From Idea to Product in 30 Min Using AI Agents (Full Tutorial) | Claire Vo

频道: Peter Yang
视频: https://www.youtube.com/watch?v=ikyxK6i0AL0
原文语言: en
统计: 共 84 轮

[0:00]

I don't say product management is dead because of doom and gloom. I say it's dead because I think you're going to be in a world of pain if you're not prepared for a shift in the next 18 months. I know lots of startups that have gone many, many years before hiring their first PM and kind of don't want to hire many. If you look at something like Devon, it is a language-based interface. You are telling a system what to do.

我说产品经理这行已经死了，不是在唱衰。我说它死了，是因为我觉得未来 18 个月会有一波大变化，你要是没准备好，日子会非常难过。我见过很多创业公司，干了好多年都没招第一个 PM，而且也不太想多招。你看 Devon 这种东西，它本质上是一个用语言来交互的界面——你就是在用语言告诉系统该干什么。

[0:27]

your ability to clearly articulate what to do in a way that a system can appropriately interpret it, access its set of tools and get the job done correctly is actually a really important skill. I tell all my engineering managers when they use Devon, I say, "Congratulations. I just gave you 10 interns." I would screenshot this. This looks great. Write me a purity I can send to my engineering team.

你能不能把要做的事说清楚、说得让系统能正确理解、能调动它手里的那套工具、最后把活儿干对——这其实是一项特别重要的能力。我跟手下所有工程经理说，他们一用上 Devon，我就说：恭喜你，我刚给你配了 10 个实习生。这个我得截个图，看着真不错。帮我写一份 PRD，我好发给我的工程团队。

[0:56]

Boom. Nice. It worked. It's there. Okay. Welcome everyone. My guest today is the one and only Claire Vo. Uh Claire is the chief product officer of Launch Darkly, founder of Chat PRD and also host of the brand new How I AI podcast. So, no product leader uses AI as much as Claire. So, I'm really excited to get a demo of her favorite AI workflows and discuss how AI will change product management forever. Welcome, Claire. Oh, thanks for having me. I'm really excited to be here. Yeah. So So I think you're like the first person that I actually interviewed. So So it's great to have you back.

搞定，漂亮，成了，就在这儿。好，欢迎大家。今天我请到的嘉宾是独一无二的 Claire Vo。Claire 是 LaunchDarkly 的首席产品官、ChatPRD 的创始人，同时还主持着全新的播客 How I AI。可以说没有哪个产品负责人像 Claire 这样高频地用 AI。所以我特别期待能看她演示一下最喜欢的几套 AI 工作流，也想聊聊 AI 会怎样彻底改变产品经理这个职业。欢迎你，Claire。哦，谢谢你请我来，我真的很高兴能来。是啊，我记得你应该是我第一个正式采访的嘉宾，所以这次能请你回来太好了。

[1:35]

Oh, good. So we're we're back at it. So maybe we can get right into it. Like maybe you can show us how you build new products with AI. Sure. Let's do it. Okay. We're doing the reverse how I AI move here. And I'm going to show you how I AI. So I'm going to share my screen. I'll just walk you through how I think about building products and how we might build something actually how we do build things at chat prd. So this is chatpd AI co c co-pilot for product managers and it helps you write and draft product documents and do other product work more efficiently and with better quality. And you know, one of the uh features of chat

哦，太好了，咱们又凑一块儿了。那不如直接进正题吧，你可以给我们演示一下你是怎么用 AI 做新产品的。当然，来吧。好，我们这次反着来一遍 How I AI 的玩法，我来给你们演示我是怎么用 AI 的。我先把屏幕共享出来，给大家讲讲我做产品的思路，以及我们可能会怎么做一个东西——其实就是我们在 ChatPRD 真实的做法。这就是 ChatPRD，一个给产品经理用的 AI 副驾，帮你更高效、更高质量地写产品文档、做各种产品工作。它有个功能是这样的——

[2:15]

prd is that there's a chat and then there's a a document mode. And the way we tune and improve the core experience is very similar to a chat GPT. We have, you know, little user user ratings here. And so when somebody chats with chat PRD and has, you know, a good response either from a document or message perspective, they hit the thumbs up, thumbs down um button. We actually run a lot of AB tests and experiments on different models, different prompts, different configurations. It helps us figure out how to build these great experiences. And one of the things that I do is when you hit me with a a thumbs

ChatPRD 里既有聊天模式，也有文档模式。我们打磨和优化核心体验的方式，跟 ChatGPT 很像。你看这里有用户打分，当有人用 ChatPRD 聊天、得到了一个不错的回复（不管是文档形式还是消息形式），就可以点赞或点踩。我们其实跑了大量的 AB 测试和实验，试不同的模型、不同的 prompt、不同的配置，这帮我们摸清怎么把这些体验做好。我做的一件事就是，当你给我点了个踩的时候——

[2:53]

down um I ask specifically what we could improve. So these are some common common um you know feedback that we get from users. It's too long. It's too short. It's inaccurate. And so we built this modal every time you hit the thumbs down button to say what we can improve. Now I have a problem which is actually chatard gets a lot of thumbs up and I actually don't know what people like. So when you hit a heart or a thumbs up I just say thank you. I don't bother you and all I just all I know is that you liked it. I don't know what you liked about it or um how you know how you how you liked it.

我会专门问一句：我们哪里可以改进。这些是用户常给的反馈：太长了、太短了、不准确。所以我们做了这么个弹窗，每次你点踩，它就问我们能改进什么。但我现在有个问题：其实 ChatPRD 收到很多赞，可我根本不知道大家到底喜欢什么。所以当你点爱心或者点赞的时候，我只会说一句谢谢，不再打扰你。我能知道的就是你喜欢，但我不知道你具体喜欢哪一点、喜欢成什么样。

[3:30]

So I am going to actually build a PRD for you know thumbs up and then show you how we might flow that into a prototyping tool or to Devon to actually build this feature. So, what I Yeah, what I would typically do, you know, I'm a lazy PM. So, I would screenshot this and I would probably because I'm a PM, I'm always in Slack. So, I would go in Slack and I would actually ask chat purity in Slack because I don't like to change um models and I say, "Hey, help me build a PRD for giving uh sorry, receiving um positive feedback, comments, and reasons for when people like a chat PRD chat response. We

