, although they've also reduced the size

是的，确实如此。确实如此。是的，不过他们也缩小了规模

30:10

Yeah, yeah, I know but then they grew again and then they cut

对对对，我知道，但后来他们又增长了，然后又削减了

30:14

It's a sea saw. Yeah, yeah, I think it's interesting

这就像跷跷板一样。对对，我觉得挺有意思的

30:16

You'll be surprised like what goes in there. They actually sometimes you don't use

你会惊讶里面都放了些什么。他们有时候其实你不会用到

30:20

Sometimes it actually hints at the internal workings some of the tools

有时候它其实暗示了一些工具的內部运作方式

30:25

Anything that you found there that could be cool to talk about. Yeah, no

你在那里有没有发现什么值得聊的酷东西。嗯，没有

30:28

I mean in fact yesterday when I was talking about this it was a little bit of this like

我是说其实昨天我聊到这个的时候，有点这种感觉

30:33

One of the things that I thought was most interesting about cloud code back in the day

当年我觉得 cloud code 最有趣的一点之一

30:36

And I think part of why it really blew up in terms of popularity was

而且我觉得这也是它人气爆棚的部分原因

30:39

Their focus on simplicity right so in the early days they're among the first to really lean into this

就是他们对简洁性的专注，对吧，所以在早期他们是最早真正投入这一点的

30:45

Let's fewer tools do more right lean into sequel lean into bash lean into like the things that are that are powerful like Swiss Army

让更少的工具做更多事，对吧，深入用SQL，深入用bash，深入用那些真正强大的东西，就像瑞士军刀一样

30:52

Knife tools and have that be a lot of the driver of capability

刀具类工具，并让这成为能力的重要驱动因素

30:55

I think they did a fantastic job leaning to that. I think that's why they they were as popular as they did

我觉得他们在这方面做得非常出色。我觉得这就是他们为什么那么受欢迎的原因。

31:00

I think that when I presented something of the system prompt it was like well

我觉得当我展示了一些system prompt的内容时，感觉就像是好吧

31:03

There was a different system prompt about two months ago

大概两个月前有一个不同的system prompt

31:05

It is now they have reduced in complexity and it turns out that now what was once part of the system prompt

现在他们降低了复杂度，结果发现曾经属于system prompt一部分的东西

31:10

Things like how do you even get right or how to like set up tasks or things of that nature are no longer part of the system prompt

比如你怎么才算做对，或者怎么设置任务之类的东西，已经不再是system prompt的一部分了

31:16

It's actually expected to be more kind of captured in the models like the model weights itself

实际上，这更多是被捕捉在模型里，比如模型权重本身。

31:20

So they've kind of baked in a lot of those learnings more at the model level

所以他们在模型层面已经把很多这些经验内化了

31:23

And that just raises interesting questions around you know the model was the harness right and how do you kind of think through that trade off

这就引出了有趣的问题，你知道，模型本身就是harness对吧，你怎么去权衡这个取舍

31:28

What do companies invest in what you invest into really

公司该投资什么，你真正该投入什么

31:32

Push the the frontier of your own capabilities

去推动你自己能力的前沿

31:34

And so I think we still even as a as a community we see saw between the two a lot

所以我觉得即使作为一个社区，我们也经常在这两者之间摇摆

31:39

I think you don't want to fault them for growing a system prompt and then cutting it at some level

我觉得你不能怪他们把system prompt越写越长然后在某个层面又砍掉

31:44

It's also how the model was right

这也是模型本身的问题对吧

31:47

Weaker older models you need to give them examples of this is good. This is bad

更弱的、更老的模型，你需要给它们例子，告诉他这是好的，这是坏的

31:51

And then obviously you cut that away at some point right and that's just part of

然后显然你在某个时候会把它去掉，对吧，那只是学习的一部分

31:56

Learning what has been done right you can apply the same towards if you're

学习已经做了什么，对吧，你可以把同样的方法应用到如果你

31:59

Using a local coding agent with not the strongest model

用一个本地coding agent，模型不是最强的

32:02

You might need to add these examples, right? That's right. That's not correct. Yeah

你可能需要加这些例子，对吧？那是对的。那不对。是的

32:06

Yeah, and then and then of course there's a interesting AI learnings to fall out of that which is why

是的，然后当然从中会得出一些有趣的AI学习成果，这就是为什么

32:11

You know as the models get better

你知道随着模型变得更好

32:13

When are you over constraining it by adding those examples right and then have you pull away the over constraining

你什么时候会因为加那些例子而过度约束它，对吧，然后你又怎么撤掉过度约束

32:18

And that's I think part of the learning as well

我觉得那也是学习的一部分

32:20

Okay, we too you're going to talk about context engineering

好，我们也要聊聊context engineering

32:24

Repet I haven't seen but there's many versions of spectrum and development

Repet我没看过，但spectrum和development有很多版本

32:30

What do you would you highlight to a broader audience about

你想向更广泛的受众强调什么

32:33

What you're putting in here that people should know. Yeah, this is like the you know

你在这里面放了什么大家应该知道的东西。对，这就像是

32:37

One hour version of the course

课程的一小时版本

32:44

That's fair. Yeah, let's say it's also a trailer, right? It's like well

说得对。也可以说这是个预告片，对吧，就像是

32:48

But next next year next fall for courses, you know, this is your one hour

但明年明年秋天正式开课，你知道，这就是你的一小时

32:53

This is like this is like it was like the quibi of the class, you know

这就像是这课的Quibi版，你懂的

32:55

I mean, whether it's it's too long to be a trailer. It's too short to be the class is this like what it would I guess people will double play part

我是说，不管是啥吧，当预告片太长，当课堂内容又太短，这大概就是那种，我猜大家会反复播放的部分吧

33:02

If it's a trailer by the next time you teach it, it'll change

如果下次你讲课的时候它还当预告片，那它就会变了

33:07

Yeah, well, yeah, so it's okay. So it's a good call it on the news. It's it's funny

是啊，嗯，是啊，所以还行。所以这是个好主意，在新闻上提一嘴。挺有意思的

33:11

The I think that the main thing that I just want to reinforce here is

我觉得我在这儿最想强调的主要一点是

33:14

And I see this even what I'm like interviewing folks

而且我甚至在面试别人的时候也看到这一点

33:17

In the very few people in industry even progress beyond like just vanilla prompting right there's like a vendor forum like hey

业内很少有人能超出最基础的vanilla prompting，就好像有个vendor forum说嘿

33:23

Do me the thing right like do do the thing and and

帮我做那个事，就是做那个事，然后

33:26

That certainly as the models get good that can care you a certain distance

当然随着模型越来越好，这能带你走一段距离

33:30

But inevitably, you know, any focus I speak to that aren't come out of like a Google or someplace where there's a massive like code base

但不可避免的是，你知道，我交流过的任何一家公司，只要不是出自像Google或者那种有庞大代码库的地方

33:37

The reality is you have to be a lot more thoughtful about how you're like ingesting context right and so thinking through

现实是你必须更加深思熟虑地考虑如何摄取context，对吧，所以要仔细思考

33:42

What are some of those principles and those techniques around how to like point, you know

其中有哪些原则和技巧，关于如何指向，你知道

33:46

And I sort of like you know the rapid framework that I'm sort of describing here is is functionally similar to like dex's

我在这里描述的这个rapid框架，在功能上类似于dex的

33:51

RPI right or something of that nature where you have this like research kind of proposal

RPI之类的，就是你有这种研究类的提案

33:55

It's sort of a spec and then you and then you do like a your implementation and then you kind of tied together with some testing and whatnot

它有点像一份spec，然后你进行实现，然后通过一些测试之类的把东西串起来

34:01

so

所以

34:02

What I just wanted to reinforce here is that like

我在这里想强调的是，就像

34:05

You should have this step especially when you're entering a familiar code bases when you're trying to

尤其是在你进入一个熟悉的代码库时，你应该有这一步，当你试图

34:09

Sort of infuse into the context something that an agent needs to understand about a very large code base

把agent需要理解的关于一个非常大的代码库的东西灌输到上下文里

34:14

There's a value in something that is a research phase and then in the case of this particular sort of framework that I'm putting forward

研究阶段是有价值的，然后在我提出的这个特定框架的情况下

34:19

It's like also then suggests some proposal where there's a few alternative ways of solving a problem once you've done the research

它还会在完成研究之后建议一些方案，其中会有几种解决问题的替代方式

34:25

And so this is where I sort of also invite the human and the loop to sort of be the decision maker on one approach versus the other

所以这也是我邀请人类进入loop、在一種方法和另一种方法之间做决策者的地方

34:32

And then you enter the plan which is kind of the spec all to say the the the TLDR is like

然后你进入plan阶段，也就是spec，总而言之，TLDR就是

34:38

There is a way to do this that actually does scale out to code bases that are you know millions of lines long

确实有一种方法可以真正scaling到那些你知道有数百万行长的代码库

34:43

Um, and it's not just the vanilla prompt and I think that's what I want to highlight if you're going to go working these big code bases

而且它不只是vanilla prompt，我想强调的就是这一点，如果你要去处理这些大型代码库的话

34:47

You should use something a little bit more a little bit more involved

你应该用稍微更复杂一点的东西

34:51

And you think that add on context engineering

而且你觉得这涉及context engineering

34:54

I think that the probably the weirdest thing about how I think about context right now is when you split out to a sub agent

我觉得我现在对context最奇怪的理解是当你拆分出一个sub agent的时候

35:03

because

因为

35:04

sub agents

35:06

Can benefit from inheriting your entire context of the main thread or not

是否能从继承主线程的整个context中获益

35:11

Sometimes there's a regular disagreement on that and then sometimes you want to condense what the sub agent does and researchers

有时候对此存在常规的分歧，然后有时候你想压缩sub agent所做的工作，研究人员

35:18

put it back onto the main thread and

把它放回主线程，然后

35:22

Very hacky. I don't like anything I'm doing. So I don't have any answers. I know it's like

太hacky了。我做的这些东西我都不喜欢。所以我也没有什么答案。我知道这就像是

35:27

When you want to fan out work

当你想把工作分发出去

35:30

It is not just about

这不仅仅是

35:32

You have a big spec and you you know you do RPI you do our rep it whatever

你有一个大的spec，你知道你做RPI，你做我们的rep什么的

35:37

It is not one agent anymore. It is about fanning out work and make sure they have all the right context

这不再是一个agent了。而是关于把工作分发出去，确保它们都有正确的context

35:42

And they all have the right guidance if you have a redo agent reviewing all the other words

而且它们都有正确的指导，如果你有一个redo agent在审查所有其他的工作

35:46

And make that guy have have everything and you always go through this phase of expanding contracts like oh my god

让那个家伙拥有一切，你总是会经历这个扩展contracts的阶段，就像天哪

35:52

These guys don't have enough context. I'm going to give them more context. Okay, great

这些家伙没有足够的context。我要给他们更多的context。好的，太好了

35:55

And then they're going to overtune to some random one-off comment that you made a while ago

然后他们会对某个你很久以前随口说的一句话过度调整

36:00

And then make some really stupid mistakes because you you didn't mean it that way

然后犯一些非常愚蠢的错误，因为你根本不是那个意思

36:03

But like if I took it as the law

但就像如果我把它当成了铁律

36:06

Something like that. So I don't know what the word is. I'm saying right. It's just like multi agent management for context

大概就是这样。所以我不知道该用什么词。我是说对吧。就像是多agent管理中的上下文问题

36:12

Yeah, yeah, yeah, yeah, I think that that's absolutely right there

对对对对，我觉得这完全说得对

36:14

There's like some aspect of this is maybe the part that's

这里面有些方面可能是

36:18

More maybe among the most arthand science, right? It's like some combination. I think to your points

也许是最偏艺术和科学结合的部分，对吧？像是某种组合。我觉得结合你说的几点

36:22

Yes, the fan out for the exploration the fan out for like the subtasks and then and then the consolidation

是的，为了探索而fan out，为了子任务而fan out，然后再做整合

36:27

Combaction whatever, you know, you might want to call it

Combaction什么的，你知道，你可能想这么叫它

36:29

I think being thought about refreshing of the context understanding when things are getting a little bit too bloated

我觉得要思考一下上下文中理解的刷新，当东西变得有点太臃肿的时候

36:34

I think you know Dex had this concept of like the dumb zone and sort of like being a little bit more thoughtful about the dumb zone

我觉得你知道Dex有这个概念，像是dumb zone，然后对dumb zone稍微多一点思考

36:39

It's good at these equating things. Oh, yeah, yeah, I think he's a diagram. Yeah. Oh, yeah, he's been on

它擅长这些等同的东西。哦，对对，我觉得他有个图。对。哦对，他上过节目

36:44

Yeah, well, I think he's I mean he helped definitely informed a lot of the curriculum here

对，嗯，我觉得他是，我是说他肯定帮助启发了这里很多课程内容

36:47

And so he's just been asking I think think around all this

所以他一直在问，我觉得是在思考所有这些

36:50

No, I think the diagrams are are huge

不，我觉得那些图非常关键

36:52

Messy useful. Yeah, I would add

混乱但有用。对，我会补充

36:54

Breadcrumbs for people that are trying to look into this. It's good to see

给想要深入了解这个的人一些线索。很高兴看到

36:58

Best context engineering preferences like on your sub agent inheriting

最好的context engineering偏好，比如你的sub agent继承的

37:02

Cloud managed agents has a version where sessions are separate and it's not inherited

Cloud managed agents有一个版本，session是分开的，不会被继承

37:07

This used to be a big topic topic before reasoning models when

这在reasoning models之前曾经是个大话题，那时候

37:11

agents didn't run as long

agents运行时间没那么长

37:14

We had shorter context windows

我们的context window更短

37:16

Everyone knows very very

大家都知道非常非常

37:18

Managing that precious context window right

管理好那宝贵的context window对吧

37:21

Link chain used to write a lot of blogs about this in good context engineering all this stuff is good for people to know

Link chain 写了很多博客讲这个，好的 context engineering 这些东西对大家来说都值得了解

37:27

I think in the student context as well

我觉得在学生这个语境下也是

37:30

You can expect the average student to take your class because it's a great class and they'll know all these things

你不能指望普通学生选你的课是因为这是一门好课，然后他们就都会这些东西

37:35

So you should know them know them better as well

所以你自己也应该更了解这些

37:39

This is a fun space for research as well

这也是一个有趣的研究空间

37:41

One of the papers research papers presented at AIE they looked at the delta in the cloud code system prompt

在 AIE 上展示的一篇研究论文，他们研究了 Claude Code 系统 prompt 的 delta

37:49

What got changed and they like

哪些地方被改了，然后他们

37:52

They plotted a whole bunch of changes in performance. They tried to reverse engineer it

他们画了一大堆性能变化，试图去逆向工程它

37:56

It's just good research that you can do

这是你可以做的很好的研究

37:58

There's another one that also looked at context window management and how it affects performance downstream

还有一篇也研究了context window管理以及它如何影响下游性能

38:04

They weren't the best papers, but like it's quite accessible and it puts out good blog posts

它们不是最好的论文，但挺容易上手的，而且发的博客文章也不错

38:09

So people that are trying to work on it, you know, just recommend it's a it's a good area. You can easily poke into yeah

所以对于那些想在这方面做工作的人，推荐一下，这是个不错的领域，你可以很容易地切入，是的

38:16

Agent skills CLI. Yeah, yeah, I mean, I think all this is

Agent skills CLI。对对对，我觉得这些其实都是

38:20

So a little bit of an extension from a single single kind of

稍微延伸一下，从单一的那种

38:24

Function tool calling I'll like mcp right using interaction with like third-party external services

Function tool calling，比如mcp，对吧，利用与第三方外部服务的交互

38:29

And then extending into like what you would call it procedures, right?

然后延伸到所谓的procedures，对吧？

38:31

Like what how do you automate some set of actions over and over again that you then formalize in the form of like a skill

比如你如何把一组动作反复自动化，然后把它形式化成一种skill

38:36

I'm dear something like that so I think just making that a little bit more

我亲爱的，类似这样的东西，所以我觉得就是让那个稍微更

38:40

Giving people exercise in

给人们一些练习

38:42

codifying some of those procedures

把其中一些流程编纂成文

38:44

I think what I'll also sort of like include here is is a few few kind of fun skills that that I think are good to use

我觉得我在这里还会稍微包含几个我觉得很好用的有趣的skills

38:50

Like I think matte matte paccox like real me is like a good one

比如我觉得matte matte paccox像real me就是一个不错的

38:53

And then I there's ones that I also like from more fun the educational perspective like Jeffrey Litz

然后还有一些我也喜欢的，从更有趣、更有教育意义的角度来说，比如Jeffrey Litz

38:58

He has like explained if which I think becomes especially interesting once you start talking about code review because

他有像explained if，我觉得一旦你开始谈论code review，这就变得特别有意思，因为

39:03

You know the whole understand the code understand the code exactly yet

你知道整个理解代码、精确理解代码这件事

39:05

You got to understand the code and and I don't you guys have used that but effectively like you run you know

你得理解代码，而且我不……你们用过那个，但实际上是，就像你运行

39:10

You have your you're kind of that the the dip you've put it in your code and then you run the skill and at the end of it

你有你的，你把那个东西放进代码里，然后你运行那个 skill，在最后

39:15

You have this not only like kind of a visual representation of what's happened

你不仅有对发生了什么的可视化呈现

39:19

So you can you know kind of skim it and understand what's what's actually at the core of it

所以你可以浏览一下，理解它核心到底是什么

39:22

But then what I really liked was this almost like quiz at the end of

但然后我真正喜欢的是最后几乎像测验一样的部分

39:26

At the end of it where you go through and sort of like really reinforce whether you understand what's happening

在最后你过一遍，真正强化你是否理解了正在发生什么

39:30

And I think that that level interactive sort of back and forth. I think is is quite nice

我觉得那种互动式的来回交流，我觉得挺不错的

39:34

Yeah, there's there's a couple blog posts in this domain that I really like to bring up

对，这个领域有几篇博客文章我特别想提一下

39:38

This one. I think recently which is like kind of there's no situation in which turning your brain off is a good idea

这一篇。我觉得最近，就是那种“任何情况下把脑子关掉都不是好主意”

39:44

like you know

你知道吧

39:45

And I think these skills actually force you to turn your brain on

我觉得这些技能其实会逼着你把脑子打开

39:49

And when everyone wants to use coding you just turn your brain off

而当大家都想用coding的时候，你就直接把脑子关了

39:51

They're just like make me a app and get read my mind and guess we get everything correct without me telling you about it

他们就是那种“给我做个app”，然后指望我读心，猜我们不用我说就能把一切都搞对

39:57

That's right, but like a lot of these skills are mostly in the genre of like no no like ask me more questions please like

没错，但这些技能很多基本上都是那种“不不不，请多问我一些问题”的风格

40:04

Yeah, I think there's a lot more output than people are used to like I can type in a one-line prompt and get

对，我觉得输出量比人们习惯的要大得多，比如我可以输入一行prompt就得到

40:11

tens of thousands of lines of code and all structure and like look at my productivity guys

几万行代码和所有结构，就像看看我的生产力吧各位

40:16

Um, but you know

嗯，但是你知道

40:18

More effectively you should know what you're working on yeah

更有效的是你应该知道你在做什么，对吧

40:22

You can really be put out of touch out of the loop you should spend more time

你真的可能会变得脱节、被排除在循环之外，你应该花更多时间

40:26

Thinking about what you're actually doing

思考你实际在做什么

40:29

Doing the work and getting the code ridden is much easier than it was

做工作和把代码写出来比以前容易多了

40:33

Yeah, yeah, the understanding there's no no excuse for sorry. What is a web skill?

对对，理解这一点没有借口，抱歉。什么是web skill？

40:38

Yeah, so I mean effectively like a skills that interact with with like the web itself

对，所以我的意思是，实际上就是那种与web本身交互的skills

40:42

Computer use

40:43

Yeah, things like this

对，就是这类东西

40:45

I mean mainly things that involve like pulling pulling information from the web

我是说主要涉及从网上抓取信息的事情

40:48

their best practices on

他们关于……的最佳实践

40:51

Skills that will interact with the web versus I think this is more of like a bit of a

会与网络交互的技能，相比之下我觉得这更像是

40:55

Survey or kind of a demo of stuff that I found that was kind of interesting

一份调查或者说是对我发现的一些挺有意思的东西的演示

40:58

I think it was it might have been out of was it like a versel. I'm trying to remember now

我觉得那可能是出自……是不是像 Vercel 来着？我现在有点想不起来了

41:03

I'd have to look back at the the full like reading materials

我得回去翻一下完整的阅读材料

41:05

But there was like one site that I found where I think that maybe it was it might have been browser based actually

但我找到过一个网站，我觉得那可能其实是基于浏览器的

41:10

They might have like done. Ah, do you remember this?

他们可能做过类似的东西。啊，你记得这个吗？

41:12

Well, it would be them that been quite interesting

嗯，那应该会挺有意思的

41:15

It's just I was like did I miss something?

就是我当时在想，我是不是错过了什么？

41:16

This was like the big part of the skills community. Yeah, it's good. He's being educated

这算是skills社区很重要的一部分。对，挺好的。他正在接受教育

41:22

Yeah, I think they like made some sort of because I guess they were yeah, I mean browser is obviously like

对，我觉得他们好像做了某种……因为我猜他们是……嗯，浏览器显然就像

41:27

They like browsers. They like browsers and um, yeah, they had like a good

他们喜欢浏览器。他们喜欢浏览器，而且嗯，对，他们有一个挺好的

41:30

I think website where they serve some of these interesting capabilities

我觉得那个网站提供了一些挺有意思的能力

41:33

Seems legit

看起来挺靠谱的

41:34

CLI is a fun one

CLI挺有意思的

41:35

Do you have a lot of students?

