本集简介
双语字幕
仅展示文本字幕，不包含中文音频；想边听边看，请使用 Bayt 播客 App。
我能跟你较个真吗？人类真的是瓶颈吗？
Can I give you a hard time that humans really are the bottleneck?
因为你看从开始到提交PR的时间，只要三十五分钟。
Because if you look at kickoff to PR time, it's thirty five minutes.
但你看从提交PR到第一次人工审查，却要三个半小时。
But if you look at PR to first human review, it's three and a half hours.
而且当你在做……
And when you're doing...
我觉得上个月大概有超过两千个PR。
I think it was, like, over 2,000 PRs in the last month.
那你怎么防止团队里的东西失控，就像我说的，变得乱七八糟？
Like, how do you keep things in the team from going, as I say, like, reigns?
什么是软件工厂？
What is a software factory?
至少对我们来说，它是一个实实在在的名词。
For us, at least, it's an actual noun.
这就像一个产品概念，由一堆仓库、一堆MCP服务器、一堆配置，然后还有一堆代理组成。
It's like a product concept where it consists of of a bunch of repos, a bunch of, like, MCP servers, like, a bunch of configuration, and then a bunch of agents, essentially.
所以，比如代码审查代理设计、不同的代理、不同的自动化，而且这一切都是用代码定义的。
So, like, a code review agent design, different agents, different automations, and it's all defined in code.
评分会在所有运行中进行，但还有一个第二循环，就是现在推特上很火的自改进，你有一个观察代理，它可以对工厂进行更新，试图防止特定的失败模式。
Scoring happens across all runs, but then there's a second loop, popular thing on Twitter right now called self improvement, where you have, like, an observer agent, and it can then create updates to your factory that will try to prevent the particular failure mode.
最有帮助的一点是，你可以让这些工厂代理进行计算机使用验证。
One of the things that's most helpful is, like, you could have these factory agents do computer use verification.
所以在这个例子中，它制作了一个视频。
So in this case, it made a video.
我就是在跟这个做我过去二十年工作的东西对话，它……
I'm just, like, talking to this thing that is doing this job that I've done for the last twenty years, and it's...
现在它做得比我还要好。
Now it's, like, doing it kinda better than me.
剧透警告。
Spoiler alert.
问题出在人身上。
The humans are the problem.
欢迎回到《How I AI》。
Welcome back to How I AI.
我是克莱尔·沃，产品负责人，也是AI狂热爱好者，在这里帮你更好地使用这些新工具。
I'm Claire Vo, product leader and AI obsessive here on a mission to help you build better with these new tools.
今天，我们有Warp的CEO扎克·劳埃德，他会向我们展示他所说的“软件工厂”到底是什么。
Today, I have Zach Lloyd, CEO of Warp, and he's gonna show us exactly what he means by the software factory.
他会展示如何在Slack里启动任务，超越传统软件开发流程意味着什么，以及一位技术型CEO如何用非技术工具结合AI。
He's gonna show us how you can kick off tasks in Slack, what it means to go beyond the software development life cycle, and how a technical CEO uses nontechnical tools with AI.
我们开始吧。
Let's get to it.
本期节目由DX赞助。
This episode is brought to you by DX.
在最近一项涵盖500多个工程组织的研究中，DX发现，过去一年AI工具的开支增长了28倍。
In a recent study across more than 500 engineering organizations, DX found that spend on AI tools has grown 28 x over the last year.
AI编写的代码比例在上升，但整体创新却停滞不前。
The share of AI authored code is climbing, but overall innovation has remained flat.
随着团队生成代码的速度加快，代码审查和验证中的新摩擦正在抵消早期的速度提升。
As teams generate code faster, new friction in code review and validation is offsetting those early velocity gains.
DX在软件开发生命周期中同时追踪速度、质量和成本，让工程领导者清楚看到AI对交付的影响，以及这些投资是否真正转化为实际价值。
DX tracks speed, quality, and cost together across the software development lifecycle, giving engineering leaders clear visibility into how AI impacts delivery and whether those investments are translating into real value.
在getdx.com/howiai下载完整报告。
Download the full report at getdx.com/howiai.
网址是getdx.com/howiai。
That's getdx.com/howiai.
Zach，感谢你参加HowIAI。
Zach, thanks for joining HowIAI.
谢谢邀请。
Thanks for having me.
很高兴来到这里。
Excited to be here.
现在时间线上有很多趋势，但我觉得2026年会火的一个，或者说看起来会火的，就是“工厂”这个概念。
There are a lot of trends on the timeline right now, but one that I think is going to be big for 2026 or seems to be big for 2026 is the factory.
所谓“工厂”，就是那种很天马行空的想法，觉得我们输入一些token，就能输出企业价值之类的。
The factory, this, like, fantastical idea that we have that we're gonna, you know, input tokens and output enterprise value or something like that.
至少能输出代码吧。
At least we'll output code.
所以，对，没错。
So I Yeah.
我很期待你以CEO、工程师、开发者的身份，给我们展示一下。
You know, I'm excited for you to show us as CEO, as engineer, as builder.
对你来说，“工厂”到底是什么？在Warp里面又是什么样的？
You know, what what to you is the factory and how it looks inside Warp?
“工厂”这个词确实很火。
The factory is definitely trending.
其实我不太喜欢“工厂”这个说法，但大家都这么叫，也就统一了。
I actually don't love the term factory, but it is the sort of thing everyone is standardized on.
对我来说，这有点让人觉得没有人情味。
It feels a little bit sort of like dehumanizing to me.
不过，是的，这个想法正是你所说的那样。
But but yeah, the the idea is exactly what you said.
就是说，在一个有这些神奇代理的世界里，你怎么才能更有条理地利用它们的力量，让它们基本上从你的想法出发，为你构建软件。
It's like, how do you, in a world where we have these sort of magical agents, kinda harness their power in a more organized way to have them, you know, basically build build software for you from your ideas.
所以，是的，我很兴奋能向你展示我是如何使用工厂的，以及它对我来说意味着什么。
And so, yeah, excited excited to to to show you how I how I use factories and, like, what it means to me.
我最近参加了Lenny的峰会，我记得Marty Cagan也说过他讨厌“工厂”这个词。
I was at Lenny's Summit recently, and I think Marty Cagan also said he hated the word factory.
我在想，是不是更像圣诞老人的工坊。
I wonder if it's more like Santa's workshop.
就像，我们有神奇的精灵，他们制作出奇妙的东西，带来欢乐，并且包装好送给我们。
Like, we have magical elves that craft wonderful things that spark joy and are delivered gift wrapped for us.
所以我要说，AI魔法工坊。
So I'm gonna say the AI the AI magical magical workshop.
不是。
No.
给我们看看，这到底是怎么回事？对。
Show us, you know, what does what does this look like for Yeah.
你确实说过你输入
You did you did say, you know, you put
对。
in Yeah.
你的想法，然后输出产品。
Your ideas, you output product.
这是真的吗？
Is that real?
这到底是怎么运作的？
How how does that really work?
差不多就是这样运作的。
So it kinda works like that.
我觉得工厂大概有两种主要的使用者类型。
The the way that I think of of factories is, like, there's there's basically two types of personas who are using them.
那我给你展示一下我们在 Warp 里是怎么操作的。
So there's I'll show you like basically how we do this in Warp.
一种是所谓的“构建者”角色。
There's like the the builder persona.
这种角色的人主要是贡献想法，并且想动手做一些东西。
So the builder persona is someone who is contributing ideas, wants to build stuff.
从这个角色来看，至少按我们目前的设置，它看起来跟团队里的工程师、设计师或产品经理用本地编码助手构建东西的方式没什么太大区别。
And from that persona, it doesn't look like, at least the way that we have it set up, it doesn't look that different from like how one of these engineers or designers or product people on our team might build with a local coding agent.
这里能看到的一个主要区别是，这是我们的 Slack 频道，大部分工作都是从公开的地方开始的。
The big difference you can see here, so this is like, this is our Slack, is that most work is starting in a public place.
所以我们把工厂的概念理解为：你基本上可以在公开的 Slack 频道里说，可以@我们叫 Wilson 的那个工厂，然后说‘好，我想让你帮我们做点东西’。
And so the way we think of factories is like, you basically can say in a public Slack channel, you can tag, we named our factory Wilson, you can tag Wilson and be like, okay, I want you to to build something for us.
当你这么做的时候，Wilson 做的不仅仅是直接构建——就像你用 Cloud Code 或 Codex 时得到的那样——它还会多做几步其他操作。
And when you do that, what Wilson will do is like, essentially do not just like the direct build, which is what you might get if you're using like Cloud Code or Codex, but we'll also like do a few other steps.