所以我现在就来给点赞这个功能做一份 PRD，然后给你们演示我们可能怎么把它接到一个原型工具里，或者接到 Devon 上去真正把这个功能做出来。我一般会这么干——我是个懒 PM，所以我会先把这个截个图。而且因为我是 PM，我天天泡在 Slack 里，所以我会在 Slack 里直接找 ChatPRD，因为我不想来回换模型。我会说：嘿，帮我做一份 PRD，是关于给——啊不对——是关于在大家喜欢某条 ChatPRD 回复时，接收正向反馈、评论和理由的。我们

[4:25]

do this for negative feedback, but I want to do it for positive feedback as well. Okay. So, I would like ask it's not super detailed, but I would ask chat pierd as sort of like my inslack um product manager to just write me PRD for this and it'll go off. It'll think through all it knows about chat pierd. it'll probably force me to be a better product manager here by saying, you know, Claire, this is just an idea. We don't have a why. We don't have a business reason. Why why should we build this? Um, so it's going to ask me, you know, here are some sections I might improve, context, what's the problem

对负向反馈是这么做的，但我想给正向反馈也做一套。我会这样去问它，问得不算特别详细，但我会把 ChatPRD 当成我泡在 Slack 里的产品经理，让它直接给我写份 PRD，它就开始干了。它会把它知道的关于 ChatPRD 的一切都过一遍，然后大概率还会逼着我做一个更好的产品经理，它会说：Claire，这只是个想法，你没说为什么要做，没有商业理由，我们到底为啥要做这个？所以它会反问我，还会给我几个可以补充的部分，比如背景、问题是什么——

[5:08]

statement, which is actually right. I have an imbalance in understanding user satisfaction, seamless integration into current workflow, etc., etc. I think this looks fine. We're a small team. I don't need to go into it too too much detail. So, I'm going to say, "This looks great. Write me a purity I can send to my engineering team." Spoiler alert, my engineering team is is is AI, too. Yeah. Um, and and this is one of the prompting tricks I I do in uh a lot of how I use AI is I give it the context of what kind of team member it's going to go to next, even if that team member is an AI. So, I often say like, "This is

问题陈述，其实它说得对：我在理解用户满意度这块有个失衡，还有要无缝融入现有工作流之类的等等。我觉得这看着挺好的。我们是个小团队，不用抠太细。所以我会说：这看着不错，帮我写一份 PRD，我好发给我的工程团队。剧透一下，我的工程团队其实也是 AI。对，这就是我用 AI 时常用的一个 prompt 小技巧：我会告诉它这份东西接下来要交给什么角色的团队成员，哪怕那个成员是个 AI。所以我经常会说，比如这份——

[5:47]

great. Write me a PRD." So I can send it to design. That's the PRDS that I would build to send something to like Vzero. Um sometimes if I'm going to send it straight to Devon, I'm going to say, "This looks great. Write me a PRD I can send to my engineering team." And what I found is that it will tune the content of the document to what they anticipate that sort of uh persona might. This episode is brought to you by Vanta.

很好，帮我写一份 PRD，我好发给设计——那我做出来的就是发给类似 v0 的那种 PRD。有时候如果我打算直接发给 Devon，我就会说：这看着不错，帮我写一份 PRD，我好发给我的工程团队。我发现它会根据它预判接收方那种角色的需求，去调整文档的内容。本期节目由 Vanta 赞助。

[6:14]

Whether you're a startup founder navigating your first audit or a seasoned security professional scaling your GRC program, proving your commitment to security has never been more critical or more complex. That's where Vanta comes in. Businesses use Vanta to build trust by automating compliance for indemand frameworks like SOCK 2, ISO 2701, HIPPA, and more. And with automation AI throughout the platform, you can proactively manage vendor risk and complete security questionnaires up to five times faster, getting valuable time back. Venta not only saves you time, it also saves you money. A new IDC white paper found that

不管你是第一次面对审计的创业公司创始人，还是在扩展 GRC 项目的资深安全专家，证明你对安全的投入从来没有像今天这样关键，也从来没有这么复杂过。这就是 Vanta 的用武之地。各家企业用 Vanta 来建立信任，它能自动化满足 SOC 2、ISO 27001、HIPAA 等热门合规框架的要求。借助平台里贯穿始终的自动化和 AI，你可以主动管理供应商风险，填安全问卷的速度最高能快五倍，把宝贵的时间省回来。Vanta 不光帮你省时间，还帮你省钱。IDC 一份新的白皮书发现，

[6:47]

Vanta customers achieve $535,000 per year in benefits and the platform pays for itself in just 3 months. Go to vanta.com/peter to meet with the Vant expert about your business needs. That's vanta.com/peter. Neat. So, it's going to go write the doc. It takes a little long to write the docs because we're now upgraded to GPT41 and it's just a real yapper. People love it. Product managers like to write but it takes a little bit longer time. So what I would do is I'm gonna wait for this to be written. It'll come back and link me to the doc. What I would decide is if I need to send this to a prototyping tool and because the you

Vanta 的客户每年能获得 53.5 万美元的收益，平台只需 3 个月就能回本。去 vanta.com/peter，和 Vanta 的专家聊聊你的业务需求，网址是 vanta.com/peter。好嘞。它现在要去写这份文档了。写文档要花点时间，因为我们现在升级到了 GPT-4.1，它特别能叨叨，大家就爱这点。产品经理都爱写东西，所以会多花点时间。我现在要做的就是等它写完，它会回来给我一个文档链接。然后我要决定的是，要不要把这个再发给原型工具。因为这个——

[7:27]

know the modal already exists in my codebase because I already have sort of a component that I can reuse. I'm probably going to skip the prototyping step because it's like an unnecessary step in the workflow. Instead, what I'm very likely to do is take this PRD, export it. I'll show you how I do that and then send it directly to Devon who can actually write write the code for me. Okay. Now, now it's writing the PRD here. Okay. So, you don't so you use Vzero for prototyping and you don't you don't actually use cursor or windsurf.