你有很多学生吗？

41:37

This is their first time in a terminal in a CLI or is it kind of the norm?

这是他们第一次在CLI里用终端，还是说这已经算常态了？

41:41

No, I don't I don't think so I think that given that most of them again

不，我不……我不这么认为，我觉得考虑到他们大多数人……

41:46

Last year, I would say that some of them had probably still

去年，我想说他们中有些人可能还记得

41:49

Remember the days of of finger on keyboard typing the this is

在键盘上敲finger命令的那些日子，这是……

41:54

Most I think students have like interacted with cloud code at this point

我觉得大多数学生到现在应该都用过cloud code了

41:57

I mean, I'm curious the split of

我是说，我很好奇现在的分布是怎样的

42:00

People that are now just using the apps right cloud code also has a desktop app. You know

现在大家就是直接用这些app，cloud code也有桌面版app了。你知道吧

42:04

So does codex codex is primarily an app now. Yeah, you're kind of not in the CLI not in the terminal

codex也是，codex现在主要就是一个app了。对，你基本上不在CLI里，不在终端里

42:10

It's a thing that puts off a lot of average people, but

这一点会让很多普通人望而却步，但

42:14

I'm just yeah, no

我就是，对，不是

42:16

Yeah, I would say you know, I think there's also another thing about skill

对，我想说，你知道，我觉得关于skill还有另一点

42:21

There are but skills that are bundles of skills where they sort of cross reference each other. Yeah, like the plugins right

有一些是skill的bundle，它们之间会互相交叉引用。对，就像plugins那样

42:26

Like plugins and then there's also I guess

就像plugins，然后还有，我猜

42:30

The art of writing a skill, which is a thing that I've also focused on very IE

写skill的艺术，这也是我一直非常关注的东西，IE

42:35

I think a lot of these are focusing on skill-consuming

我觉得这里面很多都聚焦在消费skill上

42:38

Yeah, and you have all these influencers like Macpokog and one have you there like this skill. We'll see if you resolve all your problems

是啊，然后你有那些网红比如Macpokog，还有那个谁，跟你说这个skill。看看你能不能解决你所有的问题

42:44

But no, you should write your skills yourself. Right. And actually when you try to write them naively

但其实不对，你应该自己写你的skills。对吧。而且实际上当你天真地尝试去写它们的时候

42:49

They're usually bad and actually it takes a little bit of skill to write a skill

它们通常都很糟糕，而且写一个skill其实需要一点技巧

42:55

That's absolutely right. That's absolutely right. There are switches best tips on writing. I have a skills

完全正确。完全正确。关于写skill有一些最好的建议。我有一个skills

42:59

I have a public skills people I went through it definitely a phase of psychosis of like skill everything

我有一些公开的skills，我经历过那个阶段，绝对是经历了一段精神病般的阶段，什么都想搞成skill

43:03

No, no, don't give us skills. How do you write good skills? Yeah, exactly. So I have a skill for cutting skills

不不不，别给我们skills。你怎么写出好的skills？对，没错。所以我有一个关于写skills的skill

43:09

So I I like there so every like

所以我我挺喜欢这样的，就是每一个像

43:12

Concord and codex they all have skill creation skills, right?

Concord和codex它们都有skill creation的skills，对吧？

43:16

So then there's I have a skill cutter skill that cuts it down because they they're all

然后就有了，我有一个skill cutter skill来把它精简掉，因为它们都

43:20

It just documents every way in which those default skills that are created are bad

它只是记录了那些默认创建的skills有多糟糕的每一个方面

43:25

So you always run this after any skills created to cut it down, right?

所以你总是在任何skills创建之后运行这个来把它精简掉，对吧？

43:30

So yeah, you it's really

所以是的，你它真的

43:34

Annoying because sometimes there is slop and like but there's good in the slop

很烦因为有时候会有slop然后像但slop里面也有好的东西

43:38

And then you have to like sort of whittle down what they're with right right? Oh interesting as you're saying like even the slop

然后你就得有点像削减掉它们是什么对吧对吧？哦有意思像你说的就连slop

43:44

It's still in the language of the agent so that the slop means something

它还是用agent的语言，所以那些slop才有意义

43:47

It's like it's like a it's yeah, okay, okay got it

就像，就像，是，对对，明白了

43:49

It's when everyone has their own version. I used first

就是当每个人都有自己的版本。我先用的

43:52

There's very easy to like vibe check a skill, right? You just call it run it a few times

很容易对一个skill做个vibe check，对吧？你直接调用它跑几次

43:57

Yeah, and then I distill these preferences into a skill that's like this is what I like this way

对，然后我把这些偏好蒸馏成一个skill，就是我喜欢这样

44:02

He doesn't like and it it adjusts over time and then you know

他不喜欢，然后它会随着时间调整，然后你知道

44:06

The problem is that yeah, you're calling it but the actually the worst part is when a skill triggers when you're not

问题是，对，你在调用它，但实际上最糟糕的部分是当一个skill在你没预料到的时候触发

44:12

Expecting you to like you wrote it one time to fix the thing and then like now

就像你写了一次来修那个东西，然后现在

44:16

You're working on a completely different thing and it triggers why because it's value-ridden because the triggers is like

你在做一件完全不同的事情，然后它触发了，为什么？因为它充满了价值判断，因为触发条件就像是

44:21

Oh, you call me whenever blah blah blah happens and you're like okay reads good

哦，只要发生什么什么你就叫我，然后你说好，读起来不错

44:25

And then you do a different project and it and it triggers and you're like

然后你去做另一个项目，它触发了，你就想

44:29

Yeah, there's a there's like a few people that are just like okay

对，有那么几个人就是那种，好吧

44:32

Have a bunch of skills like keep adding on sometimes run this up and then I know some people like a Eugene

有一堆skills，不断往上加，有时候跑一下，然后我认识一些人像Eugene

44:37

They are pretty particular about how many skills they have they prudent them if not used they clean them every few weeks

他们对自己有多少skills非常讲究，他们会修剪，如果没用就每隔几周清理一次

44:45

Yeah, yes, so yeah, this is like a huge problem if you're if you're sort of in that skills world

对对对，所以是的，如果你处在那个skills的世界里，这确实是个大问题

44:51

I think this is actually why I suspect

我觉得这其实正是我怀疑的原因

44:54

They're good. They're very smart people at opening and topic that one skills to die

他们挺好的。他们是非常聪明的人，在开场和话题上，那个技能真是要命

45:00

Which is not a popular statement no now they they won't say this publicly

这不是一个受欢迎的说法，现在他们不会公开这么说

45:03

But it just feels sloppy this feels sloppy. I know it's yeah, I can see my own skills people being sloppy. It doesn't feel

但就是感觉很粗糙，这感觉很粗糙。我知道，是的，我能看到我自己技能的人很粗糙。感觉不

45:11

Uh, AI peeled right I mean yeah, yeah, it's it's not it's not what EGI was supposed to look like a

呃，AI剥离得对，我是说，是的，是的，这不是AGI应该看起来的样子

45:18

It's like everyone having a giant bunch of markdown files

就像每个人都有一大堆markdown文件

45:21

I mean I think

我是说我觉得

45:23

There's a gardening problem. Yeah, I think in some sense a lot of people actually don't look at their markdown files right like the

有一个 gardening 问题。是的，我觉得在某种意义上，很多人实际上不看他们的markdown文件，就像

45:29

the

那个

45:30

UX of it of just chatting what you want creating a skill and chat using it seeing output

就是聊天的那种UX，你想创建什么skill就创建，然后聊天用它，看输出

45:36

There's a lot you can there's a lot of value you can get without ever looking at your markdown

你不用看你的markdown就能获得很多价值

45:41

Don't don't over complicate it at some level for yourself

别把它搞得太复杂，在某种程度上对自己来说

45:45

You can get a lot of value if you do stuff repeatedly just be like look at the last five times we did this make me a skill and

如果你反复做某些事，就能获得很多价值，比如说看看我们过去五次做这个的情况，给我做一个skill

45:51

You can iterate it with chat. There's a good side of it as well, right? There's another thing

你可以通过聊天来迭代它。这方面也有好的一面，对吧？还有另一件事

45:56

There's like a another way to look at it as well because there's a there's the use of the skill as just procedure that the agent kind of runs whenever to

还有另一种看待它的方式，因为skill的使用就是作为一种procedure，agent在需要的时候运行它

46:03

You know do like a code review or whatever but then in the cases something like Devon right where we've codified skills as a way of

你知道，比如做code review之类的，但在像Devon这样的情况下，我们把skills编码为一种

46:09

of sort of representing our best practices right and then making and exposing that to the agent the the code review system to then you know

表示我们最佳实践的方式，对吧，然后把它暴露给agent，给code review系统，然后你知道

46:18

Institute the APIs the way that we like to define them institute kind of the service of the boundaries the way we define them

按照我们喜欢的方式定义这些API，按照我们定义边界的方式来建立服务

46:22

That's like an alternative way of sort of representing like or global knowledge and maybe not so much this like

这就像一种替代方式，用来表示某种全局知识，也许不太是那种

46:29

Arbitrary thing you want the agent to do so you don't have to write the problem each time

你想让agent做的任意的事情，这样你就不用每次都写问题

46:33

And so I like you I like that

所以我喜欢你，我喜欢这一点

46:35

I feel like that positioning might that that purpose might exist longer than just

我觉得那个定位，那个目的可能会存在得更久，而不只是

46:40

You know these kind of for random procedures. I just google monoco AI skills and it's giving me the the country

你知道这些随机的流程。我刚谷歌搜了一下monoco AI skills，它给我的是那个国家

46:48

Maybe the country one day

也许有一天是国家

46:50

Well, we're gonna be the money. Oh, that's for sure. You just buy the country

嗯，我们会成为钱的。哦，那是肯定的。你直接把国家买下来

46:54

Well, here's the thing. Okay on this

嗯，事情是这样的。好，关于这个

46:56

We uh, we're up out to but no, we're not about the country but but but um, you know, we did

我们呃，我们本来要出去，但不对，我们不是要聊那个国家，但是但是但是嗯，你知道，我们确实

47:01

We dug into a little bit of flak when we bought the monoco.com domain. I will say that we did get

我们买下monoco.com这个域名的时候遇到了一些麻烦。我得说我们确实被他们

47:06

Sweet by them. So

恶心到了。所以

47:08

What yeah, there's there's like a whole public

什么，对，就是有一整个公开的

47:11

There's a whole public thing about this right

关于这事有一整个公开的事情，对吧

47:13

Yeah, so they didn't like that but but here we are and uh, we want we want so it's okay

对，所以他们不喜欢这样，但是但是我们现在在这儿了，呃，我们想要，我们想要，所以没关系

47:18

Then we're keeping it in. I was about to say this is spicy, but you won. Yeah. Yeah. No, no, no, it's

那我们就留着它。我正想说这挺劲爆的，但你赢了。对对对，不不不，就是

47:23

It's not really is that way we see the flags flying around

其实并不是我们看到的旗帜到处飘扬那样

47:27

Because it's like a victory one. It's good to victory. It's our victory. That's funny

因为那像是一个胜利的标志。胜利是好事。这是我们的胜利。挺有意思的

47:31

For those that are in SF just look up and know. Yeah, yeah

那些在SF的人抬头看看就知道了。对对

47:36

So uh for those who are on audio, we're showing this on youtube

所以呃对于那些在听音频的人，我们在YouTube上展示这个

47:39

It's you fly a plane around. Yeah, I think you're still one of the few that actually do this on a regular basis

是你开着飞机到处飞。对，我觉得你仍然是少数真正经常这么做的人之一

47:45

Sandblown came up with it six thousand dollars a day

Sandblown想出来的，一天六千美元

47:48

Uh, I don't necessarily think it directly

呃，我不一定觉得这直接

47:52

Causes me to google what Monaco is, but that all good

导致我去谷歌Monaco是什么，但都挺好的

47:56

Monaco's ads are just monaco dollar sign

Monaco的广告就是monaco美元符号

47:58

Right. Yeah, we stand for like dollar sign in the next number

对。是的，我们就是代表美元符号加下一个数字

48:02

It starts the conversation right all the the whole point of the ad is what the hell is monaco

它开启了对话，广告的全部意义就是“monaco到底是什么鬼”

48:07

Yeah, and and you know you you could you could argue like that's a debate to be had

是的，你知道你可以争论说这值得讨论

48:10

But but in some sense like what is the purpose of marketing if not to like get attention right and yes

但某种意义上，如果营销不是为了吸引注意力，那它的目的是什么，对吧，是的

48:15

Maybe there's one level beyond this which is if somebody so you know

也许还有更高一层，就是如果有人，你知道

48:18

I'll go to events and people like oh you guys are the ones with the planes

我去参加活动，人们会说哦你们就是那些有飞机的人

48:20

You guys are the ones with like the big number on the billboard blah blah

你们就是那些广告牌上有个大数字的人之类的

48:22

193 you know million whatever it is um and and that but they're like but what do you do and then but that's that's but that's

193 你知道百万不管是什么嗯然后但他们会说但你是做什么的然后但那那但那

48:30

But maybe that's the whole point right is that I just needed someone to come and use it as an icebreaker

但也许这就是重点对吧就是我只需要有人来把它当作一个破冰的话题

48:36

I will say it's what I said as you walked in as well. I was like oh the planes

我得说你走进来的时候我也是这么说的我当时就想哦那些飞机

48:40

Yeah, yeah, and now I know what do you guys do? Well, well, let me tell you that what we do

对对现在我知道了你们是做什么的好吧好吧让我告诉你我们做什么

48:43

So you know as someone who's been in those meetings

所以你知道作为参加过那些会议的人

48:46

Deciding what we do on the the Devon billboards. We spend so much time on the words

决定我们在Devon广告牌上做什么我们花了太多时间在措辞上

48:50

And maybe we should just be like Devon

也许我们就应该直接写Devon

48:54

Just so fucking dumb but like

就他妈的这么蠢但就像

48:57

Okay

48:58

You know look I think Sam had another post this past week where he was just talking about like you know

你知道，我觉得Sam上周还有一条帖子，他就是聊了聊你知道的

49:03

I mean Sam is effectively like our you know marketing sort of mastermind here

我是说Sam实际上就像是我们这里的营销大师

49:07

And uh, yeah, I think there's a little bit of history behind but bend the planes as well

而且，嗯，我觉得这背后有一些历史渊源，但bend the planes也是

49:10

But but the TLDR is that

但但TLDR就是

49:12

His whole stick is like I don't have all the answers

他的核心观点就是，我并不是所有答案都有

49:14

You just got to try things and I think a lot of this is yeah, we just try things and and if it sticks great

你只需要去尝试，我觉得很多情况就是，对，我们就试试看，如果成了那太好了

49:19

If it like did we know this is gonna blow up as much as it did I mean not really but but now that it did it's an idea

如果它像——我们知道它会这么火吗？我是说其实不太知道，但但既然它火了，那就是个好主意

49:23

Yeah, exactly and minimal is start small go high instead of start big go small right way easier to get start small go go big

对，没错，最小化就是从小做起往大了走，而不是一开始就搞大的然后缩小，对吧，从小做起往大了走容易多了

49:31

Yeah, exactly

对，没错

49:33

It's six K a day like it's oh yeah, it's very much. Just try things. I think I think people cheaper than billboard cheaper than a billboard for sure

一天六个K，就像，哦对，非常是。就是去尝试。我觉得我觉得人比广告牌便宜，肯定比广告牌便宜

49:40

Yeah, and it's fun. Yeah, not much harm done, but you know, it does it's um

对，而且很有趣。对，没什么大损失，但你知道，它确实，嗯

49:44

But you know the kind of simplicity here that that's worth mentioning is then when we're trying to like one up the plane

但你知道这里值得提的那种简单性，就是当我们想要在飞机上再加一个的时候

49:48

We're like, you know what's better than one plane two planes

我们就想，你知道什么比一架飞机更好吗，两架飞机

49:52

Two planes what

两架飞机什么

49:54

My my way to rocket well, okay

我的上火箭的方式，好吧，行吧

49:56

I also had that I have seen that idea

我也看到过那个想法

49:59

I would like for that that happened as well at some point, but uh, we'll see

我也希望那件事在某个时候能发生，但呃，我们拭目以待吧

50:03

This is the next basics launch we just see

这是我们接下来看到的基础版发布

50:10

Got it warm up to our space x friends. Okay, I'm have

好了，跟我们SpaceX的朋友们热热身。好，我来

50:14

You don't want to

你不想

50:16

Is it worth talking a little bit about monaco? Do you leave it? Yeah? Yeah

值不值得稍微聊聊Monaco？你要离开它吗？是啊？是啊

50:20

Because there's just like an intermission between the topics

因为就像话题之间有个中场休息

50:23

You decide okay, we're we mentioned monaco right like let's let's give

你决定吧，好，我们提到了Monaco对吧，那就让我们

50:27

I think it's I think it's a little bit annoying to like mention what the house monaco and then not answer it

我觉得这有点烦人，就是提到那个摩纳哥的房子然后又不回答

50:32

Sure. Can you answer it? Sure. Well, I hope so. Yeah, so

当然。你能回答吗？当然。嗯，希望如此。对，所以

50:37

This is the country so it's a small little

这是一个国家，所以它是一个小小的

50:42

There you go there you go. Okay, that this is monaco

这就对了，这就对了。好，这就是摩纳哥

50:45

This is the monaco that that I know and love

这就是我所知道和喜爱的那个摩纳哥

50:48

Yeah, so we are you know, we're a series v company

对，所以我们是一家Series V公司

50:51

Here based in sf

总部在SF

50:52

We are building effectively like an all-in-one sales platform

我们实际上在打造一个一体化的销售平台

50:56

fully i native agents doing all the fun stuff

完全是原生agents在做所有有趣的事情

50:59

Or at least the stuff that humans don't like to do but but that agents are do like to do because they don't really have opinions

或者至少是那些人类不喜欢做但agents喜欢做的事情，因为它们其实没有自己的意见

51:04

But do in a less air prone fashion

但以不那么容易出错的方式来做

51:06

So the sales people and sales teams can focus on doing what humans are good at which is you know closing deals

这样销售人员 and 销售团队就可以专注于做人类擅长的事情，也就是你知道的，成交

51:12

So presumably

所以大概是

51:14

Mark bandy off doesn't like you guys

Mark Bandy off不喜欢你们

51:16

Probably not but he's like also

可能吧，但他也

51:19

Honestly a very i native person very aggressive and agent force and everything. Yeah, what are you doing that he doesn't do or he cannot do well

说实话是一个非常原生的人，非常激进，agent force什么的都用上了。 yeah，你们在做什么是他不做或者他做不好的

51:27

Okay, so so right now we are we started the company in a sort of like we're the the market that we went after was not enterprise when they get go right

好，所以现在我们是，我们创办公司的时候，某种程度上我们瞄准的市场不是企业级市场，对吧

51:34

We sort of say like we're gonna start by like if our goal is to build the all-in-one

我们基本上是这么说的，我们先从——如果我们的目标是打造这个all-in-one

51:38

Then we need to kind of build this in an early fashion with a group of customers

那我们就需要以一种早期的方式，和一批客户一起构建

51:43

And we need to like grow with them into this

然后我们需要和他们一起成长到这个阶段

51:44

It's much harder to pull some away from from sales force than to say like hey if we meet you at the point when you are still just a

从Salesforce那里抢客户要难得多，不如说嘿，如果我们在你还只是用

51:50

Google spreadsheet tracking pipeline and we sort of show you a

Google spreadsheet追踪pipeline的时候遇到你，然后我们给你展示一个

51:54

You know more AI-pilled sort of answer the problem

你知道的，更AI-pilled的解决方案

51:57

Then we can grow with you and in a way it kind of follows the same the same model of like you know

那我们就可以和你一起成长，某种程度上这跟那个模式是一样的，你知道的

52:02

Brecks are all the like YC companies where you kind of sell through batch mates you kind of grow with them and then you go up up market

Brecks都是像YC公司那样的，你通过同批的伙伴来销售，跟他们一起成长，然后往更高端的市场走

52:07

And so we are I would argue are the you know the I would I would claim that among the best if not the best sort of like a startup focused

所以我认为，我会说我们算是，可以说是最好的，如果不是最好的话，那种专注于初创企业的

52:15

Sales platform and so all of our customers are like early stage startups at this point probably through like

销售平台，所以我们所有的客户现在都是早期初创公司，大概到

52:20

Series A is is kind of where where we primarily been focused kind of see to series A and then now we're going more up market

Series A左右是我们主要专注的阶段，从种子轮到Series A，现在我们正在往更高端的市场走

52:26

And we've kind of learned a lot about the all-in-one in the last year and half of doing that

在过去一年半做这件事的过程中，我们对all-in-one学到了很多

52:30

And you are the head of AI. Yeah, that's right. What's some cool AI stuff you guys have built?