所以它会做的是，它会经历所有步骤，比如，首先它会分类处理这个请求。
So what it will do is it'll kind of go through all the steps of like, first it will triage whatever this is.
所以它会把它当作工厂的输入，而不是直接要构建的东西，意思是，对于这个例子，Harry 说他想改变这个功能的外观。
So it'll treat it as like an input to the factory, not just like a direct thing to build, meaning it will, you know, for this one, Harry's saying he wants to change the way this feature looks.
他给出了相当详细的指示。
He gives like pretty detailed instructions.
他附上了一张图片。
He, you know, attaches an image.
他说使用计算机操作，这会启动一个流程，工厂做的第一件事实际上是打开一个问题。
He says use computer use and this kicks off a flow where the first thing that the factory does is actually kind of open an issue.
所以你可以在这里看到，我们除了使用 Slack 之外，还使用 Linear，这样所有事情都能被追踪。
So you can see over here, it's like, we use Linear in addition to using Slack so that everything gets tracked.
所以工厂不仅与 Slack 集成，它还会打开一个问题。
So the factory integrates not just with Slack, it opens an issue.
然后它进行实际的实现。
Then it does the actual implementation.
对于这个来说，它比较简单，也比较模糊，所以我们实际上能实现它。
For this one, it's like, it was simple enough and ambiguous enough that we could actually implement it.
然后它会创建PR。
Then it does, it will create the PR.
所以它也跟GitHub集成。
So it integrates also with like GitHub.
然后它做QA，我觉得这在现在真的很重要。
Then it does the QA, which I think is like really important these days.
如果你不想把所有时间都花在代码审查上，最有帮助的一件事就是让这些工厂代理做计算机使用验证。
Like if you wanna not spend all your time doing code review, one of the things that's most helpful is like you could have these factory agents do like computer use verification.
所以在这个案例中，它制作了一个完成功能的视频。
So in this case, it like made a video of the completed feature.
它展示了整个过程，展示了所有的按键操作。
So it's like showing the thing, showing all the keystrokes.
所以你得到的不仅仅是合并，对吧？
And so what you get here is is not just like, and then it gets merged, right?
不只是像构建、构建东西这样的一步。
Is not just like the single step of build, build the thing.
如果你在工厂出现之前用编码代理来做这件事，通常的做法是打开你的本地编码代理，在那里做修改，在本地测试，然后推送到GitHub。
Like if you were to do this with like a coding agent in like the pre factory world, what you would typically do is like pull up your local coding agent, do the change there, test it locally, and then like push it up to GitHub.
但你得到的是完整的流程，包括工单、代码审查、PR、视频，所有这些都一起完成。
But you get the whole thing of like the ticket, the the code review, PR, the video, all all done together.
然后另一个真正有点神奇的地方是，这一切都是公开进行的。
And then the other, like, really, kinda magical thing is that it's all done in public.
所以如果你是……
So if you are...
其他人想进来看看这个，看看任务是怎么完成的，甚至想参与进来。
You know, someone else wanted to come in and like, look at this and see how the task was done, or even contribute to it.
所以我们都有很多人，比如在Slack线程里的多个人，你作为一个构建者，不再是在本地孤岛中工作。
And so we all have people like multiple people on these Slack threads, you get to a world where as like a builder, you're no longer sort of working in this local silo.
相反，你在一个公共空间里工作，每个人都能看到你在做什么。
Instead, you're working in a public space where everyone can see what you're doing.
所以，从开发者的角度来看，他们就是这样使用工厂的。
And so that's like, from the builder's point of view, how they, like, use the factory.
到目前为止，这样讲清楚了吗？
Does that make sense so far?
我觉得我在帮那些可能还在纠结的人理清思路，比如，到底什么是编码代理？
And, you know, I think what I'm reflecting for folks who are maybe still trying to grapple with, okay, like, what's a coding agent?
什么是工厂？
What's a factory?
它们有什么区别？
What's the difference?
我听到你说的是，这个所谓的工厂，是全面的……
I think what I'm hearing from you is the factory, quote unquote, is comprehensive...
它是根据你对软件开发生命周期的理解，全面设计出来的。
Comprehensively designed to reflect your version of what the software development life cycle should be.
所以它不仅仅是，从想法到代码，再到拉取请求，最后推送。
And so it's not just, like, idea to code to PR to push.
对，没错。
It is, okay.
一个想法需要经过好几个步骤。
An idea needs to go through several steps.
它需要沿着产品的传送带走下去，进入问题跟踪的漏斗。
It needs to go down the conveyor belt of product and into, like, the funnel of issue tracking.
然后我们需要编码，然后需要做质量保证，然后需要有非常具体的验证循环，我们可以拿我们的贴纸，说质量由一号、二号、三号控制。
And then we need to code it, and then we need to QA it, and then we need to have these very specific verification loops where we can take our sticker and say quality controlled by, you know, number one, two, three.
所以对你来说，按照你的描述，它超越编码代理或副驾驶的地方在于，它实际上是设计来处理非常具体的端到端产品步骤，而不仅仅是工程那种输入输出。
And so to to you and the way you're describing it, where it goes beyond a coding agent or a copilot is that it is actually designed to take take the very specific end to end product steps, not just the engineering kinda, like, input output.
那绝对是其中的一部分，当然。
That's that's definitely part of it for sure.
它确实做了更多软件生命周期的部分。
It's like it does it does more of the the software life cycle.
所以它做了更多这些步骤。
So it does more of those steps.
对我来说，还有更大的一部分，那就是……
There's a whole other bigger part of it to me, which is...
所以我是从个人开发者的角度给你展示这个的。
So I showed you this from like the perspective of the individual builder.
如果你进我们的 Slack 看看，你会发现这种情况整天都在反复发生，我们团队的人都是这样工作的。
If you were to go into our Slack, you would see like this is happening over and over like all day long, the people on our team are building in this way.
而且顺便说一句，有时候不只是工程师这样发起工作，也可能是自动发起的。
And sometimes by the way, it's not just like the engineers who were kicking off work like this, it could be work that's being kicked off automatically.
比如，我们有一个叫 Sentry 的系统，不知道你知不知道，它是个崩溃报告系统。
So for instance, we have, I don't know if you know Sentry, but it's like a crash reporting system.
所以当我们的终端应用有崩溃报告时，我们会收到信号，然后尝试自动修复，这些也会进入工厂流程。
And so we have signals when there are crash reports coming from like our terminal app that we try to automatically fix those and those go into the factory too.
所以它可以是人工发起的，也可以是外部系统发起的。
So it can either be human initiated, it can be initiated by like an external system.
实际上，它可以在这些工具中的任何一个里发起。
And it can be initiated actually, in any of these tools.
所以，你可以发起它，比如我们可以在Linear里启动整个过程，也可以在GitHub里启动。
So it's like, you can initiate it, like, we could have started this whole thing in Linear, we could have started this whole thing in GitHub.
所以它已经集成到所有工具中了。
And so it's it's integrated into all the tools.
但在我看来，工厂模式更重要的部分不在于个人开发者如何工作，因为这与那些可能直接进入终端、使用编码代理做事的个人开发者并没有太大区别。
But the bigger piece of the factory approach, in my opinion, is not like, it's not so much how it, the individual builders work because it's not all that different from individual builders who might be using like, like, who might be going into the terminal, for instance, and using a coding agent to do something.
它有所不同。
It's different.
它是公开的。
It's in public.
但工厂模式真正不同的地方在于，一切都集中在云端。
But what's really different in the factory approach is that everything is like centralized in the cloud.
还有另一个方面，不是针对开发者，而是针对那些试图在团队中扩展软件开发的管理者。
And there's a whole other aspect to it, which is for not the builders, but for the people who are, like, the managers who are, like, trying to scale software development on their team.
给管理者的东西。
Something for the manager.
这是大家一直想要的。
It's what everyone's been wanting.
对开发者的关注太多了。
Too much attention on the builders.
不。
No.
但说真的，如果你……
But in all seriousness, if you're...
比如，作为公司经营者，我担心的是，我想要软……
Like like, my concern, as somebody who's running a company, it's like, I want I want soft...
这市场竞争超级激烈。
Like, it's a super duper competitive market.
我想看看我们进展有多快。
I wanna see how quickly we're moving.
我想确信我们开发软件的方式确实在随时间改进。
I want confidence that the way that we're building software is actually improving over time.
我想确保我们没有浪费太多钱。
I want to make sure that we're not like wasting too much money.
而且在这个每个开发者都有自己的本地环境的世界里，这真的很难做到。
And in the world of like all of like, every builder has their own individual local setup, that's very, very hard to get.