这个弹窗在我代码库里已经有了，我已经有一个能复用的组件，所以我大概率会跳过做原型这一步，因为它在这个工作流里是个多余的步骤。我更可能直接拿这份 PRD 导出来——我待会儿演示怎么导——然后直接发给 Devon，让它真正帮我把代码写出来。好，现在它正在写 PRD 呢。好，所以你做原型用的是 v0，你其实不用 Cursor 或者 Windsurf。

[8:00]

You just go to Devon for coding for something like this. Yeah. So don't don't tell my I love cursor. Um I've tried a lot of wind surf but actually since uh Devon has really improved over the past couple months and claude code came out I have dropped my cursor usage by like 70% or so. I still start fresh projects in cursor. sometimes things that I just want to explore or build myself or have an opinionated view on how to build. But I would say seven out of ten times I'm going to Devon to start a PR for me. Um it's it's for me it's less about the technical capability of the codegen and it's actually more about the user

像这种事你就直接交给 Devon 来写代码。对，不过别告诉别人——我其实很爱 Cursor。Windsurf 我也试过不少，但说实话，自从 Devon 这几个月进步特别大、再加上 Claude Code 出来之后，我用 Cursor 的频率大概降了 70% 左右。开新项目的时候我还是会从 Cursor 起步，有时候是我自己想随便探索一下、想自己动手做、或者对怎么做有很强的个人想法。但十次里大概有七次，我会让 Devon 帮我起一个 PR。对我来说，这更多不是因为代码生成的技术能力，反而更多是因为用户体验。

[8:49]

experience. So, you know this, I'm very busy. Lady's busy. And so, I can't be synchronously working in an IDE all the time. I'm actually not at my computer all the time. So, yesterday morning I went for a early morning walk. I'm not at my computer. I got like a ping from our support that there was a bug. I can't pull up an IDE on my phone. I can't log into C even though it was a simple bug to fix. And so I but I can't at mention Devon and say go fix this bug for me and it happens elsewhere. And so it's really less about the quality of the code although I think the quality code is sufficient for most of the use

你懂的，我很忙，是个大忙人。所以我没法一直同步地待在 IDE 里干活，我也不是一直都在电脑前。比如昨天早上我去晨走，人不在电脑边，结果我们客服那边给我发了条消息，说有个 bug。我手机上又打不开 IDE，也登不进去——哪怕那只是个很简单就能修的 bug。但我可以直接 @ Devon，跟它说去帮我把这个 bug 修了，它就在别的地方自己搞定了。所以这真的不是代码质量的问题——虽然我觉得对大多数场景来说代码质量是够用的。

[9:27]

cases. It really is about the synchronous asynchronous nature of how I can interact with the tool. Okay. So I had this lovely PRD. Um I'm sure it's very very compelling. Yes. Uh I want to quickly indicate I liked a response. I want to specify what I liked. I want to skip adding a comment if I'm busy, but still give feedback. That's pretty much that's pretty much it. So, what I would do here is actually export this as markdown. You know, RAI loves markdown. So, now I have a markdown file. And I would actually go into my very long Devon thread. You can see what I've been working on. And I would say, Devin,

它真正的好处在于我跟这个工具交互时那种同步对异步的差别。好，现在我有了这么一份漂亮的 PRD，我相信它一定写得特别有说服力。是的。我想让用户能快速表示我喜欢这条回复，并且能具体说出我喜欢哪点；如果我很忙，我想能跳过写评论这一步，但还是把反馈给出去。基本上就这些。所以我接下来要做的，就是把它导出成 markdown。你知道的，AI 特别爱 markdown。这样我就有了一个 markdown 文件。然后我会进到我那条很长很长的 Devon 对话里，你能看到我一直在忙活什么。我会说：Devon，

[10:09]

um, I want to capture positive feedback commentary the same way we capture negative feedback commentary. Um, I'm going to pull up I have cursor. So this is the actual code for the um the negative feedback. So I'll say we already have a component called shared feedback modal. We can use the same component for positive feedback as well. Love, like, and what else would I say here? You can make up the positive feedback categories. I don't want to think about that.

我想用跟我们采集负向反馈评论一模一样的方式，去采集正向反馈的评论。我现在打开了 Cursor，这就是负向反馈那块的实际代码。所以我会说：我们已经有一个组件叫 shared feedback modal，正向反馈也可以复用这同一个组件。爱心、点赞，还有……我这儿还该写点啥呢？正向反馈的分类你自己编就行，我不想费脑子去想这个。

[10:56]

um review the attached PRD for other context. So I would attach this PRD and send it off and then what Devon Yep. Go ahead. It has so so you can read that component in your code, right? Yeah. It has access to my repo. Okay. Okay. Yeah. Yeah. So, uh Devon's friendly friendly. He wakes up. Um, and then what's really interesting about the um, the interface here for the product designers out there is it takes time.

再看一下附带的那份 PRD，了解更多背景。所以我会把这份 PRD 附上、发出去，然后 Devon——你说，请讲。它能读你代码里那个组件，对吧？对，它能访问我的代码仓库。好的好的。对对。然后呢，Devon 很友好，它“醒”过来了。这个界面有意思的地方，给在场的产品设计师们说一下，就是它是要花时间的。

[11:31]

You saw my chat PRD. It t it takes a time to get a response and you know, you want to know that your message is acknowledged. You want to know it's thinking. You're kind it's it's kind of like waiting for a colleague where you're like, are are you there? Do you have five minutes? And so you'll see here that pretty quickly it will come back with a plan and say, "Hey, here's my plan for implementing this." And then um it'll ask if I want to wait to confirm the plan or if it can just yolo go straight out and then um once it it continues, it will go ahead and build and issue the PR. So we'll wait for that

你刚也看到我那个 ChatPRD 了，它出回复是要花时间的。这时候你会想确认一下你的消息被收到了，想知道它在思考。这有点像在等一个同事，你心里会犯嘀咕：你在吗？有空吗？能给我五分钟吗？所以你会看到，它很快就会带着一个方案回来，说：嘿，这是我实现这个功能的计划。然后它会问我，是想等着确认这个方案，还是它可以直接 yolo 一把往下冲。等它继续之后，它就会动手去构建并提交 PR。所以我们就等它

[12:07]

to happen. Wow. Okay, that's pretty cool. Let's see. and I'll pull up. There we go. Okay. So, we can see here Devon's working on a plan. I can say wait for my approval. I don't want to wait for approval. So, it can just go and it'll actually build the plan. And what it's doing behind the scenes is it's looking in the repo for the right um project. It's finding the code. It's designing a plan and then it's going to go ahead and make um make the experience. And what's really interesting here is I'll pull it over. I can also follow along here in the Devon console. So, it's making it's making this plan. You can see uh it

做完。哇，好的，这挺酷的。咱们看看，我把它调出来。来了。好，我们这儿能看到 Devon 正在拟一个方案。我可以让它等我审批，但我不想等审批，所以它就直接往下走、真的去把方案做出来。它在背后干的事是：在仓库里找到对的项目、定位代码、设计一个方案，然后接着去把这个体验做出来。这儿还有个挺有意思的地方——我把它拖过来——我还能在 Devon 的控制台里跟着看。所以它正在拟这个方案，你能看到它——