你是AI负责人。对，没错。你们做的一些很酷的AI东西是什么？

52:35

Yeah, I mean

对，我是说

52:36

We have built our own agentech harness for for sales use cases. So all this I mean to call back to the coding harnesses

我们为销售用例构建了自己的agentech harness。所以所有这些，我是说回到coding harnesses

52:45

Taking a lot of inspiration from from the systems like the codecs is in the in the clouds of the world and sort of building

从codecs这类系统中汲取了很多灵感，它在全球的云环境中，然后构建

52:49

The spoke tools that are very sales focused bespoke workflows that are all kind of geared toward GTM use cases

那些非常以销售为导向的定制化工作流工具，都是面向GTM用例的

52:56

So we built our own version of that that power is pretty much every single thing that that powers the monical platform. So from you know

所以我们构建了自己的版本，它几乎驱动了monical平台上的每一个功能。所以从你知道的

53:03

scoring of scoring of Tams

Tams的评分

53:05

Scoring of ICPs and develop defining by CPS

ICPs的评分以及通过CPS来定义

53:08

Through to pipeline evolution through to like all the mundane kind of things that maybe a salesperson doesn't want to do

到pipeline的演进，再到那些销售人员可能不想做的所有琐碎的事情

53:14

Which is like oh update my meeting you know pre-meeting. No, it's update my summaries like all these things

比如哦更新我的会议，你知道会前准备。不，是更新我的摘要，所有这些事情

53:19

You know we have agents powered powering all those workflows

你知道我们有agents在驱动所有这些工作流

53:22

And so we've we've done that it's it's kind of comes comes out of the box

所以我们已经做了这个，它算是开箱即用的

53:28

And yeah, so long long agents long horizon agents for long horizon agents and short horizon agents some cases for for for GTM use cases

然后是的，所以长周期agent，长周期agent和短周期agent，某些情况下用于GTM用例

53:36

And it's pretty domain specific built on foundation models. Would you do any of your own models?

而且它相当领域特定，建立在foundation models之上。你们会做自己的模型吗？

53:42

Yeah, it's a good call out because so far I mean, I think this is generally true of a lot of like vertical AI companies

是的，这是个好问题，因为到目前为止，我觉得这对很多垂直AI公司来说普遍成立

53:48

It's like you kind of start where you sort of use the frontier models like this third-party services to see what's possible

就是你一开始会用前沿模型，比如第三方服务，来看看什么是可能的

53:53

Because they are like the very frontier

因为它们就是最前沿的

53:55

And then I think we've seen this I guess in the past week with the whole like Harvey sort of where they like the 50 plus 50 was in my speed

然后我觉得我们过去一周看到了这一点，就是整个Harvey那件事，他们那个50 plus 50在我的速度里

54:01

We were sent margin kind of question

我们被问到了margin的问题

54:03

I mean, I think in your respective of the numbers

我是说，我觉得从你们各自对数字的角度来看

54:05

I think what is interesting is like at a certain point things do become interesting or cost perspective

我觉得有意思的是，在某个节点上事情确实会变得有意思，或者从成本角度来看

54:10

You you start to wonder can you do it faster on you know open-weight models

你开始想，能不能在开放权重的模型上做得更快

54:14

And then that's a lot of that then a lot of what I think the the purpose of the AI

然后很多就是，我觉得AI这部分的目的

54:17

Pieces are is just to posturing those models build environments for you know kind of to posturing using our own things of that nature

就是对这些模型进行后训练，构建环境，你知道，用我们自己的东西来做后训练之类的

54:23

So we're kind of in that phase right now

所以我们现在大概处于那个阶段

54:25

Oh fun

哦，有意思

54:27

Yeah, I've been calling this table to the agent lab playbook

是的，我一直把这叫做agent lab的playbook

54:30

But that's kind of my coining off like you know like agents start from the user first and then they

但这算是我的一个造词，就像你知道的，agent是从用户端起步，然后它们

54:36

They use whatever models you need to build products

它们用任何你需要的model来打造产品

54:39

But then you can backfill them the model side once you have the user base

但一旦你有了用户基础，你就可以反过来补上model那一侧

54:43

And then the model labs go kind of good the other way they start the big patrons and then they try to find users

而model lab走的是相反的路，它们从大客户起步，然后试图去找用户

54:48

Um, and it's like very you know

嗯，这就像你知道的

54:50

It's hopefully working as a as a as a split a monoco would be on the sort of revenue or gtm side. That's right

希望在revenue或GTM那一侧能形成一种拆分，一种monoco式的拆分。没错

54:55

Yeah, um, and I think it's interesting

是的，嗯，我觉得这很有意思

54:58

Uh, they would want to serve you while also try to compete with you

呃，它们既想服务你，又想跟你竞争

55:02

Mm-hmm as as model labs that is correct one of the things that I've been I was having I dabble in in some of this

嗯嗯，作为模型实验室这是对的，我一直在涉猎这方面的一些东西

55:09

I think one of the hard problems is actually enrichment

我觉得其中一个难题其实是数据丰富化

55:12

mm-hmm

嗯嗯

55:14

I would say I spent a lot of my time doing that for for basically SEO and like

我想说我把很多时间花在做这个上面，基本上是为了SEO之类的

55:21

Research and what have you and I imagine for you is actually like a core part of that

研究和什么的，我想对你来说这其实是核心部分

55:25

That's right. Do you find that models are good at it?

没错。你觉得模型擅长这个吗？

55:29

Do you have any vendors that like you like working with you want to give them a shout out?

你有没有什么供应商，你比较喜欢合作的，想给他们打个招呼？

55:32

Yeah, yeah, absolutely. Yeah, so we we have used we were early adopters of parallel

对对，当然有。我们用的是，我们是Parallel的早期采用者

55:37

And then I think they're very strong. Okay, so just general search. Yeah, I mean not like zoom info or something

然后我觉得他们非常强。好吧，就是普通的搜索。对，我的意思是不像ZoomInfo之类的

55:42

No, no, no, um, but yeah, we use parallel. We also use X uh, um, so we'll give it to give a shout to both of those folks

不不不，嗯，不过对，我们用Parallel。我们也用X，呃，嗯，所以我们要给这两家都 shout out 一下

55:48

Uh

呃

55:50

They're both friends

他们都是朋友

55:52

They're both folks that uh that we like they'll be very interested if you like one of them

他们都是我们喜欢的人，如果你喜欢其中一个，他们会非常感兴趣

55:57

Yeah, we will. I'll keep it. I'll keep it a neutral. I'll stay I'll stay Switzerland in this today

对，我们会的。我会保持，我会保持中立。我今天会保持，我会保持瑞士立场

56:02

But yeah, no, so and then of course we do our own version of enrichment as well

不过对，不是，所以然后当然我们也做我们自己版本的 enrichment

56:05

So we it's like a most of our data systems

所以我们，就像我们大多数数据系统

56:07

I mean part of like one of one I think the most valuable piece of software that we build are most valuable systems

我是说，其中一个我觉得我们构建的最有价值的软件或者最有价值的系统

56:12

Is currently like our world database like the database of companies enrich companies enrich accounts

目前就是我们的世界数据库，就是公司数据库、丰富公司信息、丰富账户信息

56:17

Enriched contacts so like that whenever we are constructing ICPs for folks

丰富联系人信息，这样每当我们为客户构建ICP的时候

56:21

We have the most update information about who you should reach out to

我们拥有关于你应该联系谁的最新信息

56:25

At which company at this point in time like I would say that one of our like I think where we are like world-class hands down

在哪个公司、在这个时间点，我想说我们的其中一个，我觉得我们绝对是世界一流的

56:30

Is just demand like we are very good demand company like we help companies

就是需求方面，我们是一家非常擅长需求的公司，我们帮助企业

56:34

Increase their meetings booked in a way actually fine by your at flying by or by at the right time, right? And it's kind of like the combination

以正确的方式、在正确的时间增加他们预订的会议，对吧？这有点像一种组合

56:41

Like there's there's times when like here's an example like if you are

有时候，举个例子，如果你

56:44

If you are targeting someone that just you are selling to

如果你的目标对象是你正在推销的人

56:48

VPs of sales, let's say okay like the the right time to reach out to them is not when they are like sort of in the middle of some

比如销售副总裁，合适的联系时机并不是当他们正处于某种

56:56

Like they've been at the same place for the last like five years

比如说他们已经在同一个地方待了五年

56:58

If they are now in the middle like a job like a good signal is like once if they've gone from one job to a new job

如果他们现在正处于比如换工作的阶段，一个好的信号是，一旦他们从一份工作换到了新工作

57:03

They're now like entering a new place where they're more likely to be to purchase or be open to purchase of a new sales platform

他们现在进入了一个新环境，更有可能购买或愿意购买一个新的销售平台

57:09

Because they're in the middle of the job transition. So

因为他们正处于职业转换期。所以

57:11

Going from company to company B for like the sales a good signal of like hey

从一家公司跳到另一家公司，对于销售来说是一个好信号，比如嘿

57:15

Should you reach out to them at this point in time is did you detect that they're linked and changed or something like that?

你应该在这个时间点联系他们吗，你是否检测到他们的LinkedIn发生了变化之类的？

57:19

How much of this is your

这其中有多少是你们的

57:21

Expertise put into model and harness versus models are good. There's generic best practices like

专业知识融入模型和harness， versus 模型本身就很好了。有一些通用的最佳实践，比如

57:27

Is there there's obviously some value in both? Yeah

显然两者都有一些价值？是的

57:32

Is it just the glue getting it reliable doing it all in one platform doing it well? It seems like there's also

是不是就是那些胶水工作让它变得可靠，在一个平台上把所有事情做好？看起来还有一些

57:38

Stuff that we wouldn't know that you guys are also domain experts in that's picked in yeah

我们不会知道的东西，而你们同时也是那方面的领域专家，被融入进去了，是的

57:43

I think without the former it would it would be like a fully commoditized

我觉得没有前者的话，它就会变成一个完全商品化的

57:47

I mean not completely because there's still a lot of value in the all in one

我的意思是不完全是这样，因为一站式整合仍然有很大的价值

57:50

I think like that is something that we definitely pioneered like most companies in the sales space started as just like point solutions

我觉得这确实是我们率先开创的东西，销售领域的大多数公司一开始都只是单点解决方案

57:55

And I think they're they tried more of like the YC approach of you kind of like land and expand

我觉得他们更多是尝试了YC那种land and expand的方式

57:59

You kind of pick a niche and then you expand from there

你先选一个细分市场，然后从那里扩展

58:01

And we from the get go I mean that's why we stayed in stealth for the greater part of a year and some months was just building

而我们从一开始，我的意思是这就是为什么我们在stealth模式下待了大半年，就是一直在build

58:06

Because we knew we wanted to be the all in one from the start

因为我们知道我们从一开始就想做all in one

58:07

But you know that alone is is powerful. It's meaningful

但你知道光这一点本身就很有力量，很有意义

58:11

But I think the real power is the fact that today Monaco as Monaco has like a world-class go-to-market team

但我觉得真正的力量在于，今天Monaco作为Monaco拥有像世界级的go-to-market团队

58:16

Like a truly like world-class go-to-market team selling Monaco of course

真的是世界级的go-to-market团队在卖Monaco，当然

58:19

But we also have the same go-to-market people that sell Monaco are the ones that are like what we call four-to-poi account executives

但我们也有同样的go-to-market人员，卖Monaco的人同时也是我们所谓的four-to-poi account executives

58:24

They're FDA ease. They're effectively like one is a human human sales people that are like the consultants of sorts that help

他们是FDA ease。他们实际上就像一个是人类销售人员的角色，类似于顾问之类的，帮助

58:32

Onboard all our customers get them like building agentic workflows

引导我们所有的客户入驻，让他们建立agentic workflows

58:35

And are effectively there to like help guide them through every part of their own sales go-to-market life cycle

并且实际上在那里帮助他们引导他们完成自己销售go-to-market生命周期的每一个环节

58:39

And so to your question

所以回答你的问题

58:41

It's the all-in-one yes, but it's also the fact that we have these like

它是一体化的，是的，但同时也是因为我们有这些

58:44

You know Sam and and you we have brought on a lot of like former Sam

你知道Sam和，你我们引进了很多像前Sam

58:48

Sam's team members and stuff like that who were like the best people in GTM at Brex and other places

Sam的团队成员之类的，他们曾是Brex和其他地方GTM方面最优秀的人

58:53

So we're codifying all of that in the platform

所以我们把所有这些都codify到平台里了

58:56

Yeah, it's interesting to have you talk very differently than a normal hit-of-the-eye

嗯，挺有意思的，你说话的方式跟一般的随便聊聊很不一样

59:00

I might also talk to you because you're also a former founder

我可能也想跟你聊聊，因为你也是曾经的创始人

59:03

Right and like you're you're also think you used to thinking about these things in this way and it's like

对，而且你也是，你习惯用这种方式思考这些问题，这就

59:08

Interesting to it's important to have that right in your role. Yeah, cool

挺有意思的，在你的角色里有这一点很重要。对，酷

59:13

I I think you've introduced Monaco enough there and then obviously I think as Monaco the story develops we can revisit

我觉得你已经把 Monaco 介绍得够多了，然后显然我觉得随着 Monaco 的故事发展我们可以再聊

59:21

Maybe the bridge back is also what is coding

也许回连的桥梁也是什么是 coding

59:26

Or agenated coding or modern software development like inside of Monaco

或者说 agentic coding 或者现代软件开发在 Monaco 内部是什么样的

59:30

um, you know, I think you're when you're teaching you're maybe trying to be here's a lay of the land

嗯，你知道，我觉得你在教学的时候，你可能是在试图呈现一个全景图

59:36

But maybe inside of Monaco you guys are more opinionated on one thing versus the other

但也许在Monaco内部，你们对某件事的看法比其他事更有倾向性

59:40

Yeah, yeah, no, I mean I we have so we made investments in coding agents from basically like day one

对对，不是，我是说我们有，我们从基本上第一天起就在coding agents上做了投资

59:46

And we've like learned a lot we have our own like scars that we sort of like picked up and in learnings

然后我们学到了很多，我们有自己的伤疤，算是积累了一些经验和教训

59:50

And I think we've now been getting much deeper into sort of sharing some of our learnings

我觉得我们现在越来越深入地分享我们的一些经验

59:55

So like in our blog, which might be we're sharing here. We've been

比如在我们的博客里，可能我们会在这里分享。我们一直

59:58

Increasingly posting some of what maybe not that so that one is more like a product for the one at the very top of this stuff

越来越多地发布一些，可能不是那个，那个更像是这个最上面的一个产品

60:03

I would say like

我会说

60:05

Yeah, when agets that spread in before like scaling contributions builder by SRE

是的，当agents扩散之前，像scaling contributions builder by SRE

60:09

So you know, we spent a lot of time looking at different providers and vendors and mended building a lot of things in house

你知道，我们花了很多时间研究不同的提供商和供应商，也自己内部搭建了很多东西

60:15

So for us something like a background agent platform

所以对我们来说，像后台agent平台这种东西

60:17

We have our own version of like a ramp inspect that we call Monacoader

我们有自己版本的类似ramp inspect的东西，我们叫它Monacoader

60:21

Which is is built on the code or platform

它是基于codeer平台构建的

60:23

But it is what we use to essentially scale up our engineering product

但它就是我们用来规模化提升工程生产力的工具

60:26

Codeer like codeer.com codeer.com. Yeah. Oh, yeah

Codeer就是codeer.com，codeer.com。对。哦，对

60:29

We actually yeah, but have sort of adapted it to some of our use cases

我们确实用了，不过针对我们的一些用例做了适配

60:32

And yeah, it's where a lot of our agent development happens

对，我们很多agent开发都是在那里进行的

60:35

It's just this really well designed piece of software for essentially isolating of sort of

这只是一个设计得非常好的软件，本质上用来隔离某种

60:40

development environments scaling out to multi agents

60:43

You know being able to run these like long long horizon agent coding agent tasks

你知道能跑这些像很长很长horizon的agent coding agent任务

60:47

So Monacoader shout out to John Tracy and and co for for building that out

所以Monacoader，感谢John Tracy和团队把它搭建出来

60:52

And and then yeah like from there

然后然后对，就像从那里开始

60:54

We've also built our own versions of like a you know like a slack bot, which is

我们也做了我们自己版本的，你知道，就像一个Slack bot，它是

60:58

Call how but um, you know effectively how is like this part AISRE part like customer success part like

怎么称呼都行，但是嗯，你知道实际上怎么运作的呢，就是一部分是AISRE，一部分像客户成功那样的。

61:07

Kind of QA engineer because it lives in our slack

有点像QA engineer，因为它就住在我们的slack里

61:11

But it serves all the purposes of

但它在各方面都很有用

61:14

Helping us debug things and put up PRs when things are broken

帮我们调试问题，在出问题时提交PR

61:16

But also used by our own sales people to find out how the product works find out like how to get to the bottom of something

同时我们自己的销售人员也用它来了解产品怎么运作，搞清楚怎么深入排查某个问题

61:21

And so there's like interesting questions. We had to answer around the different personalities that how has

所以就有一些很有意思的问题。我们必须回答关于不同角色性格的问题

61:26

Depending on whether it's talking like a sales person versus like a developed developer

取决于它是在像销售人员那样说话，还是像开发者那样说话

61:31

And so systems like this that we built out over time

像这样我们随着时间搭建起来的系统

61:33

a lot in the last like year. So

过去一年里做了很多。所以

61:36

Shout out to you and Jeff or for how and John Tracy as well

向你和Jeff致敬，还有How和John Tracy

61:40

So you're building a lot in house. Yeah

所以你们内部自己做了很多开发。对

61:43

You also use other tools. We do use other tools. So what is the line?

你们也用其他工具。我们确实也用其他工具。那界限在哪里？

61:47

Yeah, so was that yeah the builders by it. It's a good question and we

对，所以那是……对，就是那些自建的人。这是个好问题，而且我们

61:53

Yeah, I think there's this

对，我觉得有这么一个

61:55

uh

呃

61:56

Like I think the general principle that I generally which is like you build to learn and you buy to keep or sorry

我觉得我一般遵循的总体原则是，你自建是为了学习，你购买是为了保留，或者说抱歉

62:01

You buy to learn and you build to keep yeah the other way around here

你购买是为了学习，你自建是为了保留，对，是反过来的

62:05

So you you buy to learn and you build to keep and I think that's generally true here where in the early days when we were still trying to figure out like you know

所以你购买是为了学习，你自建是为了保留，我觉得这在这里总体上是对的，早期我们还在摸索的时候，你知道

62:12

Get our feet under us a little bit. We were you know

让我们稍微站稳脚跟。我们当时你知道

62:15

Could reviews example one of them where we spent a lot of time looking different could review platforms

代码审查平台就是一个例子，我们花了很多时间研究不同的代码审查平台

62:18

It turns out today we still use Devon because we've actually still find a lot of value in how the system's been built

结果今天我们仍然在用 Devin，因为我们确实发现这套系统的构建方式仍然有很多价值

62:22

So you know shout out to the cognition team

所以你知道，要感谢 cognition 团队

62:24

But then along the way we did try out some like SRT tools

但在这个过程中我们也尝试了一些 SRT 工具

62:27

And you know, we tried some sort of like off the shelf may be more like background agent platforms

你知道，我们试了一些现成的、可能更像是后台 agent 平台的东西

62:31

Decided that it was just given our infrastructure given how we had set up our own cloud platform

最终觉得，考虑到我们的基础设施、考虑到我们自己云平台的搭建方式

62:35

It just made more sense to kind of build this out internally

在内部自建这套东西显然更合理

62:37

Of course, we're not dogmatic. We we are open to revisiting them if if something comes out

当然，我们并不是教条主义的。如果出现了新的情况，我们愿意重新审视这些决定。

62:41

But here I

但在这里我

62:44

Felt I think usually we ended up building in house when we felt there was no mature offering out there

觉得，通常当我们觉得市面上没有成熟的方案时，最终就会选择自研。

62:49

So

所以

62:50

And could review is maybe like an interesting example because we did do a whole survey of could review tools like maybe over a year ago

代码审查也许是个挺有意思的例子，因为我们确实在大约一年多以前对代码审查工具做过一次全面的调研。

62:56

And at the time I'll show it out like yeah, yeah could wrap it and then and you know that there's also like could write graphite

当时我展示了一下，对对，代码审查，然后你知道还有像 Graphite 这样的

63:02

Graphite you graphite for a while before they got acquired

Graphite，你们用了一段时间的 Graphite，后来他们被收购了

63:05

Like ellipsis and which like a YC company before the fact see that but there's a bunch

像 Ellipsis，还有一家 YC 公司，在被收购之前也看过，但这类东西有很多

63:09

So so there were a lot of folks that we did try and they were okay

所以，所以我们确实试了很多人，他们还行

63:12

But I don't think they had like crossed the threshold of like they're immediately useful yet

但我觉得他们还没有跨过那个门槛，就是还没有到立刻就能用的程度

63:15

And so then we kind of like we sort of like put it aside a little bit

然后我们就有点把它放到一边了

63:18

This was you know, I'm saying like well over a year ago

这是你知道的，我说的是 well over a year ago

63:20

Then we kind of went back to this and sort of organically

然后我们又回到了这个上面，有点自然地

63:23

I mean again shout out to the to the Devon team and cognition folks where we tried it again

我的意思是，再次 shout out 给 Devon 团队和 cognition 的人，我们又试了一次

63:27

I think I hit sort of like gotten that somebody told me like hey you tried this and you know

我觉得我有点像是，有人告诉我，嘿你试过这个，你知道

63:31

It seems like it's gotten a lot better met and I've been actually during silences talk during the class last year

它似乎变得好多了，而且我实际上在 silences 期间，去年课上讲过

63:36

And it was just good and it was organically like like that so purely vibes right like this is there's no way to do this

而且它就是很好，而且它就是那种自然有机的，纯粹靠感觉，对吧，这个没办法量化

63:42

Yeah, this this was like I mean okay, okay, here's what I'll say when we did that first evaluation

对，这个就是，我是说好吧好吧，我这么说吧，我们第一次做那个评估的时候

63:48

The first time around where we didn't really pick any vendor

第一轮我们其实没选任何vendor

63:51

And we were like there's no vendor that we're super happy with I did do like an analysis where I looked at our PRs

然后我们说没有哪个vendor让我们特别满意，我确实做了个分析，看了看我们的PR

63:55

Basically, it's sort of like a you know set of high diagrams like here's what this is solving are we

基本上就是一堆high diagram，就是这里在解决什么，我们

64:00

Are we happy with the kind of the trade off of verbosity and how often it's posting with utility?