所以这简直是工程经理的梦想。
And so this is like an engineering manager's dream.
也是CTO的梦想。
It's like a it's like a CTO's dream.
不是说这是世界上最酷的东西。
Like, not that this is the coolest thing in the world.
但现在我有了集中查看每个人在做什么的能力。
But it's like, now I have the ability to sort of centralize and see what everyone's doing.
比如，在这个屏幕上，这是我们团队实际的数据，关于我们如何使用工厂的。
So for instance, on this screen here, it's like these are this is actual data from our team in terms of, like, how we're using the factory.
这里有一种感觉，就是工厂的自动化程度如何。
It's like there's a sort of sense of, like, how automated is the factory?
我们的速度怎么样？
What's our velocity?
我们发布东西需要多长时间？
How long does it take us to, like, ship stuff?
所以这里有……
And so there's this...
所有这些衡量指标。
All these measurements.
我们能先暂停一下，聊聊你的生产力吗？
Can we pause real quick on your productivity?
因为我之前真的没见过这个指标，而我们见过很多指标，比如每个PR的人机交互次数。
Because I actually haven't seen this measure before, and we've seen a lot of measures, which is human interactions per PR.
对。
Just Yeah.
跟我们说说为什么那个……
Talk us through why that...
说实话，我之前没听过这个，但我挺喜欢的。
Because, honestly, I haven't heard that one before, and I like it.
这背后的直觉是，我们会变得……
You know, this sort of instinct underlying this is that we are gonna be...
不是说得不好听，但人类在生产过程中有点像瓶颈。
Not not to put this the wrong way, but it's like humans are a little bit of the bottleneck in terms of production itself.
他们同时也是创造力的来源。
They're also the the creative force.
但总的来说，公司想要的，以及我们为Warp这样的公司努力打造的，就是如何实现更多自动化？
But in general, what companies want and we're, you know, we're trying to build for companies that Warp largely is like, how do you automate more?
怎么让那些真正可以自主完成的事情，以几乎全自动的方式完成？
How do you make things that can truly just be done agentically be done in an almost fully automated way?
所以，一般的直觉是，你需要提示的次数越多，需要引导或哄着你的代理的次数越多，那就会限制吞吐量。
And so, you know, the the general instinct is like, the more times you have to prompt, the more times you have to steer or like like cajole your agent, that's like, that's gonna be a limiter on throughput.
所以这就开始进入工厂化的世界了。
So this is where you start to get into like factory world.
就像，想象一下你在运营一个特斯拉工厂之类的。
Like, this is like, imagine you're running like a Tesla plant or something.
就像，你需要让生产线停下来多少次？
It's like, how many times do you have to stop the line?
所以我们试图给工程领导者一个视角，比如，好的。
And so, you know, we're we're trying to give engineering leaders a view of, like, okay.
如何……
How...
你是否能够随着时间的推移让事情变得更高效？
Are are you are you are you able to sort of make things more efficient over time?
快速问一下。
And quick question.
这些只是因为我的想法……
Are those just because my mind...
我……
I mean, I...
作为曾经的……
You know, as a once...
作为当过CTO的人，我心想，对。
One time CTO, I'm like, yeah.
这正是我想要的。
This is exactly what I want.
这些互动……
Are these interactions...
你会把这些互动看作是提示，或者说前期的引导吗？
Do you think of these interactions as, like, the prompts, the sort of, like, upfront steering?
你会考虑PR上有多少评论吗？
Are you thinking about, like, how many comments on PRs?
有多少轮循环？
How many loops?
比如，每个PR的互动包容度评分怎么样？
Like, how inclusive is this, like, interaction per PR score?
这些全都包括。
It's it's all of those.
所以因为……
So because the...
而且，我觉得这个肯定可以随着时间不断优化。
And, again, I think this could definitely be refined over time.
你要知道，我们正在努力搞清楚什么才是合适的指标组合，但这个实际指标，因为工厂已经集成到你所有的知识工作工具里了，它包含了所有那些东西，比如在Slack里重新提示了多少次，在Linear上评论了多少条，在代码审查中你纠正了多少次。
And bear in mind, we're like, trying to figure out so what is the right set of metrics, but this actual metric, because the factory is like, integrated into all of your knowledge work tools, it includes all those things like how many re prompts in Slack, how many comments on Linear, how many times did you have to correct the thing in code review.
所以它有点像，对，它就是一个代理指标，衡量你为了让这个东西完成任务，自己还得做多少工作。
And so it's it's like a kind of like, yeah, it's a proxy for how much work did you have to do in order to, you know, get the thing to do the job.
好的。
Okay.
等等。
Wait.
我想跟你抬抬杠。
I wanna I wanna give you a hard time.
还有一件事。
One more thing.
对。
Yes.
因为你刚才说到……
Because you talked...
但在我们开始录音之前，你问过，比如，技术性有多强？
You know, but before we started recording, you said, like, how technical?
我当时就想，好吧。
And I was like, okay.
那我们现在就戴上CTO的帽子，至少是CTO的帽子。
Well, we're gonna put on the CTO, at least the CTO hat right now.
好的。
Okay.
我能跟你较个真吗？人类真的是瓶颈吗？
Can I give you a hard time that humans really are the bottleneck?
因为你看从启动到提交PR的时间，只要三十五分钟。
Because if you look at kickoff to PR time, it's thirty five minutes.
但从PR到第一次人工审查，却要三个半小时。
But if you look at PR to First Human Review, it's three and a half hour.
三个半小时。
Three and a half hours.
所以这周期时间真是挺搞笑的。
And so it's like so funny that still cycle time, man.
这才是你真正需要关注的重点。
Like, it's it's the thing you really have to focus on.
就是那个数字。
It's that it's that number.
对。
Yep.
瓶颈还是在那里。
There's still the bottleneck.
而且我们……
And we're...
我们还在做人工代码审查。
I mean, we're still doing human code review.
你们所有的PR都会经过人工代码审查吗？
Do do all all your PRs get human code review?
目前，我们所有的PR都会经过人工代码审查。
Currently, all of our PRs get human code review.
哦。
Oh.
所以……
And so...
这其实是一个团队选择，组织层面的选择。
Now this is like, I think of a team choice, an organizational choice.
我们有……
We have...
我们在这方面唯一改变的是，不像以前那样要求……
The one thing that we have changed around this is not like we used to require...
以前的工作流程是，我们团队里的A用智能体构建东西，然后B来审查智能体写的代码。
Like the workflow used to be like person A on our team would build something with an agent and person b would review the agent's code.
我们现在不再要求这样了。
We no longer require that.
也就是说，给智能体下指令的人也可以……对。
Like the person who prompts the agent can also Okay.
审查它的
Review its
你让我觉得
You're making me feel
那样更好，但我们还是不能完全信任它。
that That's a better thing, but it's still like we, you know, we we don't have yet complete trust.
我觉得未来会发展成这样：有一部分东西，我们最终会有足够的信心，可以跳过这一步。
I think the way this will evolve is like some percentage of some percentage of stuff will, will eventually will feel confident enough that we can skip that.
我最近做了一期节目。
I did an episode recently.
我建了一个叫 Merge Mommy 的 AI 代理。
I built, an Eve agent called Merge Mommy.
然后这个，嗯，好吧。
And this is this Okay.
这是我经常看到的流程。
This is the flow that I often see.
即使是成熟的工程团队，做法基本上就是自动给每个 PR 打风险分。
And even mature engineering organizations is what you do is you basically, like, risk score every PR automatically.
所以先过一遍，给它打个分。
So go through your risk score it.
然后任何低风险或极低风险的，Merge Mommy 就直接盖章批准，然后人类就可以直接按按钮合并了。
And then anything that is low or extra low risk gets a stamped approval from Merge Mommy, and then a human's allowed to just smash the button and merge.
任何中等偏高或者有异常风险的，就需要人工审查。
Anything that's, like, medium high or whatever or has some, like, outlier on risk requires human review.
这样就能把队列底部的那部分任务处理掉。
And that just, like, lets you get that bottom tranche of the queue out.
同时，也能确保你有足够的能力去对真正重要的事情进行高质量的人工审查。
Also, just so you have good capacity for high quality human review on the things that really matter.
完全同意。
A 100%.
我觉得代码审查实际上就是风险管理。
I think code review becomes an exercise in risk management.
确实是这样。
I think that's right.
我想给你看看工厂里的其他东西，让你明白这跟普通的交互式代理有什么不同。
I wanna show you some other stuff in the factory just to show you like, how this is different than just standard interactive agents.
现在大家都很关心成本问题。
So you get, cost is on everyone's mind right now.