[13:03]

opened up the PRD and is looking through it. So, you can see it's working like an engineer would, right? It's like, okay, great. I'm going to look at the PRD and make sure I understand what it's achieving here. I'm going to go ahead and search the codebase for that um component that it referenced. All that kind of stuff, which I think is really um really neat. And but I prefer to stay in Slack because I'm product manager. I don't know. I'm in Slack. I'm on my phone all the time. So, all this stuff is happening behind the scenes while I'm just sitting here waiting for Slack to happen doing a fun podcast with with my

它把 PRD 打开了，正在通读。你看，它干活的方式就跟工程师一模一样，对吧？它会想：好，我先看一遍 PRD，确保我搞清楚它到底想实现什么；然后我去代码库里搜一下它提到的那个组件；诸如此类。我觉得这套流程特别巧妙。不过我自己更愿意待在 Slack 里，因为我是产品经理嘛。我也说不上来，反正我就泡在 Slack 里，手机也是不离手。所以这些活儿都在后台默默地跑，我就坐在这儿等 Slack 那边出结果，顺便跟我朋友录个开心的播客。

[13:38]

friend uh Peter. So, so it should come back and what it will say is basically I've done the work or I have a question and it'll issue a PR with the code change and then I'll show you how I kind of go through the the code change process. Do you actually review the PR or of course of course I review the PR. I mean we're you're required to review the PR and have an approval in our codebase. We don't deploy anything without review including AI stuff. So, we enforce approvals. So, it has to have an approval. Um, so once we get the PR, I'll show you how I do the code review.

我朋友 Peter。等它跑完回来，它基本上会告诉我两种结果：要么是「活儿干完了」，要么是「我有个问题」，然后它会带着代码改动提一个 PR，接着我就给你演示一下我是怎么走代码审查这套流程的。你真的会去审 PR 吗？当然会审了。我们公司的代码库是硬性要求的——PR 必须经过审查、必须有人 approve。我们任何东西不经过审查都不会上线，AI 写的也一样。所以我们强制要求 approval，它必须有人批过才行。等 PR 出来，我就给你演示我是怎么做 code review 的。

[14:15]

That's where where cursor maybe sometimes comes into play. Um, and then I ship it. And and let let me while we wait, let let me ask you like I I noticed your Slack channel is is only like a bunch of apps and bots. Like, do you have like real humans working for I do. I actually closed them out so you wouldn't be distracted by uh by the real humans. So, for chat PRD, I still by far ship the most code. If if you haven't guessed, we have um one and a half sort of like one full-time, one part-time engineer global that works with me on chat PRD. And then I have um a part-time growth person, although she's on uh

这时候 Cursor 有时候就派上用场了。然后我把它 ship 出去。趁我们等的工夫，我想问你一下——我注意到你的 Slack 频道里全是各种 app 和 bot，你手底下真有活人在干活吗？有的。我其实是把他们的频道关掉了，免得分你的心，那些都是真人。说到 ChatPRD，到现在为止写代码最多的还是我自己。如果你没猜到的话，跟我一起做 ChatPRD 的工程师全球加起来差不多一个半人——一个全职、一个兼职。另外我还有一个兼职做增长的，不过她现在正在……

[14:58]

leave right now. She just had a baby. And then I have a friend that's doing fractional enterprise sales for me. That's awesome. Yeah. Let let me ask you this. Like I think a lot of PMs want to, you know, build something on the side, have a side hustle. Like how do you find people these people to help help you? Oh, how do I find people to help me? Yeah. So, I actually found them all different ways, which is kind of fun.

休假，她刚生了娃。然后我还有个朋友，给我做兼职的企业销售。太棒了。是啊。我再问你一个问题——我觉得很多 PM 都想在工作之余搞点副业、搞个 side hustle。你是怎么找到这些人来帮你的？哦，怎么找人帮我啊？对。其实挺有意思，他们每个人都是通过不同的路子找到的。

[15:20]

So, the um engineers, I use a service top called Top Tall. I used Top Tall when I ran my last startup about 10 years ago. Had a really really great experience. So it allows you to hire engineering talent globally and the I would say from a product perspective I think their onboarding is pretty amazing. It actually I I um hired these engineers July 4th last year. Maybe you can empathize with this because it was like July 4th weekend. I was with my kids in Santa Cruz and I was like coding because we had this major bump in users and there was all this stuff I had to handle and it was a holiday and I needed

工程师方面，我用的是一个叫 Toptal 的服务。我大概十年前做上一家创业公司的时候就用过 Toptal，体验非常非常好。它能让你在全球范围内招到工程人才。从产品角度讲，我觉得他们的 onboarding 做得相当惊艳。我是去年七月四号招的这几个工程师，你可能能体会那种感觉——当时正好是七月四号那个长周末，我带着孩子在 Santa Cruz，结果我还在那儿写代码，因为我们用户量突然暴涨，一堆事得处理，偏偏又赶上假期，我还得……

[16:01]

to spend time with my kids. I was like what am I doing? This is bonkers. This product is profitable. Claire, you're allowed to hire people. And so I submitted sort of a request for engineers that were comfortable in our stack. same day had interviews and next day hired them and they've been working with with me for about a year. So I hired engineers through sort of a a talent connection marketplace. The growth person that I work with, Alisa, you may have seen her on our YouTube. Alisa made a YouTube video about chat PRD. Oh.