我们对verbosity和它发布频率跟utility之间的那种权衡满意吗？

64:05

And so there was it was more statistically driven than just vibes the second time around I wasn't gonna push hard on it

所以第二次更多是数据驱动的，不只是靠感觉，我第一次不想太使劲推

64:10

The first time I was kind of like hey, we should adopt an echo tool second time

第一次我有点像说嘿我们应该采用一个echo工具，第二次

64:13

I was like no, I'm not gonna do that again

我当时就想，不行，我不会再干那个了

64:15

But then organically people just like hey Devon is like good like can you like add me to this and so today

但后来自然而然地，大家就，嘿 Devon 挺靠谱的，你能不能把我加到这个里面，所以今天

64:19

People are just like getting organically added so this is a vendor where you know one of the cases

大家就是自然而然地被加进来了，所以这是一个 vendor，你知道其中一个案例

64:23

We're still leaning to the vendor you know

我们目前还是倾向于那个 vendor，你知道

64:25

We'll see in the long term what that looks like, but um

长期来看会是什么样，我们拭目以待，但嗯

64:28

Everything outside of that. We just didn't feel like maturity was there. Yeah

除此之外的一切，我们就是觉得成熟度还不到位。对

64:31

And it's not painful enough that you would like we're about having right you know, that's right. That's interesting. Yeah

而且它也没有痛到那种程度，让你觉得，我们差不多就行了，你知道，对，就是这样。有意思。对

64:37

Good

好

64:38

Good talk for Devon

和Devon聊得不错

64:40

I didn't even know

64:42

You know for the record I'm only an advisor. I'm not that involved anymore, but like it's interesting to to hear about it

你知道，声明一下我只是个顾问。我已经没那么深度参与了，但听你说这些还是挺有意思的

64:47

externally because internally

64:50

There is a certain amount of let's call it

有一定程度的，我们姑且称之为

64:54

The the bubble where like on slack you're like you only see your positive things because people only want to circulate positive things

那个泡沫就像在Slack上，你只看到你那些正面的东西，因为人们只想传播正面的东西

65:01

But like you you have no dog in the fight like you just choose best tool

但就像你，你在这件事上没有利益牵扯，你就是选最好的工具

65:03

I mean if you want to get us like frequent. That's totally that

我的意思是，如果你想让我们像常客一样。那完全没问题

65:12

Okay, so so so there's that right and then and that informs all your teaching work

好，所以所以所以有这个对吧，然后这会影响你所有的教学工作

65:16

Yes, that's right. You know, I think that's that's really good

是的，没错。你知道，我觉得那真的很好

65:19

And we can bring in more of this as you like because I think this is this lived experiences very good. Yeah

我们可以根据需要多引入一些这方面的内容，因为我觉得这种亲身经历非常好。是的

65:25

I'm happy to go back to age already code basis or born like so to week four and five

我很乐意回到已经编码的基础或者天生的那种，回到第四周和第五周

65:30

kind of

差不多

65:32

You know matching yeah, they blend a little bit. Yeah, yeah, that's right. Whatever you want to whatever you strong beliefs you have introduced

你知道匹配的，是的，它们有点融合在一起。是的，是的，没错。你想怎样都行，你有什么强烈的信念就引入

65:39

In Monaco. Yeah, and recommend people to start with. Yeah, no for sure. Yeah, week four and five

在Monaco。是的，推荐人们从那里开始。是的，当然。是的，第四周和第五周

65:44

I would say they they are a little bit um for as maybe a little bit more like

我会说它们有点，嗯，也许有点更像是

65:48

Conceptual foundations and ways to do this because there are a lot of ways to customize agents

概念基础和实现方法，因为定制agent有很多种方式

65:53

And then I would say week five is where I really can like lean into

然后我会说第五周是我真正能够深入的时候

65:56

All right, let's think about this and apply this in practice on like a true code base

好，让我们思考一下这个，并在一个真实的代码库上实践应用

66:00

And so here I will definitely you know, give a shout out to the factory team and and uh and want to just highlight the value of something like a readiness report

所以在这里我肯定会，你知道，要特别感谢factory团队，并且想强调一下像readiness report这样的东西的价值

66:08

I think that the main piece that I want to reinforce

我觉得我最想强调的核心部分

66:12

Something like agent ready code code basis is to me like a precursor and a bit of a building block toward everything to come afterwards

像agent ready code code basis这样的东西，对我来说就像是前奏，也是通向后续一切的一个构建模块

66:18

When I say everything to come average, I mean like

当我说后续一切的时候，我的意思是

66:20

People always got you know for a while and I think people still are like super hyped on like multi-agents like paralyzing agents like tens of agents doing all kinds of things

人们总是，你知道，有一段时间了，我觉得人们现在仍然超级热衷于multi-agents，比如并行化的agents，几十个agents做各种事情

66:27

It's like you actually need to set a foundation in order to do that well and the foundation is how

就像你实际上需要打好基础才能做好这件事，而基础就是

66:32

Bare five loads of code base how like well structures a code base how understandable is it by an agent

代码库的裸五行负载，代码库的结构有多好，agent对它的理解程度有多高

66:36

And so I think like agent readiness to me is like almost like piece one of the next four to five weeks where maybe until like week eight or nine

所以我觉得agent readiness对我来说几乎就像是接下来四到五周的第一部分，也许直到第八或第九周

66:43

Where where I spend more time talking about like broadly

在那之前我花更多时间广泛地谈论

66:46

You know as software teams and like kind of or global structure, but

你知道，作为软件团队以及整体结构，但是

66:50

Definitely building up the the building blocks for for background agents of the software factory

绝对是在为软件工厂的后台agent搭建积木

66:55

I think this is where it all kind of gets interesting from from that perspective

我觉得从这个角度来看，这就是一切变得有趣的地方

66:59

Yeah, one thing that I observed you talking about for monocle, but it's not here

是的，我注意到你谈到Monocle的一件事，但这里没有

67:03

Yeah, it's the SRE side when you close the loop

对，就是SRE那一侧，当你把闭环合上的时候

67:07

You actually want to have the agent read your logs for builds and then I ideally read your analytics for live production

你其实是想让agent去读你构建的日志，然后理想情况下我再去读你线上生产环境的分析数据

67:15

That's right. That's right. Yeah, and now it's not explicitly, but it's a really good shout out

没错。没错。是的，现在这虽然不是明说的，但确实是个很好的致意。

67:18

So I think that the one to weave that in in like two places one is in the background agents

所以我觉得，把它编织进去的一个方式是在两个地方，一个是在background agents里

67:23

Week and then also in the software factory week. So absolutely right

Week，然后还有software factory week。完全正确

67:26

I do it's a really good call out this idea of like trigger slash event based closing of the loop

我觉得这确实是个很好的提醒，就是这种触发/事件驱动的闭环思路

67:31

Which you know now is one of the one of the probably two or three entry points when people talk about agents kicking off tasks in the software factory

你现在知道的，这大概是人们谈论agent在软件工厂里启动任务时两三个切入点之一

67:38

That's absolutely right that'll come in there. Yeah, it's interesting that people call it trigger or event base

完全正确，那个会加进去的。是啊，有意思的是人们叫它trigger或者event base

67:42

Which I guess that is the actual

我猜那其实就是它真正的

67:46

Technical implementation of it. Yeah, it just seems to low level almost

技术实现方式。对，感觉有点太底层了几乎

67:50

Uh, but so the trigger is usually a cron job right like that's be real sure sure sure

呃，但触发通常是个 cron job 对吧，说真的，对对对

67:56

Event base is usually a cron job that like

事件驱动通常就是个 cron job，像是

67:59

Surface is an event and then it's event base, but it's all crons. Yeah. Yeah, that's right

浮现出来是个事件，然后就是事件驱动，但其实全都是 crons。对。对，没错

68:04

Yeah, like webhooks or something like that. But yeah, that's absolutely right

对，像是 webhooks 或者类似的东西。但没错，完全正确

68:07

Yes, and you know the technical term of course because at the end of the day

是的，你当然知道那个技术术语，因为说到底

68:11

Yeah, being sort of like a CS person by by trade and yeah, we we called it those things and so I guess that's uh

对，作为一个科班出身的 CS 人，对，我们就那么叫那些东西，所以我猜那就是呃

68:18

I fall back on that terminology. Yeah, and

我依赖那套术语。对，而且

68:21

Have any AI SRE companies made an impression on you?

有没有哪家 AI SRE 公司给你留下过印象？

68:25

So okay, I mean like I think the resolve folks are doing well there the resolve is doing well

好，我是说，我觉得 resolve 那帮人做得不错，resolve 做得挺好

68:30

I would say night shift is also doing quite a quite a good job there in terms of broadly like background agents

我会说 night shift 那边也做得相当不错，就是广义上的 background agents

68:35

I'm a big fan of kind of like they're gray lockbacked. I think they're doing doing well in this area as well

我很喜欢那种，他们是 gray lockbacked 的。我觉得他们在这个领域也做得挺好

68:40

Yeah, I would say that

对，我会这么说

68:42

Where I think SRE is an interesting use case and this might back to the build versus buy question

我觉得 SRE 是个有意思的用例，这可能会回到 build versus buy 的问题

68:48

I think what's interesting about it is like where do when does it become

我觉得有意思的地方在于，什么时候它会变成

68:51

Like where does where does an SRE an SRE like really see the most value

比如说SRE到底在哪里最能体现价值

68:55

It's typically when you have like a tremendous number of like service dependencies

通常是在你有大量服务依赖的时候

68:59

And these like massive sort of like systems and tons of Kubernetes and Helm files like so you know

还有这些庞大的系统和一大堆Kubernetes和Helm文件之类的

69:04

Just like tremendously large Kubernetes infrastructure Helm that's like files kind of string everywhere around

就是那种超大规模的Kubernetes基础设施Helm文件到处都是

69:08

I think that's why um resolve is done really well in like the enterprise right where you go and you sell to forward

我觉得这就是为什么Resolve在企业里做得特别好，就是你去卖给那些大企业

69:13

You go and sell to like a Walmart and you kind of say hey you have

你去卖给像Walmart这样的公司，然后你说嘿你们有

69:17

Your service all dependencies are crazy

你们的服务依赖关系乱七八糟的

69:19

Um, and you see these these like you know these diagrams of how things are set up

然后你看到这些这些你知道的这些架构图

69:22

And then they kind of go in and they're like we'll we'll try and make sense of all this

然后他们会介入，然后他们会说我们会试着理清这一切

69:25

I think they're like at smaller scales

我觉得他们在较小的规模下

69:27

It's usually still manageable that you can kind of roll your own

通常还是可控的，你可以自己搞定

69:30

But you know, I will say that for the enterprise. I think they've done a good job

但你知道，我得说对于企业来说，我觉得他们做得不错

69:33

Yeah for for people who I have never worked at this skill because it took me a while to

是的，对于从未在这个技能领域工作过的人来说，因为我花了一段时间才

69:37

Uh to to see this it is on the order of let's let's call it tens of thousands of repos more repos than developers

呃才看到这一点，它的量级大概是，我们称之为数万个repos，repos比开发者还多

69:44

These are the situations where like you don't know what code is running

这些就是你不知道什么代码在运行的情况

69:48

That's anywhere in the company because it's probably the guy who is no longer here

在公司任何地方都是如此，因为那可能是那个已经不在这里的人写的

69:52

Right and like by the way, it's still a little bearing

对了，顺便说一下，这还是个有点隐患的地方

69:56

Yeah, that's that's the scariest thing

对，那才是最可怕的事情

69:58

That's absolutely right. It's like it's and then you're the person you know

完全正确。就像，然后你就是那个你知道的人

70:01

You're the person who's on call and you're interacting with a system that goes down

你就是那个随时待命的人，你在跟一个会宕机的系统打交道

70:05

Where that person still isn't there and there's like all out of date run books

而那个人还是不在，还有一堆过时的操作手册

70:08

And you're like what the hell is going on here? It's a scary scary thing

你就想这到底是怎么回事？这真是件可怕可怕的事

70:12

And yes, you're at the enterprise level. It's like it's a whole whole different beast

而且是的，你在企业级别。这完全是另一回事

70:16

Yeah, so so like the the kind of like college hacker

对，所以就像那种大学黑客

70:19

They you know one person per evil they've never told no appreciation

他们你知道的，每个人都有自己的恶，他们从来没说过不感激

70:23

But like I think I think you just give that idea like you drop them in like a simulation of like

但我觉得，我觉得你只要把那个想法给他们，就像把他们丢进一个模拟里

70:29

Okay, like they want on your job. This happens. What now? Yeah

好吧，就像他们想要你的工作。这发生了。现在怎么办？对

70:32

Yeah, so so another shout out. I'll give here

对，所以再提一个，我在这里要提一下

70:35

So the resolve folks actually last year they made an assignment for the class where they effectively

所以Resolve那些人去年实际上给课程做了一个作业，他们实际上

70:42

Created like a sandbox for students to go and then the sandbox was like stream

创建了一个沙盒让学生进去，然后沙盒就像是stream

70:46

I mean within the sandbox there was like many many services that were doing their base things

我是说在沙盒里面有很多很多服务在做它们的基础事情

70:51

And there was even like bugs in the services

甚至服务里还有bug

70:54

Within that sandbox and then as part of the assignment the students had to like go and actually

在那个sandbox里，然后作为作业的一部分，学生得真的去

70:58

Like use the resolve agent and kind of like probe and play the that they the human on call person

就像用那个resolve agent，然后去探测和玩，就是那个人工值班的人

71:03

probing the system describing what's happening

探测系统，描述发生了什么

71:06

There's got an interesting. Yeah, it's an interesting website or not

有个挺有意思的。对，是个有意思的网站吧

71:10

That's been a while since you know

有一阵子没见了，你知道

71:12

Yeah, this is very interactive. Okay, they're doing a picture duty, you know

对，这个很有互动性。好，他们在做图片值班，你知道

71:16

Yeah, yeah, yeah, yeah, but also this like this this video. I was uh, you know, you were shouting out

对对对对，但还有这个，这个视频。我当时，呃，你知道，你在喊话

71:25

Yeah, so okay, we you know

对，所以好吧，我们，你知道

71:27

We never get a chance to talk about this side of the world except for conversations like this any other

我们从来没有机会聊到世界的这一面，除了像这样的对话之外，其他的都没有

71:32

segments because I think AI SRE is maybe too broad or even too narrow

71:36

Depending on how you look at it, right? Like too broad meaning while you there's several subdivisions of AI SRE

这要看你怎么看，对吧？比如范围太广了，而AI SRE还有好几个细分方向

71:41

That's interesting

这很有意思

71:43

Or to narrow being like well actually that's not even the biggest problem that we have

或者说得窄一点，其实那甚至都不是我们最大的问题

71:46

So it take you however you like yeah, you mean like specifically for like Monica or just like broadly for like organizations thinking about this

71:53

Broly because that is our audience

Broly，因为那就是我们的受众

71:56

Mm-hmm. Yeah, I think that it again. It's a it's a function of how we just scale right and I think that for for the enterprise

嗯哼。是的，我觉得这又是。这取决于我们如何扩展，对吧，我认为对于企业来说

72:03

They there are a number of issues that aren't always

他们那儿有很多问题并不是一直都

72:06

I almost feel like the answer is not even always

我几乎觉得答案甚至不总是

72:10

Just more agents and since we're still on the week five agent ready code basis like when I talk to companies of like a certain scale

就只是更多的agents，而且因为我们还在第五周agent ready code的基础上，就像我跟某些规模的公司聊的时候

72:16

The the things that are holding them back are really more just like low-level like DevOps CICD stuff, right?

真正阻碍他们的其实更多就是那种底层的DevOps CICD的东西，对吧？

72:21

They're like they're they're merged cues are getting backed up

他们就是他们的merge queue在堵

72:23

They it takes too long to stand up an app because there's just like actually more foundational things that haven't almost nothing to do with

他们部署一个app要花太长时间，因为实际上有更多基础性的东西，几乎跟这些完全没关系

72:29

They're like throwing in one more SRE bot or a la you know what a bot wouldn't be the answer the problem because there's like more systemic things that need to figure out and so

他们就像再扔一个SRE bot进去，或者一个la你知道bot根本不是答案，因为还有更多系统性的东西需要搞清楚，所以

72:37

So that brings us the week six. Yeah, that's that's that's what we get that's a yeah easy

所以这就带我们到了第六周。对，那就是我们得到的，那是一个，对，简单

72:42

Yeah, there we go. So the merge Q and beyond um, yeah, this is a yeah a level that are sorts to modern genetic code review

对，就是这样。那么合并Q以及之后，嗯，是的，这是一个某种程度上现代基因代码审查的水平

72:51

Yeah, I think that the the biggest thing that maybe happened between last year and this year was just

对，我觉得去年到今年之间可能发生的最大的一件事就是

72:56

If you suddenly seem very clear that code review is a bottleneck especially once you've got really a build and and people

如果你突然发现代码审查明显是个瓶颈，尤其是当你已经有了一个真正的构建和人员

73:01

There's the velocity on generation became so high that you know, you could in theory like put up 30 plus PRs a day

生成的速率变得太高了，你知道，理论上你一天可以提交30多个PR

73:08

And then what like what's where what point do you decide that

然后呢，你在什么节点决定

73:13

You either have to institute yourself and put yourself in the middle as some way to to gate and review things in the traditional way

你要么必须建立自己、把自己放在中间，以某种方式用传统方法来把关和审查东西

73:19

But then you're you're literally the constraint you're the choke point of the velocity and that's not good

但那样你就是字面意义上的约束，你是速度的瓶颈，这不好

73:24

Right, then then what was the point of generating at scale or you say well, how do we maybe get the best of both worlds where we're

对吧，那大规模生成的意义又是什么，或者你说，我们怎么才能两全其美，既

73:29

instituting ourselves enough that we have trust in what's getting merged

我们已经建立了足够的制度，让我们对合并的内容有信任

73:34

But we're also

但我们也能

73:35

Able to kind of like merge at scale, but we're also not kind of like freewheeling it here and anything goes

能够大规模地合并，但我们也不是在这里随心所欲、什么都行

73:40

So a lot of I think what I want to touch on here is just that that middle ground right which is like

所以我觉得我在这里想涉及的很多就是那个中间地带，对吧，就是

73:45

There are there is a class of problems

有一类问题

73:47

Then if you've put in thoughtful design into your agent readiness

如果你在agent readiness上做了深思熟虑的设计

73:50

You've put in thoughtful design around your bearifiability. There's a class of problems that you can feel comfortable

你在可审计性方面做了深思熟虑的设计，有一类问题是你可以放心

73:55

You know some percentage of the time essentially auto-approving

你知道，基本上有一定比例的时间是自动批准的

73:58

And what has to be true from a kind of a design perspective from like an architecture perspective

而从某种设计角度、架构角度来说，什么条件必须成立

74:03

To just auto-approved and when do you have to insert yourself and say actually here?

才能直接自动批准，什么时候你又必须介入说，实际上这里

74:06

I still want to review this carefully and even if you do review it carefully

我还是想仔细审一下，而且即使你仔细审了

74:09

Are you still doing line-by-line review or are you doing something else right?

你还是在逐行审查，还是在做别的事情，对吧？

74:12

So yeah some of this we one of our hottest topics of this year

所以是的，其中一些是我们今年最热门的话题之一

74:15

We have a guest writer program for lane space

我们有一个客座作者计划，面向 lane space

74:19

This is from Ankit from maybeator and he has a whole thing on killing the code review because I you know that that is

这篇来自 maybeator 的 Ankit，他有一整篇关于干掉代码审查的内容，因为你知道那就是

74:24

Maybe the final boss. Yeah of that's a bottleneck. He has a few

也许是最终 boss。对，那是个瓶颈。他有几个

74:30

Gates by which he does this. I don't know if you've talked with Ankit or yeah, no, I've read this post

他做到这一点所通过的关卡。我不知道你有没有和Ankit聊过，或者，对，没有，我读过这个帖子

74:35

Yeah, yeah, yeah, I haven't spoken to him personally, but I've definitely wrote the criticisms

对对对，我没有亲自和他聊过，但我肯定写过那些批评

74:40

No, I think this is right. I mean like broadly right. I guess you mentioned one thing like the last boss

不，我觉得这是对的。我是说，大体上是对的。我猜你提到了一件事，比如最后一个boss

74:43

Maybe not I think there's two choke points

也许不是，我觉得有两个瓶颈

74:45

There's the review and there's probably like the intent or like kind of the thing at the top right like this is that there's some

一个是review，然后可能还有intent或者类似最上面的那个东西，就是说有一些

74:50

There's some stats as well that I like to pull up you know everyone's stats is different for sure for sure

还有一些统计数据我也想调出来看看，你知道每个人的统计数据肯定都不一样，肯定的肯定的

74:54

But like clearly more people are merging clearly the number of reviews are not increasing in line with yeah

但很明显更多人在merge，很明显review的数量没有跟上

75:01

Really with merging. Yeah, that's right. That's right. That's right. And if you look at any modern

真的没有跟上merge。对，没错。没错。没错。而且如果你看任何现代的

75:05

Sort of company that's that's that's really invested in this and thought about this carefully

就是那种真正在这方面投入了很多、认真思考过这个问题的公司

75:09

And some people have already on post about this and you know, I'm sure we'll do post about this as well

有些人已经发了关于这个的帖子，你知道，我相信我们也会发相关的帖子

75:13

But I think recently like the amplitudes and I think cloud flare has has done a good one

但我觉得最近像 Amplitude，还有我觉得 Cloudflare 做了一个很好的

75:18

I think augment has also done a good one where they talk about the design of their air code review system

我觉得 Augment 也做了一个很好的，他们聊了他们 air code review 系统的设计

75:23

These are a bunch of really good if you want to share some of those links. I think they're fantastically written posts

这些都是一堆非常好的，如果你想分享其中一些链接的话。我觉得这些都是写得非常精彩的帖子

75:28

Yeah, so shout out to actually I'm shout out to cloud flare because I think that

对，所以要特别感谢一下，我要特别感谢 Cloudflare，因为我觉得他们

75:32

Very like kind of like you know in a sense like legacy company or like you know bigger company

很像那种，你知道，在某种意义上像是传统公司或者说是更大的公司

75:36

But I think they've done a fantastic job in terms of just their their content around the AI the AS DLC

但我觉得他们在 AI 和 AS DLC 相关的内容方面做得非常出色

75:42

So I think they have a good post there

所以我觉得他们那篇帖子写得不错

75:43

But till they are yes, I think it's true. I think that there's we're still fine finding that sweet spot

但直到他们做到之前，是的，我觉得这是对的。我觉得我们还在寻找那个最佳平衡点

75:49

What is auto-approvable?