所以你会看到，这就是亚洲的成本。
And so you get like, this is Asian cost.
目前还没有把人力成本算进去。
It doesn't factor in the sort of cost of the people at the moment.
但你可以看到，几周前我们的成本还很高，后来我们调整了模型配置，把成本降下来了，而且我们还想继续降。
But it's like, you know, you can see, like, we were really expensive few weeks ago, we made some changes to our model configuration and we've driven this cost down and we want to continue to drive it down.
但关键是，整个团队能有一个集中的视图来看这些数据。
But just like having the centralized view for this across your whole team
我能问一个关于成本的问题吗？
Can I ask you a question about cost?
关于成本，我有个小问题。
Quick question about cost.
这个看起来挺……
So, I mean, this is a pretty...
你可以看到这里的下降趋势。
You can you can see the the drop here.
我跟很多工程团队交流时，经常看到这种情况。
I I see this a lot when I'm talking to engineering organizations.
大家经常会发现，随着AI的引入，每次提交代码的成本会上升。
They often see, like, a rise in cost of PR as their...
毕竟你们已经在用AI了嘛。
Per PR as they're adopting AI, which you all already have.
然后我们通过优化，成本又降下来了。
And then we see this drop as you're doing optimization.
你觉得现在最大的降本手段是换模型吗？
Do you feel like the biggest lever here is model right now?
就像那个……
It's like that the...
是指这个调节因素吗？
Is that the lever?
是模型。
It's it's model.
我觉得模型是最关键的，还有上下文，比如那种方式
I think model is the biggest, I think, like, context, like the way
上下文管理，
context management,
上下文管理是次要的。
like context management matters as a secondary thing.
我觉得模型是最重要的。
I would say model is the biggest.
而且，真正搞清楚这个问题的方法就是测试，对，这也是我想展示的。
And that the the way to really figure this out actually is to test, which I Yeah.
这也是我想展示的。
Is something I wanna show as well.
基本上，我觉得模型是最重要的。
Basically, I think I think model is is the biggest one.
还有其他关于……的问题吗？
Any other questions on the...
就是关于支出视图的？
Just like the spend view?
太好了。
That's great.
所以，另一件事，就是这里真正变得像工厂的地方，就是我们改变开发方式的方式。
So the, the other thing, like, this is where it really gets kind of factory oriented is like the, the way that we're changing development.
我觉得这对正在观看的人来说是最相关的。
And this is, I think this is most relevant for folks who are watching.
我们试图管理这些团队并扩展编码代理，现在我们几乎在衡量一切。
We're trying to manage like these teams and scale coding agents is like we now we're like measuring everything.
所以这里的最后一项，我们称之为评分，基本上让你看到所有代理运行在不同维度上的表现。
And so this last thing in here, which we call scoring, basically gives you a view across all your agent runs of like how they're doing on different dimensions.
因为我们基本上把工厂当作一个闭环系统，每次代理在工厂里执行任务，那个任务都会被记录下来。
And so we we basically because the factory is like a closed kind of like closed loop system, every every time an agent does a task in the factory, that task gets recorded.
这带来的可能性是，你可以回溯性地查看任务完成得怎么样。
And what that opens is the possibility for you to go and like, like go back retroactively and look at how well the task was done.
所以你可以作为一个人来做这件事。
And so you could do this as as a person.
你完全可以自己回去看看那些任务，但还有一件事你可以做，就是让代理来做这件事，这也是我们正在做的。
You could literally just go back and look at at tasks, but the other thing that you can do is you can have agents do this, which is is what is what we do.
比如说……
And so for instance...
让我看看能不能找到一个有意思的例子。
Let me see if I can find an interesting one.
这里就有一个有意思的。
Like, here's an interesting one.
比如，冗余测试。
Like, redundant tests.
如果你正在处理……
So if you're if you're working with...
你知道我在说什么吧？
You you know I'm talking about here?
哦，我知道你在说什么。
It's like Oh, I know what you're talking about.
我每次提交PR，都得跑一遍并设置好135个测试。
Every PR I push, it's like, I have run and set up a 135
没错，就是测试。
tests Exactly.
这里。
Here.
就是那个……
It's it's the...
就像是……
It's like it's it's like...
之所以会这样，是因为他们写测试不只是为了防止回归，也是想顺便验证一下行为。
And the reason it does this is because, you know, they're writing tests not to, like, prevent regressions necessarily, but also just, like, test behavior along the way.
所以最后就会有一堆用不上的测试。
And so you end up with a bunch of tests that you don't need.
所以如果我们采用工厂模式，你大概能预见到这可能会出问题。
And so we, you know, if you take the factory approach, you kind of know this might be a failure mode.
你可以做的是，写一个评分器，用大语言模型当裁判，然后说，好，我要去查看所有的智能体运行记录。
And what you can do is you can write essentially like a score that uses LLM as a judge to be like, okay, I want to go and look at, you know, all of the agent runs.
我想看看，当一个智能体认为另一个执行任务的智能体写的测试有冗余时，这种情况有多频繁。
And I want to see how often we're, you know, an agent thinks a different agent than the one that did the task thinks that there's redundant tests.
这样讲能理解吗？
This makes sense?
对，完全理解。
Yeah, totally.
所以基本上你就是说，好，我想要……
And so you basically are like, okay, I want...
我会选一个裁判模型。
I'll I'll pick a judge model.
这个裁判模型属于中等聪明的那种。
This is like a kind of medium smart judge model.
你不想让它太贵。
You don't want it be too expensive.
不然的话，你会在评分上花很多钱。
Otherwise, you end up spending a lot of money on scoring.
你可以按任务来分类，看看这个任务的表现怎么样。
You have it classified in terms of like, for this task, like, how did it look?
你可以选一个采样率，来决定怎么操作。
You pick like a sort of sampling rate on in terms of how you want to do this.
然后随着时间推移，你会得到一系列运行结果，从中可以看到，有时候这个代理会认为有一些多余的测试。
And then you get over time a set of runs where you can see that, like, sometimes this agent thinks that there were some, like, surplus tests.
这就是我说的真正的衡量标准。然后你可以作为人类，亲自去看看这里发生了什么，不过我会给你演示的。
And so this is what I mean by, the real measurement, then what you can do is, like, you can go and you can actually, you know, as a human, you can go and kinda look at what's happening here, but I'll show you.
在工厂环境里还有更好的方法，就是你可以用代理来识别这些运行中哪里出了问题，然后尝试改进工厂。
There's also a better way to do this in the factory world where it's like, you can actually use agents to sort of identify what's gone wrong in these runs and try to improve the factory.
这样讲得通吗？
Does this make sense?
我很好奇。
I'm curious.
你是按每次运行来算，还是按一组PR的总体情况来算？
Do you do this, on a per run basis or on an aggregate basis across a set of PRs?
总体来看。
Aggregate.
问得好。
Great question.
那……
So...
对，我们做的就是评分，我管这个叫评分，评分是跨所有运行来进行的。
Yeah, so what we do is like the the scoring, and this is what I call like this thing, we call this like the score, the scoring happens across all runs.
但还有一个第二循环，就是现在推特上很火的所谓自我改进，你拿一个代理，然后基本上对它说，好。
But then there's a second loop, which is another like, popular thing on Twitter right now called self improvement where you take an agent and you basically say, okay.
对于所有失败的运行，你需要一个真实的样本，比如20到25个失败的运行。
For all of the failed runs and you need, like, you need, like, a real sample, like, maybe 20, 25 failed runs.
你需要一个足够大的样本量。
You need some significant sample size.
否则，它就会开始基于单一的事情过度纠正。
Otherwise, it starts to overcorrect based on single based on single things.
你让它看看……
You have it look...
你有一个观察者代理，它可以对工厂进行更新，以尝试防止特定的失败模式。
You have, like, an observer agent look, and it can then create updates to your factory that will try to prevent the particular failure mode.
所以，让我看看能不能找到一个，我不太清楚这个。
And so let's see if I can find one that's like, I don't know this.
我只是随便挑一个。
I'm just literally picking a random one here.
但这发现了一些驱动工厂的技能问题。
But this is finding some issue with our skills that are driving the factory.
它提供了证据，并说，好的，我们应该以某种方式改变其中一个工厂代理的定义。
It's presenting evidence and it's saying, okay, we should change the definition of one of our factory agents in a particular way.
如果你深入了解一下，它实际上是在改变工厂代理应该做的第10步。
And if you go under the hood and look at this, it's like, it's changing this like step 10 of what our factory agents should do.
你明白了吗？
Does it make sense?
是的。
Yep.
所以对我来说，这真的很令人兴奋。
And so this is, to me, this is really exciting.