陪孩子。我当时就想：我这是在干嘛呢？这也太离谱了。这个产品明明是盈利的。Claire，你完全有资格雇人啊。于是我就提交了一个需求，想招几个熟悉我们技术栈的工程师，当天就安排了面试，第二天就把人招进来了，他们已经跟我一起干了差不多一年。所以工程师是通过一个人才对接平台招到的。跟我一起做增长的那个人叫 Alisa，你可能在我们的 YouTube 上见过她。Alisa 做了一个关于 ChatPRD 的 YouTube 视频。哦。

[16:37]

About a year ago and somebody sent it to me. It was like somebody made this amazing YouTube video about how great chat PRD is. And so I just emailed her and and thanked her for making the video. It was very very good and very nice. And she said, "Well, if there's anything you ever need, let me know." And I said, "Yes, can you teach me how to be a YouTube influencer and help me with some of my marketing?" And so she's been helping with our newsletters, our social and growth, and our our video marketing ever since. So that was sort of an organic user. And then uh my buddy Travis and I used to work together at

大概一年前的事，有人把那视频转给我看。说是有人做了个超棒的 YouTube 视频，讲 ChatPRD 有多好用。我就直接给她发了封邮件，谢谢她拍这个视频，拍得真的特别好、特别用心。她回复说：「以后你有任何需要,尽管开口。」我说：「行啊，那你能不能教我怎么当 YouTube 网红，顺便帮我搞搞营销？」从那以后她就一直帮我们做 newsletter、社媒和增长，还有视频营销。所以她算是个很自然冒出来的用户。还有就是我哥们儿 Travis，我俩以前一起在……

[17:12]

Optimizely. He ran sales. I ran product and I grabbed coffee with him and I said, "Dude, I think we can get this thing to like a million in ARR." Yeah. With just us. And he found that very interesting, an interesting problem to solve. And I said, "I have too much enterprise inbound to manage. I can't do it during the day. I have work to do." And so he helps manage um my my customer relationships, including enterprise prospects and onboardings and all that kind of stuff. Wow. Yeah. I I I love the Elisa's example in particular because like she just made a video and kind of like didn't ask for any new permission

Optimizely 共事过，他管销售，我管产品。我约他喝了杯咖啡，跟他说：「老兄，我觉得咱俩光靠自己就能把这玩意儿做到一百万 ARR。」对。就咱俩。他觉得这事儿特别有意思，是个值得攻克的难题。我跟他说：「我手上的企业级 inbound 多到管不过来，白天根本忙不过来，我还有正经工作要干。」所以他现在帮我打理客户关系，包括企业客户的潜在线索、onboarding 这些。哇。是啊。我特别喜欢 Alisa 这个例子，因为她就是单纯做了个视频，根本没问能不能……

[17:49]

and then you found her just through proof of work. I didn't ask for I didn't you know ask for permission asking her to do stuff and you know I get a lot of people reaching out and saying if there's anything I can do with chat purity I'd love to be involved. Two things that I think Alisa did differently is she gave me very specific things she could do for me. like it was very obvious the particular value she would bring to the product. Not like if you think of something let me know. It was like yeah this is what I can do for you. And two I only hire people that I'm going to pay. Like I'm not going to hire

也没事先征求许可，你完全是通过她的实际成果发现她的。她没问过我，也没说「我能帮你做点啥」之类的。其实很多人来找我，都说「ChatPRD 这边有什么我能帮上忙的，我特别想参与」。我觉得 Alisa 有两点跟别人不一样：第一，她给我的是非常具体的、她能帮我做的事，她能给产品带来什么价值一目了然。不是那种「你想到啥需要的就告诉我」，而是「就这些，是我能为你做的」。第二，我只雇我会付钱的人。我绝不会去招……

[18:22]

unpaid interns. I don't do volunteers. Like I only hire people that pay and so I want those people to be incredibly high leverage in the business and they have to be able to add something I don't have and I certainly think Alisa does that. Okay. Awesome. All right. Let's let's go back to the Slack. It looks like it did something. Yeah. So, it made this nice little plan here. Let's look at it. Um, so it's going Oh, it's like a 14step plan. Very good. I would have just yoloed in the code. Um, so it's going to create a new branch, update, positive feedback options. I like these.

无薪实习生，我也不用志愿者。我只雇那种我付钱的人，正因为如此，我希望这些人在公司里是极高杠杆的，他们必须能补上我没有的能力。我觉得 Alisa 完全做到了这点。好，太棒了。行，咱们回到 Slack 这边。看样子它好像干了点活儿。对。它做了一个挺漂亮的小计划，咱们来看看。哦，是个 14 步的计划。非常好。要是我自己估计就直接「梭哈」往代码里写了。它准备新建一个分支，更新正向反馈的选项。这些我喜欢。

[18:56]

It was clear, actionable, concise, relevant examples, good tone. So, it gives me a good sense of why somebody would like something. It's going to modify to have a positive or negative feedback type. That's smart. Extensible works for my use cases. Don't have to create a new component. And then it's going to update it uh to accept the rating parameter. Go make sure it's updated everywhere and uh test it and push a PR. So I think it's going to do that. In a couple minutes, we'll get hey Claire, you have a PR waiting notification.

「清晰、可操作、简洁、举例恰当、语气得当。」它能让我清楚地看出为什么有人会喜欢某个东西。它打算改成支持正向或负向两种反馈类型。这招聪明。可扩展，正好满足我的使用场景，不用再新建一个组件。然后它会更新代码让它接收 rating 参数，确保所有地方都改到位，再测试一下，然后推一个 PR。所以它应该会这么干。再过几分钟，我们就会收到「嘿 Claire，有个 PR 在等你」的通知。

[19:31]

Then then you can just test it test it locally and then you can merge it, right? And then I can just test it mo locally and merge it. I I do look at the code. So, one of the things that I definitely do is I pull up the PR locally, I run it locally, I test it locally, and I actually do look at the code to see if things make sense. I would say with Devon I get 80% hit rate on good quality nailed the product requirements and uh does it the way I would do it in terms of extensibility but occasionally uh it does something that is incorrect and so I sort of have to give it feedback in the PR or you

然后你就可以在本地测一测，没问题就 merge，对吧？对，我可以在本地测完再 merge。不过我确实会看代码。我一定会做的一件事就是把 PR 拉到本地、在本地跑起来、在本地测，而且我真的会去看代码，确认逻辑说得通。我用 Devon 的话，大概有 80% 的命中率——质量过关、精准命中产品需求，而且在可扩展性上也是按我自己会写的方式来做的。但偶尔它也会做出不对的东西，那我就得在 PR 里给它反馈；要么……

[20:11]

know probably one time out of 10 I just trash the branch and move on move on with my life. Oh and it's pushing it's pushing the PR right now I think. Okay. So, so maybe we'll check in in a couple minutes, see what come comes up with. Great. Um, so let me ask you while we're waiting. Um, so like you know the reality is like a lot of PMs still have to make a bunch of internal documents like strategy and and like this kind of stuff, right? Like they're not like as much as we all want to just prototype all day like that's not the reality of the job right right now. So I'm also curious how you like work on like

大概十次里有一次，我就直接把这个分支扔了,该干嘛干嘛去。哦，它正在推 PR，我觉得它现在正在推。好。那我们过几分钟再回来看看出了什么结果。很好。趁等的工夫我问你一下。你知道，现实情况是很多 PM 还是得写一大堆内部文档，比如战略啊这类东西，对吧？再怎么说我们都想整天做原型，但这并不是这份工作眼下的真实状态。所以我也挺好奇你是怎么做……