什么是可以自动批准的？

75:52

What is not really auto-approvable? And if it is not auto-approvable

什么是不能真正自动批准的？如果不能自动批准

75:55

Is even in that world of something that has to be somewhat human reviewed? What do we review?

即使在那样的世界里，是不是也得有人工审核的部分？我们审核什么？

76:00

Are we still doing the line by line? Are we doing some some combination of the spec and you know, I think this is yeah

我们还在逐行审核吗？还是在做某种spec和你知道的某种结合，我觉得这是，是的

76:06

Yeah, there's a whole separate question there as well, but are people keeping

是的，那里还有一个完全独立的问题，但人们是否在保持

76:11

Are you guys keeping the spec up to date at all times?

你们是否一直保持spec是最新的？

76:15

The spec is in the way right? Yeah, no well, I mean okay

规范挡路了，对吧？对，不是，嗯，我是说好吧

76:18

In a few different places, right? So like there's a spec that as it exists in like a linear issue

在几个不同的地方，对吧？比如说有一个规范存在于 linear issue 里

76:21

And there's a spec as it exists in like the context of a PR description, right?

还有一个规范存在于 PR description 的上下文里，对吧？

76:25

So you can like you can have your peer description be the place where you like describe

所以你可以让你的 PR description 成为你描述的地方

76:29

More or less the entirety of the implementation in a more condensed format and as long as that's the most rough

或多或少把整个实现用更精简的格式描述出来，只要那是最粗略的

76:33

That's most updated thing representing this point in time of the code base

最能代表代码库这个时间点的最新东西

76:36

I think that's sufficient actually

我觉得那其实就够了

76:37

So we're thoughtful about like our peer descriptions

所以我们很用心地对待我们的 PR description

76:40

We make them be not just like

我们让它们不只是像

76:41

You know one paragraph. It's more like it is effectively like a spec of what has happened in the PR and there's even like kind of line

你知道的一段话。它更像是实际上相当于一个关于PR中发生了什么事的规格说明，甚至还有那种行级别的

76:48

File level pointers to things that were things have changed so you not only get the high level of what's happened

文件级指针，指向那些发生了变化的东西，所以你不仅能看到发生了什么的高层概览

76:53

But even the kind of the construction of the files and sort of the service level boundaries that you're touching

还能看到文件的构造以及你所触及的服务层边界

76:58

So you know, I think that's where we we've invested quite a bit

所以你知道，我觉得这就是我们投入相当多的地方

77:00

So I push back on that because to me

所以我对那点持保留意见，因为对我来说

77:03

Of course, it's you know

当然，你知道

77:05

Bare minimum it should be internally consistent with the PR spectrum. Yeah

最低限度它应该与PR的范围内部一致。是的

77:08

The no what I meant is the spec for the feature the code base because that is the summation of all PRs

不，我的意思是这个feature的spec，相对于整个code base来说，因为code base是所有PR的总和

77:14

That is like if you're if your PR spec got it wrong or miss something

这就好比，如果你的PR spec搞错了或者漏了什么

77:17

It's going to be the delta between what the whole code base or the whole history of the code base is supposed to do versus what your PR is trying to do

它就会变成整个code base或者说code base的整个历史应该做什么，跟你的PR试图做什么之间的delta

77:25

So so I like I mean, I don't expect you to have an answer

所以我就是，我不指望你能给出答案

77:27

Yeah, yeah, because like I really stress about this myself. I tried to instruct models to do it. They're bad at it

对对对，因为我自己就特别为这个焦虑。我试着让模型去做这件事，它们做得很差

77:33

That's it

就是这样

77:34

So the humans are the the spec gods of like well

所以人类才是spec的上帝，就像

77:39

Five weeks ago. I did this other thing and now you're reversing it. Have you seen the the post from basis like they have a they had a tweet

五周前我做了另一件事，现在你又在推翻它。你看到basis的那个帖子了吗，他们发了一条推文

77:46

Tweet little thing

推文小东西

77:47

Tweet article. I thought they did a great job talking about

推文文章。我觉得他们讲得很好

77:50

Like

就像

77:51

Different sort of file artifacts. They they they track via get and some of them are like the specs the intent of what is to be

不同种类的文件产物。他们通过git来追踪，其中一些像是规格说明，是关于要做什么的意图

77:57

And the other is like what they call like the canonical state of the code

另一个是他们所谓的代码的规范状态

78:00

Which is like a little bit more maybe what you're describing which is

这也许更接近你所描述的

78:03

Implicit knowledge implicit intent that also has to be codified somewhere right so that part

隐性知识、隐性意图，这些也必须被记录在某处，对吧，所以那部分

78:08

That's all version controlled and I think they found that to work for them

全部都在版本控制之下，我觉得他们发现这对他们很有效

78:11

So that might be another thing. Yeah as well if you was trying to pull that this is basis the accounting company

所以那可能是另一件事。对，而且如果你试图把那个说成是这家会计公司的基础

78:15

Exactly right. Yeah, yeah, we're we have a Mitchell with my newer conference amazing soon

完全正确。对对，我们有一个Mitchell和我新一点的会议，很快很棒

78:20

Yeah, I like that they're coming up with this my solution for is is

对，我喜欢他们提出这个我的解决方案是

78:26

I don't know if like how ready I am to talk about it because I mean ever ship it

我不知道我有多准备好谈论它，因为我的意思是曾经发布过它

78:30

Is something that is replacing Git in order to capture all of this

是某种替代Git的东西，为了捕捉所有这些

78:37

So I call it a program database where it captures all traces and all and and rayify aspects to something that that we that we can ignore and keep

所以我叫它程序数据库，它捕捉所有痕迹和所有以及可射线化的方面，变成我们我们可以忽略并保留的东西

78:45

And it also controls the skills that are attached to it as well

而且它也控制附加到它的技能

78:49

Because I think like basically what Git Git is only focus on code and then you just need a source of truth that is

因为我觉得基本上Git只专注于代码，然后你只需要一个真相来源

78:55

Not only

不仅仅是

78:57

Everything else outside the code including the implicit what would you like implicit assumptions are like the constraints the intent yeah other stuff

代码之外的所有其他东西，包括那些隐含的——你想要的隐含假设，比如约束条件、意图，对，还有其他东西

79:03

Yeah, but also

对，但还有

79:05

Multi like highly concurrent right because Git Git by introduced by design introduces mutex is everywhere which sucks

多重——像高度并发的，对吧，因为Git在设计上到处引入了mutex，这很糟糕

79:13

Yeah, so so you need like super concurrency

对，所以你需要超级并发

79:16

Have you seen um to get Butler is it there's like yes

你有没有见过，嗯，Git Butler，是不是有这么一个——对

79:20

Who used to be CT?

谁以前是CT？

79:22

Ian I think is his name forget his name but yeah Scott Scott

Ian，我觉得这是他的名字，忘了他的名字，但对的，Scott，Scott

79:25

So yeah, so I think there's a there's a lot of people going after this

所以是的，我觉得有很多人在追这个方向

79:30

um, and I think there's the there's the question of like really what is the ideal representation of tasks?

嗯，然后我觉得问题在于，任务到底什么才是最理想的表示方式？

79:36

Yeah, and I think the winner probably not me linear

是的，而且我觉得赢家大概率不是我，Linear

79:40

Really

真的吗

79:41

Because like that's where the task are yeah, yeah, uh, and like clearly they have decided to become a

因为那才是任务所在的地方，对对，而且很明显他们已经决定要成为一家

79:47

A coding agent company right and well, you know that they have

编程agent公司了，对吧，而且你知道他们有

79:51

All the sorts of truth there as far as you as as long as you buy into linear

各种各样的真相，只要你在用Linear

79:55

Yeah, yeah, that that's true. That's true

对对，确实是这样。确实是这样

79:57

I wonder I had like a this somewhat like kind of held belief that I think a lot of companies are gonna evolve into some version of coding agent company

我一直在想，我有个有点像是一直以来的信念，就是我觉得很多公司都会演变成某种版本的coding agent公司

80:04

Which is any company that has like a interact with b2b software where you're collecting something about the user

就是任何一家有跟b2b软件交互的公司，你在那里收集一些关于用户的东西

80:09

Or some like product insight. I think you can also close the loop because ultimately like what a coding agent is the purpose of coding agents to act on some

或者一些类似产品洞察的东西。我觉得你也可以闭环，因为归根结底，coding agent是什么，coding agent的目的是对某些

80:15

Adventure thing that's happening where you users involved or customers involved and they use that to like fix

正在发生的事情采取行动，用户参与其中或客户参与其中，然后他们用那个来修复

80:20

Ideally some master to the product right? I think what linear is doing

理想情况下某种对产品的主控，对吧？我觉得Linear正在做的事

80:23

I imagine the same thing like an amplitude could do as well right because they're also getting insights about user analytics user problems

我想象同样的事情，像Amplitude也可以做，对吧，因为他们也在获取关于用户分析、用户问题的洞察

80:29

I imagine even like you want to really stretch it you could even have like a decogon or serum do something similar where you know

我想象甚至，如果你想真的延伸一下，你甚至可以让Decagon或Serum做类似的事情，你知道

80:34

Here's what's broken go close the loop

这里是坏掉的东西，去闭环

80:36

Yeah, all right

好，行吧

80:38

So set all those things up for agent ready code base and set up code review securities week seven. It's not a week one

所以把那些东西都设置好，做成agent ready的代码库，然后把代码审查安全放到第七周。这可不是第一周

80:45

What do you mean security is not important

你说安全不重要是什么意思

80:48

I make my choice. I always say like you know at amazon job zero is security and you always say this

我做出我的选择。我总是说，你知道在amazon，job zero就是安全，你总是这么说

80:53

Yes, yeah, yeah security review at the end. I was at amazon and I don't remember that, but I believe this is the thing Andy Jesse says

对，对对，安全审查放在最后。我在amazon待过，我不记得这个，但我相信这是Andy Jassy说的

81:00

Okay, okay, okay, yes, yes, yeah, and I was I guess he I was there partly during Jeff's time and partly

好，好，好，对对对，然后我当时，我猜他，我在那儿一部分时间是Jeff的时代，一部分是

81:06

When this transition to Andy so but yes, I I think um

过渡到Andy的时候，所以但是对，我觉得嗯

81:11

This is a thing a topic last year as well, and it's it's also won this year and I think that there I

这也是去年的一个话题，今年也赢了，我觉得那里有

81:17

Isaac also spoke last year. I find that he's an incredibly thoughtful thinker

Isaac去年也讲过。我觉得他是一个思考非常深入的人

81:21

Thoughtful thinker. I was about it. You know a thoughtful thoughtful leader in sort of this space

思考深入的人。我当时就在想。你知道，在这个领域里算是一个很有深度的领导者

81:25

I think Sam grub is an interesting company because they've been around even before sort of an agents, right?

我觉得Sam grub是一家很有意思的公司，因为他们在agents出现之前就已经存在了，对吧？

81:29

I mean they're the OSS repo was well

我是说他们的OSS repo远远早于

81:32

before like kind of traditional sast kind of

像传统sast那种

81:35

Defense mechanisms

防御机制

81:36

But I think they've been a good job with pivoting and are not pivoting really augmenting their offering into this space

但我觉得他们在转型方面做得很好，其实也不算转型，更像是把他们的产品扩展到这个领域

81:41

You know, I think it's it's an interesting time, right?

你知道，我觉得现在是个很有意思的时期，对吧？

81:45

There's just a lot going on where these these attack vectors are now kind of come to the surface

现在有太多事情在发生，这些攻击向量现在算是浮出水面了

81:49

There's a lot more slop and I generated code and I think

有更多的slop和AI生成的代码，我觉得

81:53

Maybe one of the core insights that I want to drive here is just the idea of like

也许我想在这里强调的一个核心洞察就是

81:58

dependency kind of dependency tracking and supply management basically like supply chain, right?

依赖追踪和供应管理，基本上就是供应链，对吧？

82:03

You have to be careful about like your dependencies and pin things appropriately and don't just arbitrarily hit NPM install and expect the chain to go

你必须小心你的依赖关系，适当地pin住版本，不要随便npm install然后指望整条链正常运转

82:10

Yeah, so potentially the most secure code base is one of no dependencies. Yes, right?

对，所以潜在来说最安全的代码库就是没有依赖的代码库。是的，对吧？

82:15

You look at the dependency. You're like cool

你看看那个依赖，你说 cool

82:18

Cool implementation, bro, but I'm just going to make my own that's upset right that's upset right and and that's an interesting insight because

实现得不错兄弟，但我要自己写一个，这让人不爽对吧，确实让人不爽，这是一个有趣的洞察因为

82:25

It's sort of to the limit and you know, I think it's it's probably easier said than done

这差不多到极限了，你知道，我觉得这说起来容易做起来难

82:30

But if you believe that with code now being commoditized code being easy cheap

但如果你相信现在代码已经商品化了，代码变得容易又便宜

82:34

Why not just build everything in house, right? Like that like of course

那为什么不全部自己内部构建呢，对吧？就像那样，当然

82:37

And that would be the safest thing because you've seen everything you control everything instead of you know

那会是最安全的做法，因为你什么都见过了，你控制一切，而不是你知道

82:41

Depending on axios or whatever

依赖 axios 或者别的什么

82:42

But you know, we'll see how many people actually act on that that is the safest thing I agree with you

但你知道，我们看看有多少人真的会照做，那是最安全的做法，我同意你

82:46

No, I so that statement is dangerous

不，我所以那个说法很危险

82:49

Because then you would just make your mistakes instead of making up

因为那样你只会犯自己的错误，而不是弥补

82:52

Relaying other people to discover mistakes the whole point of open source is that you know, what is it like sunlight is the best disinfected

让别人来发现错误，开源的核心就在于此，你知道，就像阳光是最好的消毒剂

83:01

It's like Linus Torval's thing

就像Linus Torvalds说的那样

83:02

And it's like well

然后就像是

83:04

Yeah, you will be very slow to discover your own custom software dependencies because you only have your users and they won't report it to you

是啊，你发现自己的定制软件依赖问题会非常慢，因为你只有你的用户，而他们不会向你报告

83:10

But if this this thing has a hundred thousand users worldwide someone will find something

但如果这个东西在全球有十万用户，总会有人发现问题的

83:15

And then you can depend on it more right

然后你就可以更依赖它了对吧

83:17

Yeah, this is

是啊，这是

83:18

Yeah, there's no that basically I don't be I'm trying to be a voice of

是啊，基本上没有，我不想，我是在试着做一个声音

83:23

Reason right and you cannot be absolutely involved and you I kind of trapped you there

你推理得对，你不可能绝对地卷入其中，而我在那点上有点把你困住了

83:28

I talked I talked to you into the talk track of yeah vendor everything right right and there is a bug there

我跟你说过，我跟你说过那个话术，就是什么都外包，对吧对吧，那里有个bug

83:33

Yeah, yeah, yeah, it's an interesting one

对对对，这是个有意思的问题

83:36

I wonder how much that that how much is that I buy some aspect of that but how much of that is impacted by the fact that everything

我想知道那有多少，那有多少，我买账其中某些方面，但其中有多少是受这个事实影响的，就是一切

83:42

I have his belief that like most OSS repos if not all them will just become software factories that are effectively agent run in the future

我有一种信念，就是大多数OSS仓库，如果不是全部的话，未来都会变成软件工厂，实际上由agent来运行

83:49

then where like

那到时候就像

83:51

I mean even

我是说甚至

83:53

Like even if you look at warp right like warp which is a monitoring agent

就像甚至你看warp，warp这个监控agent

83:55

Of course, they're also you know selling a software factory product

当然，他们也在卖一个software factory产品

83:58

But but they are running their warp open source warp terminal as a software factory

但他们是在把他们的warp开源warp terminal当作一个software factory来运营

84:02

And that almost suggests that there's a part where yes, he's are so involved

这几乎暗示了有一部分是，是的，他参与得非常深

84:06

But increasingly less so certainly not every single person is reviewing every single thing that comes into the maybe the same exact level that we did in the past

但越来越少，肯定不是每一个人都在审查进来的每一件事，达到我们过去那种完全相同的程度

84:13

Does that make this could base more or less

这会让这个代码库更容易还是更不容易

84:16

Vulnerable to securities. It's a bit hard to say right

受到安全漏洞的影响。这有点难说，对吧

84:18

I think that's just where there's a little bit of nuance that I think we're still

我觉得这就是有一点细微差别的地方，我认为我们还在

84:22

And even the folks that I know that it's that working out assess or trying to figure out exactly what that stable point is

甚至我认识的那些正在努力评估或试图弄清楚那个稳定点到底是什么的人

84:26

Yeah, there's a couple of related arguments and then I'm happy to keep going

是的，有几个相关的论点，然后我很乐意继续聊下去

84:31

Like security is like a non-stop like huge huge huge huge surface area thing

比如安全就是一个永不停歇的、巨大巨大巨大巨大的攻击面问题

84:35

There's one argument from Malta of Versailles who's like okay guys

Malta of Versailles有一个论点，他说好吧各位

84:40

Let's take every single like highly dependent highly dependent on software that's written in cc++

让我们把每一个高度依赖、高度依赖C和C++写的软件

84:47

Ported to rest and it's like will be faster more secure

都移植到Rust，这样会更快更安全

84:50

And like it won't probably take like six million dollars and whether you get it done and and then there's there's another

而且大概不会花掉六百万美元，不管你能不能搞定，然后还有另一个

84:57

Argument for this is also things to pacing the frontier from the open source from the model labs

这方面的论点是关于从开源、从模型实验室那边控制前沿的节奏

85:02

Who are like okay, let's slow down so that we

他们说好吧，让我们放慢速度，这样我们

85:04

Or like we'll do daybreak. We'll do what's a glass wing will make sure everyone's patched

或者像我们会做daybreak。我们会做的是，glass wing会确保每个人都打好补丁

85:10

And then we can let loose on rampant acceleration

然后我们才能放手去搞那种狂飙式的加速

85:14

So I think like there's there's there's an inherent belief that the models will fix all security

所以我觉得，有一种内在的信念，认为模型会解决所有安全问题

85:19

And I think it's just unproven to me because what if they introduce new bugs and they're just different bugs

我觉得这对我来说就是未经证实的，因为万一它们引入了新的bug，而且只是不同的bug呢

85:26

That's right that's that's that's so so I mean well

没错，就是就是就是，所以我的意思是

85:29

I'm looking for pushback. I'm looking for like one. No, you're you're an idiot like obviously in models are way better than this

我在找反驳意见。我在找那种，不，你你个白痴，很明显模型比这强多了

85:37

Yeah

是啊

85:39

I guess like that's almost like in the limit. There's the question is like is is a model

我猜这几乎就像在极限情况下，问题在于，模型是不是

85:44

Do we believe model will create perfect software right if the model is pretty softer than then you're such a thing

我们相信模型能创造出完美的软件吗，如果模型比那还软，然后你就有这种东西

85:48

Then then that's done right then then we have then we're done here at all

然后那就完成了，然后我们就彻底完事了

85:51

Like then then then we're just gonna go home and and chill in the beach with us to realize I think there's also the

就像然后我们就回家，跟我们一起在海滩上放松，然后才意识到我觉得还有那个

85:56

Question off is it good enough to first give it for security?