让这一切能够运转起来的关键是，你用代码来定义工厂。
Like the way, the thing that's, like, enabling this whole thing to work is that, you define the factory in code.
代码。
Code.
对。
Yep.
我的意思是，如果你去看定义，你会问，工厂是什么？软件工厂是什么？
And so, you know, what I mean by that is, like, if if you were to look at the sort of definition, you would say, okay, the factory what like, what is a software factory?
对我们来说，它至少是一个实际的名词。
It's like, it's like, for us, at least it's an actual noun.
这就像一个产品概念，它由一堆代码仓库、一堆MCP服务器、一堆配置，还有一堆代理组成。
It's like a product concept, where it consists of a bunch of repos, a bunch of like MCP servers, like a bunch of configuration and then a bunch of like agents essentially.
比如一个代码审查代理的设计，不同的代理，不同的自动化。
So like a code review agent design, like different agents, different automations.
而且这一切都是用代码定义的，在Warp工厂里这样做的好处是，你可以测试不同的配置，并且知道，你基本上是在某个时间点冻结了工厂的状态。
And it's all defined in code and the the value of doing it that way with, like in Warp factories is that you get the ability to actually you can like, test different configurations and know, like, you're basically freezing the state of the factory at a given point.
所以你可以说，好吧，如果我们改变工厂的定义，然后重新运行所有这些测试，我们就能看到事情是否变得更好。
So you can be like, okay, if we were to change the factory definition, then like run all these tests again, we could see if things were better.
这也使得代理实际上可以……
It also makes it so that like, an agent can actually...
因为这些都是……
Because these are all...
全都是代码。
It's all code.
所以编码代理实际上可以更新工厂，让它变得更好。
Like coding agents can actually update the factory, to make it better.
展开剩余字幕（还有 305 条）
你明白我的意思吗？
Does this make sense?
本期节目由OpenArt Arena赞助播出，这是一个创意智能领域的全球排行榜。
This episode is brought to you by OpenArt Arena, the global leaderboard for creative intelligence.
每周都有新的AI模型发布，每个人都声称自己是最好的。
Every week, new AI models launch, and everyone claims to be the best.
但最好是指哪方面呢？
But best at what?
OpenArt Arena就是为了回答这个真正重要的问题而建立的。
OpenArt Arena is built to answer the question that actually matters.
哪个模型最适合你的具体任务？
Which model is best for your specific job?
OpenArt Arena不是评选一个总冠军，而是根据真实的创意任务对模型进行排名，涵盖广告、电影、动画、产品、平面设计、编辑和口型同步，同时覆盖图像和视频。
Instead of one overall winner, OpenArt Arena ranks models across real creative tasks from advertising and film to animation, product, graphic design, editing, and lip sync, covering both image and video.
而且这些排名不是基于炒作。
And these rankings aren't based on hype.
这些排名是由专业人士、行业领袖和在职创作者通过盲评来评判的。
They're judged by professionals, industry leaders, and working creators through blind evaluations.
所以评委们根本不知道哪个模型生成了哪个作品。
So judges never know which model produced which output.
这意味着你可以看到模型在你所做的创意工作中实际表现如何。
That means you can see how models actually perform when it comes to the creative work you're doing.
所以别再猜该用哪个模型了。
So stop guessing which model to use.
去查看基于真实创意工作的排名，找到适合你项目的模型，节省时间和成本。
Explore rankings based on real creative work and find the right model for your project and save time and cost.
去 OpenArt Arena 看看排名吧。
See the rankings at OpenArt Arena.
我想稍微总结一下我们目前的进展，因为我觉得这里面有很多丰富的内容，尤其是对工程领导者和开发者来说。
I just wanna, like, you know, kinda sum up where we are so far because I think there's so much rich stuff in here, especially for engineering leaders and builders.
而且我知道我们接下来会讨论一些不专注于工程的工作流程，这让我很期待。
And then I know we're gonna get to some workflows that are not engineering focused, which I'm excited about.
但是……
But the...
你讲的时候，我看到了几个趋势和主题。
You know, a couple trends themes that I've seen as you talk through this.
第一，工作是在公开场合进行的。
One, work happens in public.
所以，我注意到一个转变：以前工作是分配成任务单，然后在笔记本电脑上处理，现在工作是在公开场合进行的，比如在Slack或其他渠道上。
So I'm seeing, you know, this move towards instead of work being assigned in tickets and then tickets being worked on on laptops, work is happening in public, whether it's Slack or some other channel.
然后，再在云端执行，这样基本上任何人都可以参与其中。
And then and then executed in the cloud, so sort of anybody can interact with that.
这个工厂确实有类似软件开发生命周期的定义。
The factory does have this software development life cycle kind of, like, definition to it.
但除此之外，你还在对所有行为进行汇总性的元分析，正如我们所说，这给了人们，也就是管理者，他们想要的东西，那就是：我们是否在随时间变得更高效？
But on top of that, it has this aggregate meta analysis that you're doing across all the behaviors, which is giving, you know, as we said, giving the people, the managers, what they want, which is, are we getting more efficient over time?
这个工厂真的有效吗？
Is this factory actually effective?
有多少人类？
How many humans?
有多少智能体？
How many agents?
两者之间的平衡是什么？
What's the inner balance between the two?
剧透一下，目前问题出在人类身上。
Spoiler alert, the humans are the problem right now.
然后，你还在针对工厂中的关键行为做评估，想要纠正这些问题。
And then, and then what you're doing is you're also doing these evals against key behaviors in your factory that you want to correct.
这一点我在和工程团队交流时经常提到，比如使用本地工具像 Cloud Code 时，挑战就在于你需要会话级别的遥测数据，这样才能汇总整个工程团队在会话中的失败情况。
And this is something that when I talk to engineering organizations, I tell them all the time, which is such a challenge when you're using something locally like, for example, Cloud Code, is I say you need session level telemetry because you need to be able to aggregate up the failures within sessions across your engineering team.
有些团队确实会这样做，比如把每个本地编码会话都上传到 S3，然后在自己平台上运行评估。
And so some teams do this truly by, like, sucking up every local coding session to s three and running their evals kind of like, on on their own in their own platform.
但我认为，如果你没有对每个编码会话进行会话工具调用、MCP 调用、测试失败、计算机使用级别的可观测性，你就会错过很多优化效率、成本、模型以及团队使用工具方式的机会。
But I do believe that if you do not have session tool call, MCP call, test failure, computer use level observability into every single single coding session, you're missing a lot of opportunity to optimize efficiency, cost, model, just like how your team uses the tools.
所以我觉得大家做这件事非常重要。
And so I think it's super important that people do this.
然后你又加了一层很棒的东西。
And then what you've added is this extra layer of great.
然后把这些洞察用来改进工厂，不管你怎么定义“更好”。
Then take those insights and make the factory better however you define better.
这个总结很棒。
This is a great summary.
太好了。
Great.
我做到了。
I did it.
专业播客主持人。
Professional podcaster.
有，有……
There's there's...
嗯，就是想让你了解一下，有了这些数据之后，你还能怎么继续深入。
Well, just to give a sense of how you can continue to, like, you can go even further once you have this data.
我觉得你的总结很棒。
I think your summary was awesome.
另外，你还能做的一件事，我觉得你也会去做，因为未来最重要的就是优化你构建和交付软件的方式。
The other thing that you can do and, like, I think are going to do this because this is this is how, like, there's gonna be nothing more important than optimizing the way that you, like, build and ship software in the future.
还有一件非常强大的事情，如果你按照你刚才说的，把追踪和评估都设置好了，还有一个方法能让它更强大，那就是加入一个功能，让你能拿自己的数据，用不同的配置重放那些会话，来实际衡量，就像你之前问的，如果用了不同的模型，从成本或质量角度看会怎么样。
And so one other thing that is really powerful, if you set up everything you just said where it's like you have the you have the tracking, you have the evals, There's one further way that you can make it even more powerful, is adding the ability to basically take your own data and replay those sessions with different configurations to actually measure, like to your question earlier, like, well, how would it have done from a cost perspective or a quality perspective, if you had done, like, different models?
比如说，这就是我们在Warp工厂里内置的功能，但实现的方法有很多种。
And so for instance, again, this is what we have built into Warp factories, but there's lots of ways you could do this.
就像，这个想法是你可以……
Like, the the idea is you can...
就像那些公开的基准测试，比如SuiteBench和TerminalBench，它们用的是通用数据，你也可以在自己的数据上重现，比如用过去真实的工厂任务，看看事情会怎么发展。
Just like there are these public benchmarks like SuiteBench and TerminalBench that, are on, like, generic data, you can recreate on your own data, like from real past factory tasks, how would things have gone?