[20:44]

something you know like I I I don't know. I I feel like PMs think like pro strategy is like this like you know holier than thou thing like I have to go to a room and think about it for two hours. Yeah. But like that's not how you would do it with AI, right? Like how how do you do with AI? How do you come up with a product strategy or like a plan? Yeah. So I really think about product strategy as market strategy. At the end of the day, I'm trying to think about how we build something and solve customer problems in a way that has a unique market competiti like a unique market advantage that's margin accreative that

比如那种……我也说不好。我感觉 PM 们总觉得做战略是件特别「高大上」的事，好像我得进一个房间，对着它冥思苦想两个小时。是啊。但用 AI 的话就不是这么干了，对吧？你是怎么用 AI 的？你怎么用 AI 想出产品战略或者计划？嗯。我其实是把产品战略当成市场战略来看的。说到底，我想的是怎么打造一个东西、解决客户的问题，而且这种方式得有独特的市场竞争优势、能提升利润率、能在我们已有的平台上形成复利……

[21:22]

compounds on the platform that we've built and um our our team has the technical ability to execute and so I think product strategy is a set of actions you take given a unique understanding of your marketing customer. And I like it's really how I how I define product management. And so what I'm using AI for is not um necessarily to write down the things to do, but to make sure that I have a very strong understanding of what that market context is, what our um unique technical differentiators are, what might be coming. And so one of the additions to product strategy building that I found very useful is deep

而且我们团队还得有技术能力把它落地。所以我认为，产品战略就是基于你对市场和客户的独特理解所采取的一系列行动。这其实也是我对产品管理的定义。所以我用 AI，并不一定是用来列出该做哪些事，而是用来确保我对市场背景有非常扎实的理解——我们独特的技术差异化优势在哪里、未来可能会发生什么。在构建产品战略时，我发现有一个特别有用的补充，就是 deep……

[22:10]

research because it has access to such upto-date rich information that would be difficult to aggregate or or timeconuming to aggregate at an individual level really allows me to have a much more broad and deep view of the market. So a good example of this is at launch darkly. There's been a lot of market consolidation in the product um in in the feature experiment or feature management and experimentation space. I think we have a lot of adjacencies that we're going into. We've done two acquisitions just this year. We're expanding multi-product. And so as we expand our platform and portfolio, our

research（深度研究），因为它能拿到那么及时、那么丰富的信息，而这些信息靠个人去汇总要么很难、要么极其费时。它让我能对市场有一个广得多、也深得多的视角。举个好例子，就拿 LaunchDarkly 来说，在产品这块——具体是 feature management 和实验（experimentation）这个领域——市场上发生了大量整合。我觉得我们正往很多相邻的领域扩展，光今年我们就做了两笔收购，还在做多产品扩展。随着我们把平台和产品组合扩大，我们的……

[22:50]

um portfolio of competitors and partners also expand. And then there are a b bunch of macro um forces at play that could change how we might want to think about our product strategy. Yeah. And it's hard to keep track of every competitor, what they're building, what features, what those features might indicate about their overall strategy, who they're hiring, why they're hiring them, all that kind of stuff. It just it takes time and it's easy to have a very shallow view of the market when you have to do that research yourself. And so one of the things that I find myself doing quite frequently is really uh looking at

竞争对手和合作伙伴的版图也跟着变大了。再加上还有一堆宏观层面的力量在起作用，这些都可能改变我们思考产品战略的方式。是啊。要盯住每一个竞争对手实在太难了——他们在做什么、在做哪些功能、这些功能又透露出他们整体战略的什么信号、他们在招什么人、为什么招这些人，诸如此类。这太花时间了，而当你得自己去做这些调研时，很容易对市场只有一个非常浅的认识。所以我发现自己经常做的一件事，就是真正去……

[23:29]

the market and taking a step back and saying okay if we looked at these three competitors over the last three months what are they shipping? Who are they hiring? What does that indicate about their product strategy? How does that relate to our stated product strategy which is X, Y, and Z? And this is what we're building. what potential threats might emerge that we're not addressing in our strategy and where should we lean in harder that could create an edge for us and just being able to aggregate that information into a single place and really apply some critical thinking to that context allows I I think not

审视市场、退一步问自己：好，如果我们看这三个竞争对手过去三个月的动作——他们在 ship 什么？在招什么人？这透露出他们产品战略的什么信号？这跟我们已经定下的产品战略（也就是 X、Y、Z 这几条）有什么关系？而这是我们正在做的东西。有哪些潜在威胁可能冒出来、是我们战略里没顾及的？我们又该在哪些地方加码、好为自己创造一点优势？光是能把这些信息汇总到一处、再对这些背景认真地做点批判性思考，我觉得……

[24:11]

necessarily for more interesting product ideas any of that it actually allows you to identify like higher leverage points in your business that you should invest or places where um you might want to divest from because the market just isn't doesn't have the pull there. So you so you first like maybe use chatb deep research to research the competitors and what they're doing and then do you give it context of your like old product strategy or like context of Yeah. of launch darkly, right? Yeah.

倒不一定是为了想出更有意思的产品点子之类的，它真正的作用是帮你识别出业务里那些更高杠杆、值得你投入的发力点，或者那些你也许该撤出的地方——因为市场在那儿就是没有拉力。所以你是先用类似 ChatGPT 的 deep research 去调研竞争对手在干什么，然后再把你们旧的产品战略、或者 LaunchDarkly 的背景喂给它，对吧？对。

[24:39]

Yeah. Yeah. Yeah. Yeah. I I think the key thing here is like it's kind of like a collaboration with the AI, right? like like I don't I don't do well just like thinking in my room for two hours by by myself like I need to talk to some someone. Well, and I think it's one of the really it's one of the really interesting design benefits of these agentic experiences like what you see with Devon like what you saw with chat purity is especially in a distributed or remote world synchronous collaboration feels very expensive. you have to hop into a meeting, you have to make sure that everybody that you want to talk to is

对对对对对。我觉得这里的关键在于，它有点像是跟 AI 协作，对吧？我自己一个人关在房间里干想两个小时是行不通的，我得找个人聊一聊。而且我觉得这正是这些 agentic（智能体式）体验一个特别有意思的设计优势——就像你看到的 Devon、就像你看到的 ChatPRD 那样——尤其是在分布式或者远程办公的世界里，同步协作的成本变得非常高。你得跳进一个会议，还得确保你想聊的那些人都……