问题就是，它够不够好，先把它交给安全方面？

86:00

So let people patch then just delay the release and do it after or is there a problem of

所以让人们打补丁，然后只是推迟发布，之后再去做，还是说有个问题

86:06

Oh, no, it'll be much better at finding vulnerabilities. You can

哦不，它在发现漏洞方面会好得多。你可以

86:09

Full-send test it for longer even if you've patched or like

全力测试更长时间，即使你已经打了补丁，或者像

86:13

How long right? Yeah, you can give it for three months. Someone can try to exploit it for six months

多长时间对吧？对，你可以给它三个月。有人可以试着利用它六个月

86:18

The problem's not solved right right. Yeah, yeah

问题没解决，对吧对吧。对对

86:21

My whole problem with like and again like this is a whole rant that we don't need to go into but like it's like you know

我对这个的整个问题，就像，再说一次，这整个吐槽我们不需要深入，但就像，你知道

86:25

You name 100 blessed organizations that like okay, these are like priority. They they have access to cyber model that

你点名100个受祝福的组织，就像好吧，这些是优先级的。他们能访问你刚创建的cyber model

86:33

You've just created a hundred attack vectors for me to go compromise and like they're by the way they're pretty easy to compromise

你刚刚给我创建了一百个攻击向量让我去攻破，而且顺便说一句，它们还挺容易被攻破的

86:40

Yeah, yeah, I don't know anyway

对对，我不知道，反正

86:43

That above our pay grade

那超出了我们的薪资级别

86:46

It's always fun to get juicy and you know security like

搞点刺激的总是很有趣，你知道，安全这块就像

86:50

Yeah, me security is right now very juicy right like and I think actually the security folks love it because they like they usually don't get

对，安全现在非常刺激，对吧，而且我觉得实际上安全圈的人很喜欢，因为他们通常不会得到

86:55

That much attention and now they have like a ton of attention is that true like I mean I mean that they don't get attention

那么多关注，现在他们有了一大堆关注，这是真的吗，我是说我是说他们没得到关注

87:00

I find that um I find that at least from like a VC perspective. You do love

我发现，嗯，我发现至少从VC的角度来看。你确实喜欢

87:04

VCs love investing security companies, and I'm like pretty good

VC喜欢投资安全公司，而我还挺厉害的

87:07

I'm like a whiz or something like it's done very very well

我像个天才还是什么的，做得非常非常好

87:09

With this with argue with the socket

用这个跟socket争论

87:12

I literally I had this conversation of VC two days ago. Yeah, apparently wisdom socket will need security companies in the last five years

我真的，我两天前跟VC有过这样的对话。是啊，显然wisdom socket在过去五年需要安全公司

87:18

That have worked. What about like abnormal?

那些成功的。那abnormal呢？

87:21

I don't know what it worked for a VC point of view of I see from like from like an economics perspective. Yeah, yeah

我不知道从VC的角度来看什么算成功，我从经济学角度来看。是啊，是啊

87:27

Interesting

有意思

87:29

Not for me to say right obviously many like great players in the field, but it's a tough tough industry

这不该由我来说，对吧，显然这个领域有很多很厉害的玩家，但这是一个非常非常艰难的行业

87:35

I think so basically like the best time but I think it's not controversial best time for any security company is when a high profile incident just happened

我觉得基本上可以说，最好的时机，但我觉得这没什么争议，对任何安全公司来说最好的时机就是刚发生了一起高关注度事件的时候

87:43

Yeah, yeah, right right right then obviously you're gonna sell a bunch of security

对对对，没错没错没错，那显然你就能卖出一堆安全产品

87:48

Yeah, so I think I think that's

对，所以我觉得我觉得这

87:51

A fair thing like people you people should be aware people should know the the main basic mitigation

是合理的，大家应该意识到，大家应该知道最基本的缓解措施

87:56

I do think like co-scanning is important, but like I just want to give people the idea that like models introduce bugs

我确实认为代码扫描很重要，但我只是想让大家意识到，模型会引入一些

88:03

That you wouldn't make as a human and like if you're not checking those bugs

人类不会犯的bug，如果你不检查这些bug的话

88:08

It's not even that they make bugs. It's that you have less awareness of what's being written right like when

甚至不是它们会制造bug。而是你对正在写的东西了解得更少，就像当

88:15

You know you review everything. Yeah, forget bugs models are pretty good. They'll have on average

你知道你会审查所有内容。对，忘了bug吧，模型挺不错的。它们平均来说

88:20

I would say less bugs the problem is also that you don't even know what you're putting out

我会说bug更少，问题还在于你甚至不知道自己发布的是什么

88:25

I like it

我喜欢这个

88:27

Yeah, there's a lot of that. Yeah, and that's the I think we're going to do top of mitigations and like the and there's sort of a combination of security and

对，这种情况很多。对，而且我觉得我们会做最重要的缓解措施，然后还有安全和

88:34

Seerity as part of could review as well, right? I mean this is also like doing those checks. Yeah

安全性的某种结合，作为代码审查的一部分，对吧？我是说这也是在做那些检查。对

88:38

Yeah, so I think um maybe I would say one more thing here or or is check your check in check my intuition on on this thing. Yeah, is that

对，所以我觉得嗯也许我在这里再说一点，或者说是检验一下你对这件事的直觉。对，就是

88:46

not all parts of the code base is created equal the parts of the code base where security is very important is data storage

代码库的各个部分并不是同等重要的，代码库中安全性非常重要的部分是数据存储

88:54

yeah, and

是的，而且

88:55

billing so let's call it that data

billing，那就这么叫吧，data

88:58

billing

88:59

off

89:00

And auth is obviously related to all those things and those are basically don't vibe could and then everything else go go nuts

而auth显然跟所有这些都相关，那些基本上就是don't vibe could，然后其他一切就随便搞

89:08

Yeah, right. I think that's absolutely right

对对对。我觉得这完全正确

89:10

And and it's really funny how like okay, well if you don't know what the difference between auth and an off z is

而且真的很有意思，就好像好吧，如果你不知道auth和off z之间的区别

89:16

Please learn it and then you know if you if you don't know if you have never looked at AWS. I am which you have please learn it

那就去学一下，然后你知道，如果你如果你不知道，如果你从来没看过AWS。我是指你确实看过的话，那就去学一下

89:22

Like this like these are basic like

像这样，这些都是基本的

89:25

Conceptual things to understand that you can prompt for no one's you have the language you can prompt for it

概念性的东西，你得理解你才能去 prompt，没有的话，你有语言你就能去 prompt

89:30

Don't have the language you're fucked

没有语言你就完蛋了

89:33

Yeah, yeah to understand that I don't even stand that to like just general like certain certainly cloud platform of things

对对对，理解这一点，我甚至都不理解那个，就像一般的，某些云平台的那些东西

89:37

Right, I mean there was the for which company got how that hue incident huge incident where like it was just like a DNS issue that brought down

对，我是说有那么一个，哪家公司出了那个 huge incident，巨大的事故，就像只是一个 DNS 问题就把东西搞挂了

89:43

Was it do you know this was a

是不是，你知道这个吗，这是一个

89:46

DNS issues

DNS 问题

89:52

Also been many prominent

也有很多著名的

90:00

So the NS is more like a reliability uptime, sure, you know security

所以NS更像是可靠性、正常运行时间，当然，还有安全性

90:05

I think people like you know, did you lose money? Did you leak customer information? Yeah, yeah

我觉得人们会关心，你亏钱了吗？你泄露客户信息了吗？对对对

90:09

Yeah, I would yeah, probably put that about that for sure only DNS because of uptime. Yeah, general pen. Yeah

对，我会把这些放在那上面，肯定的，只有DNS因为正常运行时间。对，一般渗透测试。对

90:15

background agents

后台agents

90:17

Now you got your buddy from call for here, but I mean your guests aren't necessarily correlated with that week

现在你请来了Call for的哥们儿，但我是说你的嘉宾不一定和那一周有关联

90:21

Yeah, no, I mean like the 9-1 was a little bit of a scheduling thing where I've thrown in you know elad and and I'm

对，不是，我是说9-1那个有点像是排期的问题，我插入了你知道Elad然后我

90:27

I guess I'm John is isn't it would definitely week 9 elad I was trying to get him at the end

我猜我是John，不是吗，那肯定是第9周Elad，我试着把他安排在最后

90:31

But I think it's just timing wise I don't work it out. So yeah, it's a bit a weird one where I don't I guess

但我觉得就是时间上没对上。所以对，有点奇怪，我猜

90:35

I'm not lecturing at all week 9, but but we're coming back to those topics in week 10

我第9周完全不上课，但第10周我们会回到那些话题

90:39

So it's a little bit staggered in a bit of bizarre way there, but yeah week 8

所以那里有点交错，有点奇怪，但没错第8周

90:43

Rajesh who has been leading a lot of like the infrastructure at

Rajesh，他一直在领导很多基础设施方面的工作，在

90:46

Cloud Cloud infra

90:48

Devex it's and beyond at Cloud Flair, which as I earlier said was I think they're doing a great job here

Devex以及Cloud Flair之外的工作，正如我之前说的，我觉得他们在这里做得很好

90:53

And definitely their adoption first a company of their size and a public company of their size is fantastic

而且绝对是他们的采用，作为他们这种规模的公司，作为他们这种规模的上市公司，非常棒

90:58

And then their content I think is also really good. I mean this is fundamentally about

然后他们的内容我觉得也非常好。我的意思是这从根本上讲是关于

91:02

All the ways you can use like asynchronous background agents

所有你可以使用异步后台agents的方式

91:05

I think I'm also going to spend quite a bit of time talking about you know the stripe minions the Spotify

我觉得我也会花不少时间聊聊你们知道的stripe minions、Spotify

91:10

If I get the name for their version of this

如果我没记错他们那个版本的名字的话

91:13

The ramp inspects the you know things of that nature that are I think around

ramp inspects之类的你们知道那类东西，我觉得大概是在

91:16

Kind of showing what is it look like when you scale this out

展示当你把这个scale out的时候它是什么样子的

91:19

Instinct

91:21

Yeah, yes

对对对

91:24

Yeah, I mean I guess less on consumer more on like you know software teams using them at scale

对，我是说我想可能少聊点consumer，多聊点像你们知道的software teams在scale下使用它们

91:31

They're all coding agents under hood

它们本质上都是coding agents

91:32

Everything is a code yeah, it's like rock bot is one of those interesting things where like it's very very much presented as consumer e

一切都是代码，对吧，Rock Bot 就是那种很有意思的东西，它非常非常被呈现为消费级产品

91:41

It's it's rock kind of it's you know, it's it's cursor under hood. Yeah, yeah, it's a cursor team

它就是 Rock，你知道的，它就是底层是 Cursor。对对，是 Cursor 团队做的

91:44

It's no no for sure like so yeah, I've seen the like you know cursor working and in the like the data dots

不不，肯定的，所以是的，我见过 Cursor 在工作中的样子，还有那些数据点

91:49

But but but they are not they don't have like a code base integration, right?

但但但它们没有代码库集成，对吧？

91:52

They do okay, I haven't used that

它们有，好吧，我没用过那个

91:55

I've only been using for like open table reservations exactly

我只用过它来订 OpenTable 的位子，没错

91:58

What I think that's the point of what they're trying to do I see yeah as in like they're trying to be like

我觉得这就是他们想做的东西的核心，我明白了，就是他们想做成那种

92:05

Well, you can use it for consumer stuff

嗯，你可以把它用在消费级的东西上

92:07

But here's the extra power if you know how to use it well

但如果你知道怎么好好利用它，这里还有额外的能力

92:10

Because this is actually cursor under hood. I think this goes to all coding agents right like

因为这实际上底层就是cursor。我觉得这适用于所有coding agents对吧

92:14

Not code cloud design

不是code cloud design

92:16

People using codex not knowing they're using codex right a random code interpreter window pops up and you're just asking

用codex的人不知道自己用的是codex对吧，一个随机的code interpreter窗口弹出来你就直接问

92:24

Just to be a question at some level. There's coding agent under the hood, right? Yeah, so I'll give you another sort of practical example

某种程度上就只是问个问题。底层有coding agent对吧？对，那我再给你一个比较实用的例子

92:32

So I run all my SEO and a yo inside of codex and it has a code base even though like a traditional marketer would just check a address or something

所以我把所有的SEO和a yo都跑在codex里面，它有一个code base，尽管传统营销人员可能就查个地址什么的

92:41

Because I write all these scripts for doing all the all sorts of crazy stuff. Sure

因为我写所有这些脚本来做各种疯狂的事情。当然

92:44

And and also I need to store data and also I need to visualize data and I pretty soon you go like as long as you're

而且我还需要存储数据，还需要可视化数据，很快你就会发现只要你

92:51

Doing anything on an ongoing basis that has like maybe a lot of steps or a lot of data you need a code base

在做任何持续性的事情时，如果有大量步骤或大量数据，你就需要一个代码库

92:58

Hmm, that is an interesting take. I think models also like building code basis

嗯，这是个有趣的观点。我觉得模型也喜欢构建代码库

93:04

No, but that could be a temporary thing

不，但那可能只是暂时的事情

93:06

I guess maybe my my photos more on the interface right like will are you going to so like are you going to go so far to say that a

我猜也许我的照片更多是在界面上，对吧，就像你会不会，你会不会走到那么远说

93:12

Grockbutt will have a place where like the power user. We want to see the code under the hood

Grockbutt会有一个地方让高级用户，我们想看到底层的代码

93:16

Is that or is it?

是这样还是说？

93:18

Maybe they'll just jump to cursorize. I guess is that what you say exactly or because I'm like otherwise then then you basically

也许他们会直接跳到cursorize。我猜你是这个意思吗，还是因为我觉得否则的话你就基本上

93:23

Cannibalize cursor like if in a limit what you're saying is Grockbutt is the cursor

蚕食了cursor，就像如果最终你说的是Grockbutt就是cursor

93:27

Then you don't even need like a cursor if all under the hood is just becoming

那如果底层全都变成这样，你甚至都不需要一个cursor了

93:30

It's the same team so maybe right now they don't care

是同一个团队，所以也许现在他们不在乎

93:32

But yeah, I buy that I mean I do accept that even things that seem outwardly just like non-technical use cases under the hood

但是的，我信这个，我是说我能接受即使是那些表面上看起来非技术性的使用场景，底层

93:42

Our technical can be like written and designed and loaded in code like that. That much is absolutely true

我们的技术可以像那样被编写、设计和加载到代码里。这一点绝对是真实的

93:47

Right the question is just more like is a developer is that they're a single point of entry

对，问题更多是在于开发者是不是一个单一的入口点

93:52

Yeah, I think it's not even developer right like you think of a regular person doing marketing that is using clutch code or just

是的，我觉得甚至不是开发者，你想想一个做市场营销的普通人，在用clutch code或者就是

93:58

Claude in the app now is working in a code base that they would never know never touch but

现在app里的Claude，他们在一个他们永远不会知道、永远不会碰的代码库里工作，但

94:04

There's a code base management

有一个代码库管理

94:06

Yeah, I think cloud tag is a good one for

是的，我觉得cloud tag这个说法挺好的，对于

94:08

background agents. It's a big way and throw pick works and

background agents。这是一个很重要的方式，而且 throw pick 也能行，并且

94:12

You know they make it you should you see how they work. Yeah, good job

你知道他们是怎么做的，你应该看看他们是怎么运作的。是的，干得好

94:17

Okay, I need a team we talked a lot about these topics of like gateway

好，我需要一个团队，我们聊了很多关于gateway这类话题

94:21

We just put a routing episode out today and mcp portal

我们今天刚发布了一期关于routing的节目，还有mcp portal

94:24

I would say mcp portal is like the least

94:27

Establish on mcp's there's not bullish of bearish. It's just is

在MCP上不存在看涨或看跌。就是客观存在而已。

94:33

Is a part of the landscape now, right? Yeah

现在已经是整个格局的一部分了，对吧？是的

94:36

Um, that's a statement. There was a whole mcp is so back mcp is dead people people like extreme mcp

嗯，这话说得挺绝对的。当时有一大波人说什么MCP回归了、MCP死了，人们就喜欢走极端MCP

94:43

Portals is an unproven business. I thought about starting one

Portals是一门尚未被验证的生意。我考虑过创办一个

94:47

I actually pitched a guide in topic on potentially starting one and then I lost interest

我其实提过一个关于可能创办一个指南的选题，后来就没兴趣了

94:51

I lose interest in a lot of a lot of things, but I what's what's in there?

我对很多事情都会失去兴趣，但那里面到底是什么？

94:55

Okay, so what I'm thinking here is if this is if this is the week does a more kind of gear toward

好，所以我这里想的是，如果这周更偏向于

95:00

Scaling this out to like what happens when you take some principles and a lot of this was informed by like the composition

扩展到这种情况，就像当你把一些原则拿出来会发生什么，而这里面很多是受到了组合的启发

95:06

I had with folks that were you know leading and eight of teams and it teams at

我曾经和那些人交流过，你知道，他们当时在领导八个团队以及IT团队

95:10

At respective companies and so here

在各自的公司，所以在这里

95:14

I think that something mcp portal really becomes interesting once you're at a certain scale where

我觉得mcp portal真正变得有意思的时候，是你达到一定规模之后，那时候

95:18

You effectively have to have a single point of entry to do things like governance security auditing cost optimization

你实际上必须有一个单一的入口点来做治理、安全审计、成本优化这些事情

95:25

And so you know I

所以你知道我

95:27

I'd be curious to hear what you see is the future or maybe lack there of something like this

我倒挺想听听你觉得这东西的未来是什么样，或者说可能压根就没有未来

95:31

But it to me is in this if you believe like an open router or you know

但对我来说，这在于如果你相信像Open Router或者你知道的

95:35

It's someplace like a single point of entry for for many things to happen and connect to and be effectively you know governed

它有点像是一个单一的入口，很多东西可以在那里发生、连接，并有效地被管理

95:40

I feel like there's a world where this also can easily exist

我觉得也存在这样一种可能，这个也能很容易地实现

95:43

Yeah, my only tweak on this is that this folds into the lm gateway

对，我唯一想调整的是，这个可以并入 lm gateway。

95:47

So they are the exact same platform that I will give you that I will you know, I can accept that that actually can agree with yeah

所以它们其实是完全相同的平台，我会给你的那个，你知道，我可以接受那实际上是可以同意的，对

95:54

And and then by the way, it also folds into the other thing of

然后顺便说一下，它也归入另一个问题

96:00

Whether I say it was it was off Z you know

不管我说它是不是偏离了Z，你知道

96:02

permissioning and all these things and billing

权限管理以及所有这些事情还有计费

96:04

So like I all to the limit all in for a converges to do the same exact problems where they have the box in the middle

所以就像我全都归结到极限，全都汇聚到同样的问题，就是他们在中间有个盒子

96:10

And then you have the box in two cents

然后你有那个盒子分两边

96:12

You know God box will solve everything right and for this true only if you take if you understand the job of the God box

你知道上帝盒子会解决一切，对吧，而这只有在如果你理解上帝盒子的职责时才成立

96:18

And so separating mcp portals and getways means you don't fully accept it that they are

所以把mcp门户和网关分开意味着你并不完全接受它们是

96:24

You know the the same layer problem and in a very similar way

你知道同样的layer问题，而且方式非常相似

96:28

I also find that people don't understand database querying

我也发现人们不理解数据库查询

96:33

To have a lot of the similar things where they will separate filtering from ranking from

有很多类似的情况，他们会把filtering和ranking分开

96:38

Or just like general recommendation if they're the same problem

或者就像一般的recommendation，如果它们是同一个问题

96:42

And so like then you just come because like when you take all these things you combine them in the same layer

然后就像，当你把所有这些事情放在同一个layer里组合起来

96:47

Then you can sort of cross depend on things in useful ways if they're different layers then they cannot

你就可以以有用的方式交叉依赖这些东西，如果它们是不同的layer就不行

96:52

Yeah, yeah, I actually will not disagree with this. I think um

对对，我其实不会不同意这个。我觉得嗯

96:57

Maybe the yeah, so it's a slightly pedagogical. I think that this breaking up of like separate terms versus one

也许，对，所以这稍微有点教学性质。我觉得这种把不同术语拆开 versus 合成一个

97:02

I think I it is absolutely correct that the underlying architecture is the the God box

我觉得底层架构就是那个God box，这绝对是对的

97:08

Where one thing goes in and there's a bunch of other stuff that is some third party some internal some first party that all kind of exists

一个东西进去，然后有一堆其他的东西，有些是第三方的，有些是内部的，有些是第一方的，所有这些都以某种方式存在着

97:14

On that is out of the God box. So yes, I agree with this actually

那些是从God box里出来的。所以是的，我实际上同意这一点

97:17

I have a

我有一个

97:19

I have a spicy

我有一个辛辣的

97:20

Spicy tweet. I oh god. Where where is the God box tweet?