比如，如果你用了不同的模型配置，所以这就像一种帕累托图，这些图在推特上到处都是。
Had you use a different model configuration, for instance, so this is like a kind of Pareto chart, which again, these things are on Twitter all the time.
但这里酷的地方在于，如果你采用工厂方法，并且非常科学地对待，比如，我想知道，假设你想整理一批前端任务，然后看看，我能不能在这些任务上用JLM-five three而不是Opus？
But it's like, what's cool here is like, if you take the factory approach, and are like really scientific around, okay, I want to know, like, like, want to, let's say you want to curate a bunch of front end tasks and then see, like, can I be using JLM-five three on those instead of using, you know, Opus?
是的，当然可以。
Yes, definitely.
我应该用Gemini three seven Flash吗？
Should I be using Gemini three seven Flash?
你会牺牲一些质量。
You're gonna take a quality hit.
所以，你可以自己权衡利弊，然后把这些反馈到模型路由策略中，这样你就有证据表明，在你自己的任务上，对于这些类型，成本和质量的最佳模型配置就是你所发现的。
And so, like, you get to choose your own sort of trade offs here and then you can feed this back into a model routing strategy where you actually have evidence that okay on your own tasks for these types of, like, the best model configuration for cost and quality is is, like, whatever you find.
你是在和实际发布的结果做比较吗？
Are you comparing that to what actually shipped?
比如，你是怎么评估质量的？
Like, how are you evaluating quality?
做CSS的方法多得是。
You know, there's, like, seven ways to skin a CSS.
比如说，你实际上是怎么判断的？对。
Like, how do you actually Yeah.
在前端任务上，怎么决定这个质量更好还是不好？
Decide, you know, this is better quality or or not on front end tasks, for example?
问得好。
Awesome question.
我们在WarpFactory里是这样做的……
So the the way that we do it in WarpFactory is...
你可能会用不同的方法，但我们用的是之前给你看的那套评分系统。
And you're gonna probably you can do different things here is we use the exact same like scoring infrastructure that I showed you earlier.
比如，我们这里有各种不同的评分维度。
So for instance, you know, we over here, we have all these different scoring dimensions.
如果你信任这套评分系统，基本上就是回放过去的任务，看看这些分数受到了什么影响。
You can, if you have confidence in this scoring infrastructure, basically you're replaying past tasks and trying to see if there was like how it affected these scores.
所以主要是用大模型当裁判，但你也可以……
So it's it's it's primarily LLM as a judge, but you can also...
这就像，你也可以用人类评审来做这件事。
It's like you you could do this with human judges.
你也可以用算法来做，但我们构建的是用LLM作为评审。
You could do it algorithmically, but the thing that we have built is LLM as a judge.
明白了。
Got it.
非常有意思。
Super interesting.
好的。
Okay.
你给了我很多关于如何进一步推进工厂这个想法的灵感。
So you've given me so many ideas on how to take the factory idea further.
我很好奇。
I'm curious.
先摘下我们作为构建者、CTO的帽子。
Just like take off our builder hat, our CTO hat.
我们戴上……
Let's let's put on...
在开播之前，我又跟你聊过。
I, you know, I was talking to you again before we came on.
就是说，大家特别喜欢看非技术人员做技术活儿，也喜欢看技术人员做非技术活儿。
Was like, people love to see how nontechnical people can do technical things and how technical people can do nontechnical things.
所以，给我们展示一些你其他的AI应用场景吧，可能不太涉及评估、基准测试或者MCP，更多是关于怎么当CEO的。
So show us some of your other, like, AI use cases that are maybe less about evals and benchmarks and and MCPs and more about, you know, being a CEO.
那我们就会，比如……
So we'll get, like...
我觉得，往大了看。
I think, like, zooming out.
这些图表看着挺酷的。
This is cool with all these charts.
但我还是不太会改Figma。
I still have a hard time making changes to Figma.
而且，我不知道你是不是也这样，但每次我想去改个Figma的东西的时候，我就觉得……
And, like, I I don't know if you're the same way, but when I go to try and change, like, a, like, a Figma thing, I'm like...
感觉就像我长了三根手指头似的。
It's like I have three thumbs or something.
就是，我……
Like, I just...
我真的搞不懂怎么用它，但我知道，如果你……
I cannot figure out how to use it, but I do know that it's like if you...
比如，我们做幻灯片就是这么做的。
Like, this is this is how we, like, do our slide decks.
比如说，我们现在用的是Figma的幻灯片功能。
For instance, it's like we we are using Figma slides.
你可以做出看起来很漂亮的东西。
You can make things that look very nice.
所以我现在常做的一件事就是，如果需要改幻灯片，我就通过Figma的MCP和编程代理来做。
So one thing that I do now is like, if I need to change a slide deck, do it through the Figma MCP and the coding agent.
所以，为了让大家看看具体是什么样子，我们这里有一个关于云执行的幻灯片，看起来有点普通。
And so like, just to kind of show what that looks like, we have this kind of semi boring like slide here on cloud execution.
那我给你演示一下我是怎么做的，让你有个概念。
So I'll show you how I would do this just so you get a sense.
我打算这样做，我在Warp里用这个。
So I'm gonna do this, I use this in Warp.
我用Warp作为编码代理。
I use I use Warp as a coding agent.
这在ClockCode或Codex里也能用。
This would also work in ClockCode or Codex.
我需要Figma MCP。
I need to Figma MCP.
然后我就把这个粘贴进去，接着我会跟它对话，这是另一回事。
And so I'll just paste this in, and then I'm gonna talk to it, which is another thing.
我猜大家都已经习惯用语音了，对吧？
I I assume people are on the voice train at
这个时候。
this point.
他们
They
用语音吗？
on the voice?
我们我们我们把所有功劳都归给希拉里·格里德利。
We we we give all credit to Hillary Gridley.
她管这叫“Yappers API”。
She calls it the Yappers API.
这是跟大语言模型对话的最高效方式。
Highest speak with way to talk to an LLM.
所以我要说，比如，我想给这张幻灯片做个新版本，用Figma MCP获取上下文，复制现有的幻灯片，而不是直接改它。
So I'm gonna I'm gonna say, like, I'd like to make a new version of this slide, use the Figma MCP to get the context, duplicate the existing slide rather than making changes directly to it.
让宿主框包含运行时框。
Let's have it be so that the host box contains the runtime box.
让我看看还有什么别的。
Let's let me see what else I got.
我得切回这边。
I gotta switch back here.
所以我们要让宿主包含沙箱。
So we're gonna have host contains our host should contain sandbox.
我回去改一下这个。
I'll go back and edit this.
我们把上下文系统做成一个像云一样的东西，环绕在这些周围，给它们提供输入。
Let's make the context system something that's like kind of like, you know, cloud around these things that feeds into them.
真正的CEO。
True CEO.
在幻灯片上加个云。
Put a cloud on the slide.
你会看到一些糟糕的设计。
You're gonna see some bad design here.
我们把启动台做成火箭主题的风格吧。
Let's make the Launchpad have a sort of, like, rocket type theme.
这是我的……
This is my...
顺便说一句，这不是我们销售团队想要的。
This is not what my sales team wants, by the way.
然后我们把追踪做成……
And let's make tracking.
我不知道，要不你想些点子，怎么把追踪做得更好。
I don't know why don't you come up with some ideas for how to make tracking better.
整体思路是让这张幻灯片比简单的五个方框更有视觉吸引力，并且从语义上展示这些方框之间的关系。
The overall idea here is to make this slide more visually appealing than the simple, you know, five boxes and semantically show the relation of the boxes to each other.
那我就这么做了。
So I'll do this.
你知道我对Figma说什么吗？
Do you know what word I say to Figma?
我说的是“语义化”。
I say semantically.
然后呢？
Now what?
对Figma MCP说的。
To Figma MCP.
真的吗？
Really?
天哪。
Oh my god.
一直这么说。
All the time.
语义化颜色，语义化。
Semantic colors, semantic.
这是我的“正交”概念。
It is it is my orthogonal.
这真的很好笑。
That's really funny.
我们走着瞧吧。
I mean, we'll see.
所以……
So...
而且我最近经常用Grok。
And I'm using I'm using Grok a lot recently.
我也是。
Same.
我不知道你偏好用哪个模型。
I don't know what your model of choice is.
我觉得Grok在成本、速度和质量的权衡上挺不错的。
I think Grok is pretty good from, like, the cost and speed and quality quality trade off.
我可以给你看看我还能做点别的，你知道的
I could show you how I could do some other, you know
为什么，为什么不……
Why why don't...
在加载的时候，你不如给我们展示点别的？
While that's loading, why don't you show us something else?