[25:16]

available and because you have this kind of limitations of space and time, you actually drag out collaboration and it takes longer to get stuff done because people just simply aren't available. That's right. And it so it makes, you know, great Rise of the Super IC, I can do it all myself, but it's very lonely, right? you're sitting there, you're like, I probably could write a pretty good product strategy by myself, but it's sure as heck not very fun to do that. And so even this kind of stuff where on the weekend, you know, the our our engineers are off on the weekend, but I'm working with something, but want

可用的时间有限，再加上有这种空间和时间上的限制，协作就会被拖得很长，事情也要花更久才能搞定，因为大家根本凑不到一起。没错。所以就出现了所谓的「超级 IC（个人贡献者）崛起」——什么都能我自己一个人干，但那其实挺孤独的，对吧？你坐在那儿，心想，我大概能自己写出一份还不错的产品策略，但这么干真的一点都不好玩。所以哪怕是这种场景，比如周末，我们的工程师周末是休息的，但我手头在搞一个东西，又想

[25:50]

somebody to collaborate with and partner with. Okay, well, at least I have good old Dev in here to hang with um and get some stuff done or give me feedback when I need it. And so I think that's a really um kind of interesting design um attribute of some of these agents that might make them, you know, might make organizations drive drive adoption. I mean, you probably talk to the stuff as much as you talk to your co-workers, right? Like Oh. Oh, I mean I talk to I talk to this stuff way more than I I talk to chatardy eco-workers. That's for sure. Yeah.

找个人一起协作、搭把手。那好吧，至少还有老伙计 Dev 在这儿陪我，能帮我搞定一些事，或者在我需要的时候给我反馈。所以我觉得这是这些 AI agent 一个挺有意思的设计点，可能正是它会推动组织去采用这些工具的原因。我是说，你跟这些工具说话的次数，大概不比跟同事说话少吧？哦，那当然了，我跟这些东西说话的次数，绝对比我跟同事说话多得多。这是肯定的。对。

[26:24]

Yeah. Um okay, I've got a PR. Look at this thing. Okay, so I have positive feedback C categories and support for love and like rating. So Devon has issued a PR here. Uh what I really like is the PR description is very nice. Um it's not a thing engineers love to write in a lot of detail, but it's very nice. It's very good how it's tested. Okay. So what I would typically do is um one I would um look at the code. So um I would say I'm just reading through. Okay, selected rating. If it's bad, it's a negative rating. Otherwise, if it's not bad, it's positive. That's correct. I would read through show. Oh, there's not. They got

好。嗯好，我这儿有个 PR，来看看这玩意儿。好，我做了正面反馈的分类，还支持了「喜欢」和「赞」的评分。Devon 在这儿提了个 PR。我特别喜欢的一点是它的 PR 描述写得非常好。工程师一般不爱把这种东西写得这么详细，但它写得很到位，连怎么测试的都讲得很清楚。好。那我通常会做的事，第一是看代码。我现在就读一遍。好，选中的评分，如果是 bad（差），就记为负面评分；否则，如果不是 bad，就是正面。没错。我会从头读一遍。哦，没了。他们把那个

[27:15]

rid of the the um the timeout, which is fine because we don't need to show the little confirmation message. It updated my analytics tracking, which is nice. I didn't explicitly ask it to do it, but it a you know, updated my analytics tracking to accommodate for the new thing. So, a good a good citizen. I don't know if you've had this experience, but engineers will sometimes write the code and then you have to fire like another Jira ticket for the tracking. The tracking is always like a afterthought. Yeah. is an afterthought.

那个超时（timeout）去掉了，这没问题，因为我们不需要再显示那个小小的确认提示了。它还更新了我的数据分析埋点，这点很不错。我没明确让它做，但它主动更新了埋点来适配这个新功能。真是个好公民。不知道你有没有遇到过这种情况——工程师有时候写完代码，你还得再开一张 Jira 工单去补埋点。埋点永远都是事后才想起来的东西。对，永远是马后炮。

[27:44]

So that's good. And then this feedback modal, what is it doing? Okay, so it's adding a new feedback type here. And then it's going to say what did you like versus what did you not like? It gave me the feedback options. Okay. So like code code generally looks pretty pretty good. So then I'm going to pull this branch locally. Let's just do this and local host. Let's see. It's always thrilling to do this to do this live. Okay, so I'm going to pull up my local host. I'm going to go to a chat and we're just going to smash that heart button and see what comes up.

所以这点很好。然后这个反馈弹窗，它在干嘛？好，它在这儿加了一种新的反馈类型，然后会问：你喜欢什么，不喜欢什么？它给了我这些反馈选项。好。所以代码整体看起来挺不错的。那我接下来要把这个分支拉到本地。我们就这么干，然后跑 local host。来看看。现场演示这种事永远让人心跳加速。好，我把本地环境打开，进到一个聊天里，咱们就使劲点那个爱心按钮，看看会冒出什么。

[28:36]

Okay, boom. Nice. It worked. It's there. clear explanation, actional advice, concise, helpful. I loved it. Submit feedback. So, so that worked. And you know, I mean, you and I had recorded a whole podcast while while doing this. So, then what I would do next is make sure my my checks have passed. They have. I would approve the PR and then it would be ready to merge. That's amazing. I I've been sleeping on Devon. I I could try it. You've been sleeping on Devon.

好，搞定。漂亮，成了，就在那儿。「解释清晰、建议可落地、简洁、有帮助。我很喜欢。」提交反馈。所以，这就跑通了。而且你知道吗，咱俩可是一边录完一整期播客一边把这个搞出来的。那接下来我会做的就是确认我的检查（checks）都通过了。确实都过了。我会批准这个 PR，然后它就可以合并了。太厉害了。我一直没把 Devon 当回事，我真该试试。你确实一直没把 Devon 当回事。

[29:12]

I'm like Devon's reply guy at this. But but you can see that. And what's nice for me just the shape of how I do things is I really can't sit and babysit an ID. Yeah. I just can't do it. And so, um, it's just when I have features well scoped, well understood, um, it gets it done. And I actually wouldn't be sitting here waiting for it to to work. I would be going off doing something else and then it would send me the slack that it's ready. So, uh the the sense of waiting and time is just um it's compressed. Yeah. I mean like I tried vibe cursor and wind surfing.