辛辣的推文。我天。那个God box的推文在哪儿？

97:24

Oh, yeah, so this is the M M M action to M plus and let me bring it up architecture

哦对，所以这是M M M action到M plus，让我把架构调出来

97:30

So anyone working in front needs to understand

所以任何在前端工作的人都需要理解

97:33

Congratulations, you not have a job in infrastructure or you want to be an infrared engineer

恭喜你，你现在没有基础设施方面的工作了，或者你想成为一名infrared engineer

97:37

You need to understand that your entire job is making God boxes and then taking them apart

你需要明白，你的整个工作就是造God boxes，然后把它们拆开

97:43

So either your problem is either you don't have enough you don't have a good God box

所以要么你的问题是你没有足够的、你没有一个好的God box

97:47

Or your problem is someone else has got box

要么你的问题是别人有了box

97:49

You don't like so now you have to take it apart and make your own. That's funny. That's very funny

你不喜欢，所以现在你得把它拆开，做你自己的。这很好笑。这非常好笑

97:53

I like that. That's very funny

我喜欢这个。这非常好笑

97:56

Into the heat that of the universe. Wow

进入宇宙的热寂之中。哇

97:58

That's funny which is that this is a

这很好笑，也就是说这是一个

98:01

Restatement of like there's only two ways to make software which is you bundle them you unbundle them

就像是在重述，做软件只有两种方式，要么把它们打包在一起，要么把它们拆开

98:05

Right, but like what are you bundling? It is the things that go together. Hopefully right then like maybe over time

对，但你打包的是什么？是那些天然配套的东西。希望如此吧，然后可能随着时间推移

98:11

People's opinions shift as what what goes together? What should not go together? Right, right

人们的看法会变化，什么该配套？什么不该配套？对对

98:16

So so like they you know there is that that the whole thing you know that maybe

所以就像，你知道，整个事情就是，也许

98:21

Monaco should be a more of a point solution

Monaco应该更像一个点解决方案

98:23

Maybe Monica should own everything right and that there's the question of like in some ways you're bundling for

也许Monica应该拥有一切，对吧，然后问题就在于，某种程度上你是在为你所服务的

98:29

The kind of startup segment that you serve in some ways you're unbundling from Salesforce because Salesforce

那类创业公司细分市场做打包，某种程度上你是在从Salesforce拆包，因为Salesforce

98:35

And then that's how you win market sure from Salesforce. You're unbundled from Salesforce

然后这就是你从Salesforce赢得市场的方式，没错，你从Salesforce拆包出来

98:39

That's interesting. Yeah, that's that's an even broader statement. I've been connecting it to broadly all of vertical AI or

这很有意思。对，这是一个更宽泛的说法。我一直把它广泛地联系到所有垂直AI或者

98:44

And now like bundling on bundling. I mean, that's like the whole like Craigslist and Airbnb you know the

现在就像是层层叠加的捆绑。我是说，那就像是整个Craigslist和Airbnb你知道的

98:50

Tender and all those are separate

Tinder那些都是分开的

98:52

Tinder pull you know like a personal is basically. Yeah

Tinder拉你你知道就像个人的基本上就是。对

98:56

Yeah, I thought about it thought about it that way, but yes

对，我想过，从那个角度想过，但是是的

99:00

And you know one of one of those things so we just shift the open router pod today, right and

你知道其中一件事，所以我们今天刚换了Open Router的pod，对吧

99:05

Open router huge success. Mm-hmm lots of people try to build gateways

Open Router巨大成功。嗯哼很多人尝试建网关

99:08

None of them is success with open router open router only builds gateways

没有一个像Open Router那样成功，Open Router只建网关

99:12

So there's an example of like this they disagree with me that the the the router layer should also do ops

所以有个例子，就是他们不同意我的观点，认为router层也应该做ops

99:20

because arises during ops

因为ops期间会产生这些问题

99:22

and observability and what have you and and they and the MCP gateway they don't have that

还有可观测性之类的，然后他们的MCP gateway没有这些

99:27

And they just they only handle model routing

他们只处理模型路由

99:30

And so like I am like love being wrong like that because they're like we are the point solution and we only do this one thing really really well

所以我其实很喜欢被这样打脸，因为他们的态度是“我们就是点解决方案，我们只把这一个事情做到极致”

99:37

And that's all right that's all engineering focus, which is good. It's also the marketing focus, which is nice

这没问题，这就是工程聚焦，很好。这也是营销聚焦，挺好的

99:43

I think the um

我觉得嗯

99:45

This becomes I think that the well you've described of like the one one stop shop for all the things is

这就变成了我觉得你描述的那种，一站式解决所有事情的方式是

99:50

ideologically what makes sense

从意识形态上讲什么才合理

99:52

But I think what ends up happening in practice and you know part is probably also marketing is that

但我觉得实际中最终会发生的情况，你知道部分可能也是营销，就是

99:56

Especially when you look at the enterprise is like this is broadly under

尤其是当你去看企业的时候，这大致归在

99:59

Infra like broadly under platform, but then there's separate teams owning different parts of the stack

Infra下面，大致归在platform下面，但然后有单独的团队拥有stack的不同部分

100:04

And I think part of that consolidation at the at the skill of an enterprise if we're going to hit MCP portal and gateway

我觉得其中一部分整合，在企业层面，如果我们要搞MCP portal和gateway

100:09

Someone has to come together and say like now where there's a single endpoint that you're hitting for both of these things

得有人站出来说，现在有一个单一的endpoint，你同时打这两个东西

100:14

And that fundamentally is like a hard thing to you know get it on board with

而这从根本上来说是个很难让人接受的事情

100:18

So oftentimes like more like political and and even like people problem rather than like a from a software pure software

所以很多时候更像是政治问题，甚至是人的问题，而不是纯粹的软件问题

100:23

That's makes it in a sense. Yeah

从某种意义上说，这说得通。是的

100:25

Yeah, but that's not it right that like software systems sometimes are bought like by people problems more than then pure software

是的，但问题不在这里，对吧，就像软件系统有时候更多是被人的问题买走的，而不是纯粹的软件

100:30

The way my my framing of this is typically you know how you ship your orchard

我通常的表述方式是，你知道你如何交付你的组织

100:34

This is I will I forget what the name of this principle is

这是我会，我忘了这个原则叫什么名字了

100:37

You need Conway's law. You ship your orchard. Yeah, you buy your orchard

你需要康威定律。你交付你的组织。是的，你买你的组织

100:41

Like like who like when you start a startup or you make a tool

就像，当你创办一家初创公司或者做一个工具时

100:45

Who in the company is your are you best friends for and if it's for everyone? It's for no one

公司里谁是你最好的朋友，你是为谁做的，如果它是为所有人做的，那它就是为没有人做的

100:51

That's interesting

这很有意思

100:53

That is interesting so like a conscious part of

这很有意思，所以就像一个有意识的部分

100:56

Layton space and AI is like okay, there's this category of software engineer that really really really

Layton space 和 AI 就像是，好吧，有这么一类软件工程师，真的真的真的

101:03

Cares a lot about AI. It's a good to make it their day job. That's only make it for them right

非常关心AI。把它当成全职工作挺好的。这样才适合他们，对吧

101:08

Right right right not a not a bundling of engineering never going to be the dev everyone

对对对，不是不是工程的捆绑，永远不会成为那个开发 everyone

101:13

Yeah, I look we talked about databases talk about sure security, but like

是的，我看我们聊了数据库，聊了安全什么的，但就像

101:17

There's a there's an AI angle to it

这有一个AI的角度

101:20

We'll see

我们会看到的

101:22

Well, one thing I'll just like shout out as well related to the point is I think

嗯，有一点我也想顺带提一下，跟这个观点相关的是，我觉得

101:28

Roll of the future that is increasingly important

未来越来越重要的角色

101:31

Platform engineer. I think this is like a huge one. You know, okay more lore

Platform engineer。我觉得这个角色非常重要。你知道，好吧，更多背景故事

101:36

Temporo creep my job before all this I would have yeah

时间蔓延，在这之前我的工作，我本来会，是的

101:39

So we wanted to own platform engineer if I hadn't started. Yeah, yeah, I would have done platform engineer

所以如果我们没开始的话，我们想拥有platform engineer。是的，是的，我本来会做platform engineer

101:43

Because that is the role that decide that decides all of this mission critical stuff

因为那是决定所有关键任务东西的角色

101:48

And then the kids over in applications can do whatever they want on top of our platform

然后应用那边的孩子们可以在我们的平台之上做任何他们想做的事

101:51

But you know the big boys do platform

但你知道，大玩家们做平台

101:55

Yeah, and I think in the modern day they own they own cloud. They own dev x they own security

是的，我觉得在现代，他们拥有云。他们拥有dev x，他们拥有安全

102:00

So I think it's it's all the things like everything where you can actually like lose the company or like it would make a break like

所以我觉得就是所有那些事情，就是那些你真的可能把公司搞没了或者会出大问题的事情

102:07

Platform guys and then product guys can do you know adjust the CSS

平台的人然后产品的人可以你知道调调CSS

102:14

That's saying and that's that you're starting to fight here

那就是说，那就是你开始在这儿争了

102:16

I mean, I'm not the CSS adjuster, but but I

我是说，我不是调CSS的人，但是我

102:20

I appreciate everyone

我很感谢大家

102:22

I mean, it's a fun time. What's what's the future? I feel like all the guys can adjust all the things

我是说，挺有意思的。未来是什么？我感觉所有人都能调所有东西了

102:27

Yeah, yeah, so there's there's a few aspects here

对对对，所以这里有几个方面

102:29

Okay, so so there's a little bit of spillover from from three weeks. So yeah coding agents big teams some of the two concepts

好吧，所以从三周前有一点点延续。所以对，coding agents、大团队，其中两个概念

102:33

But then you know, we'll touch on the software factory

但然后你知道，我们会谈到软件工厂

102:36

I want people to really start thinking about this this world of this machine that build the machine and where if at all

我希望人们真正开始思考这个这个机器建造机器的世界，以及人在哪里

102:42

Are people still in the loop there?

人还在回路中吗？

102:44

You know today, I think the answer is some some combination of

你知道今天，我认为答案是某种组合

102:48

Intent and kind of review or maybe the places like sort of intent where you're defining the specs and what needs to be built and and review where you're

意图和某种审查，或者也许像意图这样的地方，你在定义规格和需要建造什么，以及审查你在

102:55

You know, so doing some amount of your own checking of the system and of course this also

你知道，所以做一些你自己的系统检查，当然这也

102:59

This is I think that bottleneck is going to be loosened up before even the intent piece does

我认为这个瓶颈会比意图那部分更早被放宽

103:03

But for now those are the two choke points. I think in a fully automated system

但就目前而言，这些是两个瓶颈。我认为在一个完全自动化的系统中

103:07

Um, so yeah, I just want to give people supposed to that for one of the OSS partners actually a warp

嗯，所以是的，我只是想给大家介绍一下，其实其中一个 OSS 合作伙伴要做的一个 warp

103:12

They're doing this thing as part of their

他们做这件事是作为他们对

103:14

Contribution to the OSS program where they are going to essentially make a version of warp that is running the factory

OSS 项目的贡献的一部分，基本上他们要做的是打造一个运行 factory 的 warp 版本

103:19

And students will just be like building the factory contributing to the factory

学生们就会像是去构建 factory、为 factory 做贡献

103:23

And keep it running like work in our factory work in our factory exactly

然后让它持续运转，就像在我们的 factory 里工作、在我们的 factory 里工作，没错

103:26

And so that's going to be one of the ways and then they're going to yeah find things that they find better interesting there and pull it back in

所以这会是其中一种方式，然后他们会，是的，找到他们觉得更好、更有趣的东西，再把它拉回来

103:30

This is one of those things where like I'm not sure like obviously factory itself is very very keen on this

这就是那种事情之一，我不太确定，很明显 factory 本身对此非常非常热衷

103:35

Yeah, I feel like everyone wants to build their own

是的，我觉得每个人都想打造自己的

103:39

Yeah, I think that's that is it is a good question

对，我觉得这确实是个好问题

103:44

And I think they're you know, of course entitled to kind of run

我认为他们当然有权去经营

103:47

Try and sell the products. They do I think it's a hard system to own its entirety

试着把产品卖出去。他们确实在做，我觉得这是一个很难完整拥有的系统

103:50

It's a lot. It's a lot. It's a lot. Decide everything for me. Please. Yeah, it is a lot

这太多了。这太多了。这太多了。帮我决定一切吧。拜托。是啊，确实很多。

103:54

And I'm trying to you know, I've been like gone to deep and understanding like the the how they approach their GTM here

我一直在试着，你知道，我最近深入去理解他们在这里是怎么做GTM的

104:00

But I suspect you would have to do a piecemeal right you'd have to be like hey, we first you're coding agent

但我怀疑你得一块一块地做，对吧，你得说嘿，我们首先你这个coding agent

104:04

Oh, then also like you basically land and expand right I mean we talked about this bundling on bubbling piece

哦，然后还有就是你们基本上是先落地再扩展对吧，我是说我们之前聊过这个捆绑和叠加的部分

104:09

You don't start by saying here's your whole factory

你不会一开始就说这是你的整个工厂

104:11

There's no way you could get like a forward on board of that

你不可能直接在他们那儿放一个前向的东西

104:14

You know, I mean where you're gonna go on and just plop the whole thing and they're on their desk

你知道，我的意思是你要去哪儿然后把整个东西往那儿一放，就摆在他们桌上

104:18

That's at least how I would think about it is you first be their their code agent of choice

至少我是这么想的，就是你首先成为他们首选的代码 agent

104:22

You build up some background agent for them you you own some of their SRE pieces and then you're like well

你给他们搭建一些后台 agent，你接管他们的一些 SRE 部分，然后你就说好吧

104:26

Now we can just weave it all together if you define everything in code or something like that

现在我们可以把这一切编织在一起，如果你用代码或者类似的东西定义一切的话

104:30

Oh, you know coffees

哦，你知道的，coffees

104:32

There's a fun blog post on this from opening I we had buddy Ryan we had Ryan on the podcast

OpenAI 有一篇关于这个的有趣博客文章，我们请了 buddy Ryan，我们请 Ryan 上过播客

104:38

Basically, they spent five months

基本上，他们花了五个月

104:40

Making an internal tool with zero lines of manually written code

用零行手写代码做一个内部工具

104:44

They kind of walk through one. It's called harness engineering so you can learn harness engineering

他们大概带人走了一遍。这叫做harness engineering，所以你可以学习harness engineering

104:50

But you know the learnings of

但你知道那些经验

104:52

There was kind of a slump or slower at first and then you offload and you had peer reviews. It's at this point it's from February but

一开始有点低迷或者比较慢，然后你卸载出去，还有同行评审。这时候是二月份的了，但

104:59

It's you know a vision off of a software factory, right like you are

你知道这是基于软件工厂的愿景，对吧，就像你

105:04

full sending

全力投入

105:07

Letting agents fully write everything and how to do that well. Yeah, absolutely. Yeah, I mean some version of this is gonna exist

让agents完全写所有东西，以及如何做好这件事。是的，完全没错。是的，我是说某种版本的这个东西会存在

105:13

Twix's point. I do wonder about the

Twix的观点。我确实在想那个

105:17

The sales piece here like the actual business of this doesn't mention a question

这里销售的部分，实际的业务这块没有提到一个问题

105:21

But my principle is receptive. I think this is the this has been version of the future

但我的原则是乐于接受。我觉得这就是未来的版本

105:25

And then beyond that I think there's gonna be some amount of dabbling into and speaking about like okay

然后除此之外，我认为会有一定程度的涉猎和讨论，比如说好吧

105:29

What does it mean to yeah, what does it mean to be a developer in the future, right like are we

作为一个开发者，在未来意味着什么，对吧，就像我们

105:34

People talk about taste people talk about like you know

人们谈论品味，人们谈论你知道的

105:37

Understanding of systems and architecture and yeah like software taste

对系统和架构的理解，对，就像软件品味

105:41

I want to like go a little bit deeper here on maybe starting to like at least put some answers to the question around this

我想在这里再深入一点，也许至少开始给这个问题一些答案

105:47

Yeah, so this is your answer to where a software engineering goes next

对，所以这就是你对软件工程下一步走向的回答

105:51

Yeah, so I mean the factory is like the tactical, but I think beyond that there's like a there's a question around like the role

对，所以我的意思是工厂是战术层面的，但我觉得除此之外还有一个关于角色的问题

105:55

Right like what is the role of all of into if if the factory is is running and you know full steam ahead doing its thing

对，就像如果工厂在运转，你知道全速前进做它该做的事，那所有这些东西的角色是什么

106:01

They can verify about this is churning out code

他们能验证的是这个在大量产出代码

106:03

What what do we do in all this right? And I think we I think you and I we had this discussion with Jerry at the dinner one spawn a time

我们在这一切里面做什么对吧？我觉得我们，我觉得你和我之前跟Jerry在晚饭时讨论过这个

106:09

A few weeks back with Dex and Jerry yeah around around the factory

几周前跟Dex和Jerry一起，对，围绕工厂

106:13

But yeah, I think that there was this part of at least the discussion points there which I am pretty much on board with is like

但是的，我觉得至少讨论点里面有一部分是我挺认同的，就是

106:20

Increasing levels of abstraction and then

不断提高抽象层级，然后

106:22

Somehow there's someone still even a CEO of a company has to have enough oversight understanding of like different constituent parts

不知怎么的，仍然有人，甚至是公司的CEO，必须对不同组成部分有足够的监督和理解

106:29

To be able to make a judgment on direction right to be able to course correct to be able to set metrics and things of that nature

能够判断方向对不对，能够纠偏，能够设定指标之类的事情

106:34

I think there's a version of the translates of that there's that there's written that of that analogy the translates like someone who oversees the technical

我觉得有一种翻译版本，就是有人写过那个类比，翻译过来就像是负责技术监督的人

106:40

We call it a manager of agents. We call it a you know different terms. They're all buzz buzz wordy

我们叫它agent管理者。我们叫它，你知道，不同的说法。都是些流行词

106:44

But in some some respect someone who understands proud requirements constraints technical constraints kind of loops through all of those and

但在某种程度上，就是那种理解产品需求、约束条件、技术约束，能在这些之间来回穿梭的人，而且

106:51

And knows how to adjudicate. Yeah

而且知道怎么去做裁决。对

106:54

I will say that in the chair before you we just had

我想说，在你们之前坐在这把椅子上的，我们刚请了

106:58

Anthropics head of products and head of engineering for the platform to their platform

Anthropic的产品负责人和平台工程负责人，负责他们平台的

107:03

And I was like when are when are your jobs going to combine? They're like not yet

我当时就问，你们的岗位什么时候会合并？他们说还没到时候

107:07

You know

你知道

107:08

And so if Anthropics not doing not there yet. You should not feel like it's it's imminent

所以如果Anthropic还没做到那一步，你不该觉得这事马上就要来了

107:16

Did they say how long or they just not yet

他们说了要多久吗，还是只是说还没

107:18

Not yet, but who knows. I definitely there was definitely a lot of like okay. Yeah, we're I'm doing her job

还没，但谁知道呢。确实有很多那种，好吧。对，我在做她的活

107:24

She's doing my job like yeah, I think it's also an interesting perspective for students right you said

她在做我的活，对，我觉得这对学生来说也是个挺有意思的视角，你说

107:29

There's a lot of sophomore undergrads like you got two three years to go to plan out what you want to be doing

有很多大二本科生，你还有两三年时间来规划你想做什么

107:35

There's different ways to look at it from that right you're not straight in industry

从那个角度有不同的看法，对吧，你不是直接进业界

107:39

The thing that I care

我关心的事情

107:41

I'm thinking a lot about is was so one part of this discussion that resonates more with me is the loop

我最近在想很多的一件事，就是这次讨论中让我更有共鸣的一部分是那个循环

107:47

Stuff where like okay, you you started writing prompts now you no longer I promise you're in prompts

就是那种，好吧，你一开始是在写prompts，现在你不再——我保证你不再是在写prompts了

107:53

That's right right and then they just keep going up and I think to me the sort of recursive

对的对的，然后它们就一直在往上走，我觉得对我来说那种递归的

107:58

concentric circles nature going outward, but also going inward. Yeah is something that

同心圆向外扩展、同时也向内深入的本质，是的，是某种

108:04

builds the factory. I think you build up to the factory not build down

构建工厂的东西。我觉得你是往上构建到工厂，而不是往下拆

108:08

That's what I think a lot of people will

这就是我觉得很多人最终会

108:11

Eventually start building down like you start from the top level like what is my goal and then you figure out the systems

开始往下拆的方式，就像你从顶层开始，比如我的目标是什么，然后你去搞清楚那些系统

108:16

That currently creates a ton of slop and you should not do that, but someday you will

目前那会产生大量的垃圾输出，你不应该那样做，但总有一天你会那样做

108:23

Yeah, yeah, so yeah, so instead of so I mean, so you know that will be my question to prompt you

对对对，所以所以，我的意思是，所以你知道这就是我要问你的问题

108:30

is

就是

108:32

Do we do we what is the sort of prime question that people

我们，我们，人们存在的那个核心问题是什么

108:36

Exist to solve like that is your that is your thing right so in that's let's say an engineer today

就像那就是你的，那就是你的东西对吧，所以在那，比如说今天一个工程师

108:43

You are measured by your PR merge rate right PR contribution rate

你是按你的PR merge rate来衡量的对吧，PR贡献率

108:48

Well two years or not that's gonna be ridiculous

两年后或者不是，那也太荒谬了

108:51

Okay, well, what's your main thing right yeah money

好吧，那你的主要东西是什么对吧，钱

108:55

I think people always go to money too quickly. It's something in between money in right PRs has facet

我觉得人们总是太快就想到钱。它是介于钱和对吧PR之间的某个东西，有facet

109:00

C sat scores from your features

从你的特征中得出C sat分数

109:02

Even that's like a product decision not really a code decision right like I don't know like I but but if the but if the engineering is evolving until like a more of a

就连那也像是产品决策而不是真正的代码决策，对吧，就像我不知道，但是但是如果说工程正在演变，直到更像是一个

109:10

PM engineer like cry engineers, you know hybrid then they then I think that's you know

PM工程师，像是cry工程师，你知道的混合型，那么我觉得那就是你知道的

109:14

That's one answer the question of like what is the role evolved into I think it goes with that say that

那是关于角色演变成为什么这个问题的一个答案，我觉得它跟那个说法是一致的

109:19

I would today recommend to students spend more time thinking about holistic systems

我今天会建议学生花更多时间思考整体系统

109:23

Not just your like single part of the the engineering puzzle, but how that ties in with and you know

不只是你工程拼图中的单个部分，而是它如何与之关联，你知道的

109:29

And this is like like in practice ends up happening like students for their capstone projects I encourage them to go out and

而这就像在实践中最终会发生的那样，对于学生的毕业设计项目，我鼓励他们走出去

109:33

Like in fact what I give them as the seed of ideas for final projects is like the the YC request for startups

就像实际上我给他们作为最终项目想法种子的是YC的创业请求

109:39

Someone was like go build a company

有人说，去创办一家公司吧

109:41

Like you know, I mean like go go build something interesting. What do people build faster?