我再给你看一个。
I'll show you another one.
所以，我……
So I...
我也是Granola的重度用户。
I'm a big Granola user as well.
所以，如果我真的想改进这个销售演示文稿，我还会确保它讲的是客户真正关心的话题。
And so again, if I were really trying to improve this sales deck, another thing that I would wanna make sure is that the sales deck is speaking to the things that customers are actually talking about.
所以我要在这里用Granola MCP开始第二个任务，比如，你能不能通过Granola MCP回顾我过去四周的销售会议，然后整理出这些会议中客户最常问的十个问题，特别是关于Warp软件工厂的。
So I'm gonna start a second task here using the granola MCP where I'm like, can you use the Granola MCP to look back over my last four weeks of sales meetings and try to build up a list of the top ten ten frequently most asked questions in these meetings as they pertain to Warp Software Factories.
在总结里不要列出任何具体的客户信息，要匿名化，因为我在做播客。
Don't list any specific customer info in the summary, anonymize it because I'm doing a podcast.
那么
So
我们让这个也跑起来。
we'll we'll get this one cooking as well.
如果你愿意，我还可以再来一个。
I could do one more if you want.
咱们来……
Let's...
咱们把它们排队启动吧。
I mean, let's queue them up.
再次强调，我很喜欢看一个AI公司的CEO一个接一个地打开标签页，然后启动各种任务。
Again, I love to see a AI pill CEO just open tab after tab after tab and kick stuff off.
这些是……
I mean, is stuff that...
这些是……
This is stuff...
顺便说一下，这些都是我一直在用AI做的真实事情。
These are all, by the way, real things that I am constantly using AI for.
所以最后一个是我用AI做的测试，就是试着重新发现可能在过去六个月里联系过的冷线索，或者你知道，我们的产品变化很大，我可能想重新联系他们。
So the last one is like a test that I use AI for is trying to rediscover potentially like, cold leads that I might have talked to in the last, like, you know, six months or you know, our product has changed a crazy amount and I might wanna reapproach them.
所以我用来做这个的工具，或者过去一直用来做这个的，其实我最近刚开始用Instinct。
And so the thing that I I use for this or had in the past been using for this, I actually just started using instinct.
你用Instinct吗？
Do you use instinct?
哦，没有。
Oh, no.
我用Instinct有过不好的经历。
I had a bad experience with instinct.
你用这个有过不好的经历。
You a had bad experience with this.
所以我通常做所有事情的方式还是通过我的...
So the the the normal way I would do everything is still through my...
还是通过 Coding Agent，但我会……
Is still through Coding Agent, but I would...
让我们……
Let's let's...
我在这里再做一个提示。
I'll do one more prompt here.
你能用 GOG 命令行工具查找我过去六个月里与潜在企业客户的邮件和日历事件吗？这些客户可能对 Warp Factories 的再次外联有用。
Can you use the GOG CLI to look for emails and calendar events I've had in the past six months with potential enterprise leads who might be useful for another outreach for Warp Factories.
你可以在 warp.devfactories 了解 Warp Factories，并为我制作一个包含这些信息的 Google 表格，分享表格链接，但不要在这个对话中打印任何具体客户的邮件或姓名。
You can learn about Warp Factories at warp.devfactories and make me a Google Sheet with the info on them and share the sheet link, but don't print out any specific customer email or name in this thread.
我喜欢这个在元层面上的地方是，这是你大脑的工作方式吗？
What I like about this at the meta level is, like, is this how your brain works?
你就像，我需要做幻灯片，就启动幻灯片。
Where you're just like, I need to do the slide, kick off the slide.
我需要更新一些销售定位，就启动那个。
I need to, like, update some of our sales position, kick that off.
我需要跟进一下那些冷线索。
I need to, like, follow-up with cold leads.
就像，是的。
Like, is this Yeah.
当你做演示的时候，这是不是你作为CEO的思维方式？
When you panel, is this a reflection of your CEO brain?
因为这肯定也是我的思维方式。
Because it's certainly a reflection of my brain.
如果我处于销售模式的话，是的。
I mean, if I were if I were in, like, sales mode Yeah.
进入市场推广模式，而且我现在经常处于这种模式，说实话，比纯粹的开发模式或者运营管理模式更多，虽然我各种事情都做。
Go to market mode, and, like, I am in that mode quite a bit now, unfortunately, more more so than, like, straight up builder mode or engine manager mode, although I do all these different things.
我一直在想，正确的定位是什么？
I'm constantly trying to find like, what is the right positioning?
我们在这些会议中的沟通是否到位？对我来说，能有一个这样的能力真是太棒了，因为我完全不会设计或画画，现在有了这个超能力，我们可以继续了。
Are we are we communicating in these meetings in a way that lands and like, it's so amazing to have a disability, like, for me who can't design or draw at all or, like, to have this now superpower and we can go.
我们来看看这个Figma在这里进展如何。
Let's let's see how this Figma is doing over here.
它已经开始了吗？
Has it started to do it yet?
哦，开始了。
Oh, it's starting.
它还有一些工作要做
It's got some work to
做。
do.
但它会做得更好的。
But it it will it will do better.
不过，是的，如果我在销售模式，我会想，我们的定位对吗？
But, yeah, if I was in sales mode, like, I'm trying to figure out, are we positioning it right?
人们在问些什么？
What are people asking about?
对，这个说得没错。
Yeah, and this is this is right.
这真的是从我们实际的销售电话里来的。
Like this is from our actual sales calls.
而且我觉得，对我们这个工厂来说，最重要的一点就是，我们所在的这个产品类别，核心就是买还是自己建的问题，而且这个是有证据支持的，我觉得很有道理。
And like this is the what I would say is the number one thing for the factory, like the product category we're in is like buy versus build, and it's backed up by evidence and it makes sense to me.
安全问题经常被提到。
Security comes up a lot.
工作流程，具体是什么流程？
Workflow, what's the workflow?
怎么衡量效果？
What's the measurement?
怎么控制成本？
How do we manage costs?
太酷了。
It's so cool.
所以我的意思是，我想去确保我们的销售演示、网站这些都能回应大家关心的问题。
And so it's like, what I would do is like, you know, I want to get then go make sure that our our sales deck, our website, all this stuff is speaking to the questions people have.
我可能会把这个做成一个真正的常见问题解答。
I I might turn this into like a literal FAQ.
我觉得我们可能还有改进的空间，或者我……
Like, I think we we have, you know, probably ability to improve that or I...
我们就是试着把这个融入我们的定位里。
We just try and work this into our positioning.
但我一直在做这件事，努力去理解我们做的东西是不是对的，定位是不是正确。
But I'm constantly doing this and trying to understand if if what we're building is the right thing and we're positioning it the right way.
哦，我们有个火箭船。
Oh, we got a rocket ship.
抱歉。
Sorry.
我们回到Figma了，给那些没在看的人说一下。
We're back at Figma for those that are not watching.
我们有个火箭船。
We got a rocket ship.
我们有个像云一样的东西。
We got a cloud of some sort.
它正在努力。
It's trying
还不太好。
Not very good yet.
它在尽力了。
Trying its best.
它会变得更好的。
It will get better.
火箭船不算太差。
The rocket ship's not terrible.
还不错。
It's not bad.
扎克，这真的太好玩了。
Zach, this is this is so fun.
我想快速进入快问快答环节，然后我们就放你回去。
I want to go to quick lightning round questions, and then we'll get you back to
好的。
Cool.
回到工厂车间去。
Back on the factory floor.
好了。
Alright.
第一个快问快答问题。
First lightning round question.
每个人都想要工厂。
Everybody wants the factory.
每个人都想要那个愿意帮你翻通话记录的上市实习生，还有那个愿意帮你把丑丑的云图幻灯片做好的设计实习生，作为CEO。
Everybody wants the go to market intern that will happily go through your call transcripts and the design intern that will happily make your ugly cloud slides as a CEO.
但是当工厂运转顺畅，而且你们上个月处理了超过两千个PR的时候，你是怎么防止团队内部失控的呢？
But when you have the factory humming and when you're doing, I think it was, like, over 2,000 PRs in the last month, Like, how do you keep things in the team from going, as I say, like, reigns?
你是怎么掌握所有这些活动、工作、背景和任务，并且在最高层面确保你们在做重要的事情？
Like, how do you keep your arms around all that activity, all that work, all that context, all those tasks, and know at the highest level you're doing the things that matter?
这是个好问题。
It's a great question.
其实比较无聊的答案就是，我们作为软件工程师一直做的事情仍然适用，那就是我们在吃自己的狗粮。
Like, the kind of boring answer is, like, a lot of the stuff that we've always done as software engineers still applies, which is like, you know, we're we're dogfooding.
所以我们就是自己产品的重度用户，不断确保产品质量依然良好。
So we're just like huge constant users of the thing that we're building and making sure that like the quality of the thing still works well.
这里有一些很酷的知识传递在发生，我不知道你有没有注意到，就是……
There's like cool knowledge transfer that's happening, where it's, I don't know if you've seen this too, but, like, the...
这个……
This...
这些工具的使用方式有一种幂律分布，我们团队里有些人是非常非常高级的用户，而其他人则比较普通。
There's, like, a kind of power law for, like, how these tools are used in terms of, like, we have some people on our team who are super duper power users, and other people are, like, kinda more at the average.
所以就有这么个情况，当你……
And so there's this, when you...
当你在这种看起来很混乱的方式下工作，大家都在公开场合整天往这些工厂里塞东西，你确实有机会看到那些真正的高手是怎么用的，然后这就能带动团队里其他人提升水平。
When you're working in this, what seems like a very chaotic way in public with everyone slacking these factories all day, you do get a chance to sort of see how like the the really expert people are using it and that kind of up levels other people on the team.
然后，就像我之前说的，我们并不是完全混乱，不是每个人都一直往机器里乱塞东西。
And then, like I said earlier, we're still doing, you know, it's not just like chaos, like everyone like, you know, slop stuff into the machine all the time.
我们其实还是有一个产品开发流程的，在很多方面还是比较传统的。
It's like, we still have a product development process that, you know, is somewhat traditional in a bunch of ways.
比如说，我们会做大量的用户访谈。
So it's like we do a whole bunch of user interviews.
我们会观察人们怎么使用产品。
We watch people use the product.
在开始构建之前，我们会做设计工作坊，倒不是说要精确地确定界面应该长什么样，而是确保我们在解决正确的用户故事。
We do design jams before we start building to not not so much like figure out exactly like how the UI should look, but just to make sure that we're tackling the right user stories.
因为归根结底，你写软件再快也没用。
Like, cause at the end of the day, like it doesn't matter how fast you build software.
就是说，还是要解决用户的某个问题，而且要用好的方式去解决。
It's like, it's also solve some user problem and solve it in a good way.
所以我们正在想办法，既要保留人工参与的关键部分，又要利用这种神奇的技术来……
And so we we're, we're trying to figure out how to like, not lose the key parts of, like, the human input here, but also, you know, use use the fact that there's this, like, magic technology to...
让你能大幅提速。
That can make you just go much faster.
我很喜欢这个说法。
I love it.
我……
I...
说得太对了。
Amen.
没有比这更好的表达了。
I could not say it better.
好的。
Okay.
最后一个问题，然后我们就放你走。
Last question, and then we will get you out of here.
当你的AI没有按照你的想法去做的时候。
When your AI is not doing what you want.
而且，我也很好奇这个问题是不是……
And and, also, I am curious if this question is...
如果你私下用AI和在工作时用AI，你的回答会不会不一样？
If you answer this question differently when you're working with AI privately versus working in...
在Slack里面。
Within Slack.
当它没按你的想法做的时候，你的提示策略是什么？
With it's not doing what you want, what's your prompting strategy?
你会吼它吗？
Do you yell?
我最近……
I am lately...
我要向观众坦白。
I will admit this to the audience.
我经常这样，比如，你为什么这样？
I am doing a lot of like, why are why are you like this?
为什么？
Why?
为什么？
Why?
很多疑问。
A lot of questioning.
那很有趣。
That's funny.
我是个挺冷静的人。
I'm like a pretty level guy.
我想...
I'd like...
如果……
If...
我觉得如果你问我团队里的人，我不……
I think if you ask the people on my team, I don't...
我从来没有真的，像，真的吼过，那不是我的风格。
I've never really, like, really yelling not my style.
我觉得我算是那种有点阴阳怪气的。
I get like passive aggressive is kind of what I would say.
我会翻白眼，像，真的假的？
I I get I I get like eye rolly like really?
我们现在就要搞这个吗？
Like this is what we're doing now?
你搞了这么个东西，没人让你搞啊？
Like you built this whole thing that no one asked you to build?
所以我会有点烦，用那种微妙的语气，但我不是那种会吼的人。
So I'll get like annoyed like in like subtle undertones, but I I'm not a I'm not a yeller.
所以我也说不好。
And so I don't know.
而且我也尽量保持一种心态，就是这玩意儿简直太疯狂了，我居然在跟一个东西说话，而这个东西正在做我过去二十年一直在做的工作。
And I also try to just keep the perspective that this is just like a bonkers thing that I'm just like talking to this talking to this thing that is doing this job that I've done for the last twenty years.
现在它做得比我还好。
And it's now it's like doing it kind of better than me.
所以我也说不好。
And so I don't know.
我现在对它还是有点好奇的阶段。
I'm still in like, I'm in like a little bit of like a wonder phase with it.
不知道你有没有看过路易斯·CK的那个小品，他讲的是飞机上刚有Wi-Fi的时候，如果你知道这个，就打断我，就是，他……
I don't know if you've ever seen that Louis CK skit where he he talks about how when internet like Wi Fi first got onto airplanes, stop me if you know this, but it's like, he...
那是个小品。
It's a skid.
他就像，他在飞机上坐着。
He's like, he's riding on an airplane.
就像这样，二十年前，第一次有了Wi-Fi。
It's like this, like, twenty years ago, there's Wi Fi for the first time.
然后他就觉得，这太不可思议了。
And he's just like, this is incredible.
所有人……
Everyone...
就像，这太神奇了，但没人高兴。
Like, this is, like, amazing, and nobody's happy.
他讲了个故事，说旁边那个人想看点YouTube视频，然后双手一摊。
And he tells a story about how, like, the guy next to him is, like, trying to watch a YouTube video and, like, throwing his hands up.
就像，去他的。
Like, screw this.
就像，我一点信号都没有。
Like, I can't get any signal.
然后路易斯·CK就说，这太疯狂了。
And and Louis CK is, like, this is bonkers.
就像我们在飞机上一样。
Like, we're on an airplane.
这个东西正在飞向太空。
This thing is, like, going up to space.
给它一点时间。
Give it a second.
所以我觉得我们很容易失去对这项技术有多神奇的认识。
And so I think it's it's so easy to lose perspective of, like, how wild this technology is.
就像，我一般，算是比较……
Like, I'm I'm, like, I'm I'm generally, like, pretty...
可能比大多数人都更有耐心一些，我觉得。
Probably, like, kinda more patient with it than most, I would say.
我喜欢这个。
I love it.
这个观点太棒了。
That's such a great perspective.
好的，Zach，这次真的非常开心。
Well, Zach, this has been super fun.
我们在哪里能找到你，以及我们怎么才能帮到你？
Where can we find you, and how can we be helpful to you?
你可以通过个人渠道找到我。
So you can find me personally.
比如，我在推特上。
Like, I'm on I'm on Twitter.
我的推特账号是Zach Lloyd tweets。
I have Zach Lloyd tweets is my is my Twitter.
如果你想打造软件工厂，当然应该来warp.dev看看Warp。
You should obviously come check out Warp at warp.dev if you wanna build software factories.
我几乎没怎么提到它，但我们有一个非常受欢迎的开源智能终端，你也应该来看看并使用它。
You know, I I hardly even talked about it, but we have an extremely popular agentic terminal that's open source that you should also come check out and use.
那里是和交互式编程智能体一起工作的绝佳场所。
It's a great place to work with interactive coding agents.
不过，真的，这次太棒了。
But, yeah, this was awesome.
非常感谢你邀请我来。
I really appreciate you having me on.
感谢你参加《How I AI》节目。
Thanks for joining How I AI.
非常感谢大家的观看。
Thanks so much for watching.
如果你喜欢这期节目，请在YouTube上点赞和订阅，或者更好的是，在评论区留下你的想法。
If you enjoyed the show, please like and subscribe here on YouTube, or even better, leave us a comment with your thoughts.
你也可以在Apple Podcasts、Spotify或你最喜欢的播客应用上找到这个播客。
You can also find this podcast on Apple Podcasts, Spotify, or your favorite podcast app.
请考虑给我们评分和评论，这将帮助更多人发现这个节目。
Please consider leaving us a rating and review, which will help others find the show.
你可以在howiaipod.com上查看我们所有的节目，了解更多关于这个节目的信息。
You can see all our episodes and learn more about the show at howiaipod.com.
下次见。
See you next time.
关于 Bayt 播客
Bayt 提供中文+原文双语音频和字幕，帮助你打破语言障碍，轻松听懂全球优质播客。