就像你知道的，我是说去做点有意思的东西。人们做什么东西最快？

109:45

I mean that's a variety of things like someone was like

我是说那有各种各样的东西，就像有人说的

109:47

Someone came in the e department. I was in the e-school

有人来到了电子系。我当时在电子工程学院

109:50

And e was studying e and they were like they built effectively a translator for like parsing of like manuals

那个人在学电子，然后他们实际上做了一个翻译器，用来解析那些手册

109:57

to that have like

那些手册里面有

109:58

Like memory dresses in in C and then converting that into like

比如C语言里的内存地址之类的，然后把它转换成

110:02

Like code that they could then like interpret and even like visualize so so something specific to their their use case

他们可以解释的代码，甚至可以可视化，所以是针对他们自己用例的特定东西

110:07

Someone was like a designer and is now actually a versel and she like had you know

有个人本来是设计师，现在实际上在versel工作了，她就像你知道的

110:11

Like take a picture of my like design notebook and generate me an app based on what I take a picture of right

就像拍一张我的设计笔记本的照片，然后根据我拍的照片给我生成一个app，对吧

110:15

So literally she's like for her demo. I was like here. You go snaps a picture of like some random like scribbles

所以 literally 她就像为了她的demo。我就想给你。你随手拍了一张随便什么涂鸦的照片

110:19

That they're she was doing and she got a fully fleshed out of versel app and from that

那是她正在做的东西，然后她就得到了一个完整成型的versel app，从那个开始

110:22

And I think because of that probably got a job at versel doing that doing some versions. Oh nice

我觉得正因为如此，她可能因此拿到了versel的工作，做那个做某些版本。哦不错

110:27

That's a high bar actually versel very known for their design engineers

那其实是个很高的门槛，versel非常以他们的设计工程师闻名

110:31

Yeah, yeah, and I think she's now working on Eve

是的，是的，而且我觉得她现在在做Eve

110:34

I remember ah

我记得啊

110:35

No Eve in your

你那边没有Eve吗

110:37

Eve Eve is an OSS or like OSS partner

Eve Eve是一个OSS或者类似OSS合作伙伴

110:40

I mean like like they one of the the ways versel is partnering with us is that they are having students build with Eve

我的意思是，就像Versel和我们合作的方式之一，就是让他们让学生用Eve来构建

110:45

Yeah, so this is the concept of meta harnesses right right harnesses of harnesses

对，所以这就是meta harnesses的概念，对吧，harnesses的harness

110:51

You can always go one that will be brutal

你总是可以再往上走一层，那会很残酷

110:53

But I do think like it makes sense at least for me in the same way that model orders wrap model APIs

但我确实觉得至少对我来说这是合理的，就像model orders包裹model APIs一样

110:58

You you need to wrap harnesses as well if you're gonna that's a nice one between them. I agree

如果你要那样做，你也需要包裹harnesses，这是个不错的类比。我同意

111:02

I think at some level students should also stay grounded in like there's a lot of stuff happening at the edge

我觉得在某种程度上，学生也应该保持脚踏实地，因为边缘侧有很多事情在发生

111:08

There's also a lot of companies that are still like

还有很多公司到现在还是

111:11

single-call chat GPT in production playing around haven't tried it

在生产环境里单次调用 ChatGPT，只是在玩玩，还没真正试过

111:18

You know there there is a lot of that so there's there's still time for what to work

你知道，这种情况很多，所以现在做“什么才有效”这件事还有时间

111:21

But I think your point on taste is good have taste try stuff like

但我觉得你说的关于 taste 这点很好，要有品味，去尝试东西，比如

111:27

whether

不管是

111:28

Each company has their own version of a software factory

每家公司都有自己的软件工厂版本

111:32

You're building tools to help companies do it like staying on top of it actually

你是在打造工具帮公司做到这一点，比如真正跟上它

111:36

Using the tools playing with it like you'll you'll figure it out along the way, you know

用这些工具，玩起来，你会边走边摸索出来的，你知道

111:40

Yeah, and I think the the hundreds of years all that I mean first to your point that are a lot of companies still in the prompt chat GPT

对，而且我觉得那几百年……我是说，首先回应你的观点，还有很多公司仍然停留在prompt ChatGPT的阶段

111:48

Or some version of you know chat GPT maybe do they just got a cursor license and are just now getting people onboarded

或者某个版本的你知道的chat GPT，也许他们只是刚拿到了cursor的license，现在才刚开始让人onboard

111:53

That is way more true than we believe certainly here in the valley

这比我们想象的要真实得多，尤其是在硅谷这里。

111:57

That is I can I can say 100 percent confidence

那就是我可以，我可以百分之百肯定地说

112:00

um

嗯

112:01

But yes, I think for the students being supposed to this is just at a certain point

但是是的，我觉得对于学生们来说，这只是在某个节点上

112:05

This the fact that they know this and are on the edge is what will eventually allow them to also be some of the more interesting and more valuable contributors when they

他们知道这一点并且处于前沿，这正是最终让他们也能成为更有趣、更有价值的贡献者之一的原因，当他们

112:11

Go to maybe places where that it maybe aren't up to snuff in terms of being on the frontier

去那些可能在前沿性方面还不太达标的地方

112:16

Yeah

是啊

112:17

What one of the this this kind of progression type talk

那个，那个，那种渐进式的说法

112:20

Still very influential about a year old TV Aggie talking about the levels of psychosis

大概一年前TV Aggie讲精神病等级那个，现在还很有影响力

112:25

Uh, he's all the way through level 10 now um where he's he's uh, he's given up and he's moved to a farm

呃，他现在已经到第10级了，嗯，他已经放弃了，搬去农场了

112:30

But uh, yeah, like you want to you. I'm sorry. I mean, I'm making fun of him

但是呃，是啊，就像你想的那个。抱歉，我是说，我在开他玩笑

112:34

He built gas town. Yeah, I remember all these things multi-gas city now

他建了gas town。是啊，我记得这些事，现在多gas城市了

112:39

He's burned down the city to say that I'm building my game. I'm no more a gas town

他把城市烧了，说我要做我的游戏。我不再是gas town了

112:43

I thought that I mean, I remember talking to him at the engineer fair early this year and uh

我当时想，我是说，我记得今年早些时候在engineer fair跟他聊过，呃

112:48

Wasn't he his whole like thing like the only useful thing that was build a gas town was gas down there

他整个那套东西不就是嘛，唯一有用的东西就是建了个gas town，结果gas town就在那儿

112:53

Right would that that's a horrible failure horrible totally totally um and and so like people should be aware that like

对吧，那那那是个可怕的失败，可怕的，完全完全，嗯，所以大家应该意识到

113:00

psychosis is real and

精神病是真实存在的

113:02

If you're not holding yourself accountable to delivering real products, you're just screwing along with all lens

如果你不要求自己对交付真实产品负责，那你就是在跟所有lens瞎混

113:07

Yeah, this is like a maybe I mean relevant to it

对，这这有点像，我是说跟它相关

113:11

But I wonder how much of I mean, I played a little bit of gas town very involved system

但我想知道有多少，我是说，我玩了一点gas town，非常复杂的系统

113:15

Maybe a little ahead of its time um, but I'd also wonder if they were like bottled like a little bit by the by the marketing

也许有点超前于它的时代，嗯，但我也想知道他们是不是被营销给限制住了

113:20

In that it was a very difficult thing to grok from a terminology perspective

因为从术语的角度来说，那是个很难理解的东西

113:24

If there's a lot going on there. I think the gas town marketing also is one of the things that propelled it right like

如果那里有很多事情在发生。我觉得gas town的营销也是推动它的事情之一，对吧

113:31

What was right before gas town?

gas town之前是什么？

113:33

There were like

好像有

113:34

The like the like his his yeah, he did the next year talking with someone else's no

那个那个那个他的他的对，他第二年跟别人聊的不是

113:38

No, no, there were like there was a hole up in claw. There was something else and there was gas town

不不不，好像有，有一个hole up in claw。还有别的东西，然后还有gas town

113:42

And the whole manifesto also really helps spurred into the conceptual the people that would never use gas town

还有那个宣言也确实帮助推动了概念层面，那些永远不会用gas town的人

113:50

But learned about gas town because you can read the manifesto like there. There's a marketing angle to it

但因为可以读那个宣言而了解了gas town，就像那里有一个营销的角度

113:55

Yeah, I mean, so so I do so Steve is not necessarily someone to take literally but direction like you know

对对，我是说，所以我确实，Steve不一定是那种要照字面理解的人，但方向，你知道

114:01

Directionally correct, you know, and so I think he's directionally correct on this one

方向上是对的，你知道，所以我觉得他在这一点上方向是对的

114:06

And so if you're at the ID stage, it's time to move into a power user stage

所以如果你还在ID阶段，是时候进入power user阶段了

114:10

If you're a single agent you probably time to move to multi agent and blah blah

如果你还是single agent，可能该转向multi agent了，等等等等

114:13

And you probably should try to build your own orchestrator, right and that's all that

而且你可能应该试着搭建自己的orchestrator，对吧，大概就是这样

114:17

They're arguably this is not this is that kind of out of date now there is more to the software factory beyond this

可以说这现在已经有点过时了，software factory远不止这些

114:23

But this this article predates the software factory. I do think you know one of those things where like

但这篇文章早于software factory。我确实觉得，你知道，就像那种事情

114:28

I'm open to as an employer

作为雇主我是开放的

114:31

I have this problem in my company today, which is I do not want to hire anyone who hasn't been through

我今天公司里就有这个问题，就是我不想雇任何没有经历过

114:37

Psychosis and come out the other end because I am currently paying for somebody to go through psychosis

精神病，然后从另一头走出来，因为我目前正在花钱让某个人经历精神病

114:42

And it's very expensive from the his salary

而且从他的工资来看这非常昂贵

114:45

Very expensive in tokens and I'm getting nothing out of it and he doesn't get like my feedback doesn't get through

token上非常昂贵，而我什么也没得到，他也收不到，就像我的反馈传不进去

114:50

And so like

所以就像

114:52

It's almost like I would encourage your students to have psychosis and then come out and go like okay, what was wrong with me

这几乎就像我会鼓励你的学生去经历精神病，然后走出来说，好吧，我到底哪里出了问题

114:57

Right because like that's what Steve just went through

对吧，因为就像Steve刚刚经历的那样

115:00

Like a very public to your journey of this. Yeah, it's weird on the student side. There's like a lot of

就像一段非常公开的你的这段旅程。是的，在学生那一侧很奇怪。有很多

115:05

Industry you think you're being productive. You're not is probably the student version right your

行业，你以为自己在产出，其实没有，这大概就是学生版本的吧，你的

115:10

notion charting and making all these

notion图表和做所有这些

115:13

Optimization personal ass and it's like thing I went through the psychosis, but nothing has changed right right go through it go through right yeah

优化个人屁股，然后就像我经历的那种精神病，但什么都没变，对吧对吧，经历它，经历，对

115:21

I haven't seen this. I've seen a version of this from Boris actually obviously more geared toward cloud use cases

我没见过这个。我 actually 从 Boris 那里见过一个版本，显然更偏向云端用例

115:26

But I remember he made like a quad artifact which showed I think more of the team level

但我记得他做了一个类似四象限的图，我觉得更多是团队层面的

115:29

But I think he had four stages not eight stages

但我想他有四个阶段，不是八个阶段

115:32

But it kind of roughly followed the same progression of like no adoption to through to

但它大致遵循了相同的进展，从没有采用到

115:37

Yeah, only for as good. This is how this is too many levels

对，只是为了好。这就是为什么这层级太多了

115:40

Yeah, well, so we'll see if

对，好吧，所以我们看看会不会

115:42

If you can pull it up, but other than that, you know, zoom me out every you know, the scar discussed all the things

如果你能把它调出来，但除此之外，你知道，帮我缩小一下，你知道，那个scar讨论了所有的东西

115:47

That's one. Oh, that's a lot. Okay. There we go. It is okay

这是一个。哦，那挺多的。好的。来了。没问题

115:52

So it's gated assisted parallel was the last one

所以最后一个是有门控辅助并行

115:54

Supervised autonomy and AI native

监督自主和AI原生

115:57

Okay, so zero indexed so it's five levels, right? All right. I think so it's zero

好的，所以从零开始索引，一共五个层级，对吧？好的。我觉得是这样，从零开始

116:01

Yeah, wow actually it's pretty good. I need to

是的，哇，实际上还挺不错的。我需要

116:05

Retour this and more more debt

重新审视这个，还有更多的债务

116:07

And use the anator. Yeah, yeah, I would give a shout out to dynamic workflows

然后用那个anator。对对，我要给dynamic workflows点个赞

116:12

This is one of those things that is a sub agent pattern

这是那种sub agent模式的事情之一

116:15

That the opening guys are also exploring

这也是opening那帮人正在探索的

116:18

That is clearly fitting

这显然很契合

116:20

And so it's basically you're writing a harness on demand for the thing that you're trying to do and so

所以基本上就是你为你想要做的事情按需写一个harness，然后

116:28

This is maybe like a lot of sub agent

这大概就像很多sub agent

116:31

Conversation is just you're copying the the parent agent

对话就是你只是在复制parent agent

116:34

But you actually should be writing way more code and then like running that as a as a sub agent and that is what

但你其实应该写更多的代码，然后把它作为一个sub agent来运行，这就是

116:40

The anthropic people have chosen to call dynamic workflows

Anthropic的人选择称之为dynamic workflows的东西

116:43

Okay, but you know, yeah, we built a version of this as well. Yeah, it's a good pattern for Monaco

好，但你知道，是的，我们也做了一个这个的版本。是的，这对Monaco来说是个好模式

116:49

Yeah, what does it do? Yeah, I mean on the fly sort of like harness sort of use cases for GK. Yeah

对，它是做什么的？对，我是说那种即时性的，类似GK的用例场景。对。

116:54

That's great

太好了

116:56

I will note that when Boris wrote this it's for team adoption for people that are listening and also he says you know

我要指出，Boris写这个是为了团队采用，也是给在听的人看的，他还说你知道

117:04

Andthropic is on stage three. We're still trying to get to stage four. Wow. He's like I am at stage for I'm the guy

Anthropic 已经到第三阶段了。我们还在努力迈向第四阶段。哇。他就像在说我已经到第四阶段了我就是那个人

117:10

But you know, it's it's not as simple as Steve's like Steve is a very good personal do this do this do this

但是你知道，这并不像Steve说的那么简单，Steve是个很擅长“做这个、做这个、做这个”的人

117:17

This is a much more if you're running a team and you want to AI pill your

这更像是，如果你在带团队，你想用AI赋能你的

117:23

Team and not just be a user of one, you know, andthropic is still at stage three. Don't

团队，而不只是做一个使用者，你知道的，Anthropic仍然处于第三阶段。别

117:28

Don't be like, yeah, I mean you should teach this

别这样，我是说你应该教这个

117:31

This is this is great. Yeah, it's got all the stuff that is everything that we talked about. There you go

这这很棒。对，我们聊过的所有东西这里面都有。这就对了

117:35

I can't believe I haven't come across this. Yeah, um, take a research

不敢相信我居然没碰到过这个。对，嗯，去研究一下

117:39

Very, very good. Okay

非常非常好。好的

117:41

That was a wild tour one semester of Stanford and one hour

那真是一趟疯狂的旅程，斯坦福一个学期浓缩成一小时

117:46

Thank you for doing this any party thoughts any calls to action. No, this is a thanks for I mean ton of fun

谢谢你做这些，有什么想法或者呼吁大家做的事吗。没有，这就是个感谢，我是说特别好玩

117:52

It's it's been a you know, nine months in the making. I'm excited. This is end of week one

这这是你知道的，酝酿了九个月。我很兴奋。这是第一周结束

117:56

A lot to do a lot to learn. I'm excited to see what happens with the

有很多事要做，有很多要学。我很期待看到接下来会发生什么

118:00

Especially the OSS partners and just getting students like deep in the code of some complex code bases

特别是 OSS 的合作伙伴，还有让学生深入一些复杂代码库的代码里

118:04

Having them build some stuff. So no, it'll be fun. You guys supposed to you're doing another version next year

让他们构建一些东西。所以不，这会很有趣。你们应该……你们明年还要做另一个版本

118:10

You need credits for students if OSS people want to get involved they can reach out. Sure

你需要给学生提供 credits，如果 OSS 的人想参与，他们可以联系。当然

118:16

Very yeah, this credits matter. Did he want it? What did he actually want the students? Yeah

非常对，是的，这些 credits 很重要。他想要吗？他到底想让学生做什么？是的

118:21

You do have credit card. I don't think a lot of students want to spend

你确实有信用卡。我觉得很多学生不想花钱

118:24

I won't yeah, yeah, you credit card

我不会，对对，你信用卡

118:26

Yeah, no, I think look I will start like I remember you know, it was like you know

是的，不，我觉得听着，我会开始……我记得你知道，就是那种你知道的

118:30

There's a certain number of credits for sure

肯定有一定数量的 credits

118:31

But then on top of that he's like look if they want to go hog wild with this like let me know and we'll do have these like one-off

但然后在此基础上他又说，你看，如果他们想在这上面放开手脚大干一场，就告诉我，我们就搞这些一次性的

118:36

Experiments where they can just like scale to like a thousand plus agents and so this is you know from factor

他们做的实验能直接扩展到一千多个agents，所以这就是你知道的从factor

118:40

And you never write it. Yeah

而你从来不用写它。是的

118:42

No, those kind of people are very valuable. So this is Wilson from cursor right who is like I'm gonna write a browser

不，那种人非常有价值。所以这就是Cursor的Wilson，他说我要写一个浏览器

118:48

Right, yeah, those people who are ambitious and like can actually execute those are very valuable and that's basically a

对，是的，那些有野心而且确实能执行的人非常有价值，那基本上就是一个

118:54

Recruiting call that's right. Yeah, that's a good point. Yeah, I do think like students should should like understand how to be ambitious

招聘电话，对。是的，这是个好观点。是的，我确实觉得学生应该理解如何有野心

119:02

Because snapping a photo. I mean respectfully. He's snapping a photo and turning into app is not that ambitious in the this day and age

因为拍张照片。我意思是说句公道话。他拍张照片然后变成app，在这个时代也没那么有野心

119:09

Writing in bronzer is and they're like they're actually there's a lot more projects like that. Yeah, yeah, I think that's right

用bronzer写作，他们就像，其实还有更多类似的项目。对对，我觉得是这样

119:13

Yeah, it's all about raising the floor. I may think yeah last year was it would you know

对，关键就是提高下限。我可能会想，去年就是那样，你知道吗

119:18

You know credit screw to the student that of the last year, but uh, yeah, this year right

你知道去年那个学生credit screw的事，但是，呃，对，今年这个

119:22

We're gonna yeah, but that's that's a whole point right like we're leveling up right last year. It was like

我们会这样的，对，但这正是关键所在，对吧，就像我们在升级一样，对吧，去年。那时候就像是

119:29

Yeah, hey, have you heard of prompting like here's what it is

哎，对了，你听说过prompting吗，就是这样的东西

119:31

I spent some time talking about like you know k-shot prompting as one of the lessons

我花了一些时间讲了讲，你知道的，k-shot prompting作为其中一个经验教训

119:35

And like I don't even think you guys even think about that now. So

而且我觉得你们现在甚至都不会去想那个。所以

119:38

I was there's a there's a version of

我那时候有一个版本是

119:40

Thinking you stick out in your cohort of classes and then there's also a version of

觉得自己在同届同学中很突出，然后还有另一个版本是

119:45

You're also in early engineering roles of a bunch of people in industry people still trying to get jobs and

你也在早期工程岗位，行业里很多人还在努力找工作

119:51

It's not it's not uh, you know, not to call out one demo, but you can you can do a lot to stick out

不是说，呃，你知道，不是要点名某个demo，但你可以做很多事来脱颖而出

119:56

I'll give us students a freebie of a week 11 uh making benchmarks makes benchmarks in e-vals

我给学生们一个免费福利，第11周，做benchmarks，做evals里的benchmarks

120:03

This will get you a good one

这会给你带来一个好的

120:05

I'm gonna get you in the door of every frontier lab and every front every top age of lab

我能让你进入每个前沿实验室的门，每个顶级AGI实验室

120:09

Good good engineering blog post with some rigor on

好的工程博客文章，带一些严谨性

120:13

Anything interesting. There's a lot of benchmarks are laying fruit there, but the

任何有趣的东西。有很多benchmarks在那里等着被摘取，但

120:17

Engineering blog posts do really well right you you went deep on one thing yeah

工程博客文章效果真的很好，对吧，你深入钻研了一件事，是的

120:21

um and probably also there's like big a lot of money to be made in that as well

嗯，而且这里面可能也有很大的赚钱空间

120:25

So not just uh in the doors, but if you make that into businesses you you can that you can really really cash in on that

所以不只是呃入门级的，如果你把它做成生意，你你就能真的真的从中大赚一笔

120:32

Yeah, people don't know like see see bench was like undergrad project

是啊，人们不知道，比如see bench其实是个本科项目

120:35

Yes, yes, yes, that's right from from Princeton right

对对对，没错，来自普林斯顿对吧

120:38

Yeah, yeah

对对

120:40

Um great. Okay, well that's it. Thank you so much. Yeah, thank you guys

嗯太好了。好，那就这样。非常感谢。好，谢谢大家

How Stanford Teaches AI-Native Engineering — Mihail Eric, Monaco

Play Queue