我现在简直是 Devon 的「头号粉丝」了。但你能看出来，对我来说很合我做事的风格——我真的没法坐那儿盯着一个 IDE 看。对。我就是做不到。所以呢，只要我把功能拆得足够清楚、想得足够明白，它就能把活干完。而且我其实不会坐这儿干等它跑，我会去忙别的事，等它好了就会给我发 Slack 通知。所以那种等待、那种时间的感觉，都被压缩掉了。对。我是说，我试过用 Cursor 那种 vibe 编程，还有 Windsurf。

[29:48]

Yeah. You have to sit there and like wait for it to respond and you know like it's like way way more step by step than than what we did here. Okay. So, let let's kind of like switch gears a little bit. Let's talk about um I'm I'm I'm curious uh how you balance everything. Like someone was trying to balance stuff. How how how do you balance the CPO stuff with like chat PRD and the new podcast? Is is it just like doing more things as async or people really don't like to hear it? I work all the time.

对。你得坐那儿等它一步步响应，而且它比咱们刚才这个流程要琐碎、要一步一步得多。好。那咱们换个话题聊聊吧。我挺好奇的，你是怎么平衡这一切的？就有人在问怎么平衡。你是怎么同时兼顾 CPO 的工作、ChatPRD 还有新的播客的？是不是就靠把更多事情异步化？还是说大家其实不爱听这个答案——我就是一直在工作。

[30:17]

Like I get up at today I got up like 5:30 in the morning. I do I check in on the support queue from Europe. I reply to a bunch of support stuff on my phone. Um, I send a note to the engineering team, uh, who's usually about halfway through their day, uh, early ear early morning on chat, just making sure that I'm not blocking anything and that I can do PR review if they need it. My kids get up at like 600 6:15. Um, my husband's amazing. He does lunches and breakfast. I do getting the kids dressed and drop off. One of us gets a workout in and then I'm straight into launch darkly work. So then I work

比如今天我大概五点半就起了。我会先看一眼来自欧洲那边的客服工单队列，在手机上回一堆客服问题。然后我给工程团队发条消息，他们那时候通常一天过了一半了，而我这边是大清早，在聊天里确认一下我没卡住任何人的进度，需要的话我可以做 PR review。我的孩子六点、六点一刻左右起床。我老公特别给力，他负责做午饭和早饭，我负责给孩子穿衣服、送他们上学。我们俩总有一个能抽空锻炼一下，然后我就直接进入 LaunchDarkly 的工作。接着我就开始干

[30:57]

a very serious job and spend a lot of time on lunch darkly. And then after school I I pick up the kids most most days. Sometimes my husband gets the kids. We come home, we do like homework. We all do our homework. So my kids do their homework and then I do my homework. Okay. Whether that's like uh launch darkly homework, podcast homework, chat PRD homework. Um we hang out and read books and do bath time and family time. um you know until 8 or nine. I watch like one and a half episodes of television and then I'm asleep between 9:30 and 9:45 to just like do it all again. And then I work on the weekends. Like I really do work on

一份非常正经的工作，在 LaunchDarkly 上花大量时间。然后放学后，大多数日子是我去接孩子，有时候是我老公去接。我们回到家，一起做「作业」——我们全家都做各自的作业。孩子做他们的作业，然后我做我的作业。好。不管那是 LaunchDarkly 的作业、播客的作业，还是 ChatPRD 的作业。然后我们一起待着、读书、洗澡、享受家庭时光，大概到八九点。我会看一集半左右的电视，然后在九点半到九点四十五之间睡着，第二天就再这么从头来一遍。然后周末我也工作。我是真的会在

[31:41]

the weekends. So I have two jobs on the weekends. One is carting my kids around to sports and hanging out with them and making sure that they have fun. And the other is I catch up on a lot of of chat purd stuff. And you know I do have help. So on chat PD I have these two engineers. So I've hired um because the business is profitable. I'm able to hire. So that's certainly helped. So I have a whole team that sort of works the business day for me. So I can step away from stuff um during during the business day. And then podcast is a new thing. So it's a little bit more time inensive.

周末工作。所以周末我有两份「工作」。一份是开车带孩子去各种体育活动、陪他们玩、确保他们玩得开心。另一份是我会集中补一大堆 ChatPRD 的事。而且我确实有帮手。ChatPRD 这边我请了两个工程师。我之所以能招人，是因为这门生意是盈利的，所以我招得起人。这当然帮了大忙。所以我有一整个团队在工作日帮我打理这门生意，这样我白天就能从这些事里抽身出来。至于播客，是个新东西，所以会更费时间一些。

[32:13]

Yeah. Right now. But I've got a bunch of stuff queued up and hopefully it'll become more steady state moving forward. But it's it's super hard. Um, I've got a great partner, so that helps. Um, I like to work, so that helps. When When are you gonna hire a husband to help help you with all this stuff? Are you gonna Oh, I mean, he's he's he's uh my my partner in all things. We have founded I don't know, six startups together. So, he's, you know, he's behind the sca for for Chat Purity. Got it. Taxes, legal, admin. Yeah. I'm I'm trying to get my wife to do all my finances and like you know do stuff.

对，现在是这样。不过我已经攒了一堆内容，希望往后能进入更稳定的状态。但这真的超级难。嗯，我有个超棒的伴侣，这帮了大忙。而且我喜欢工作，这也帮了忙。那你打算什么时候招个老公来帮你搞定这一摊子事啊？你会不会——哦，我是说，他就是我做一切事情的搭档。我们俩一起创办了，我都数不清了，大概六家创业公司。所以你知道，他在 ChatPRD 背后管着幕后的活儿。明白了。报税、法务、行政。对。我也正想让我老婆来帮我管所有财务啊、各种杂事。

[32:52]

Yeah. I mean I have a accountant in some in some sense um you know I have to invest in the business and invest in my own scalability and we have you know we have househel too uh like I have somebody that picks up the kids two times a week. Yeah. Uh but honestly the thing that I spend the most time on is is my kids when they're when when they can. I love dropping them off to school. We spend a lot of time together in the mornings. I try to pick them up from school every day. Um, yeah, we hang out together and so it's just you got a lot of hours in the week. I just don't do much of any anything else other than

对。我是说，某种意义上我有个会计。你知道，我得往生意里投钱，也得往自己的「可扩展性」上投钱，而且我们也请了家政帮手——比如我有人一周两次帮我接孩子。对。但说实话，我花时间最多的还是我的孩子，只要他们有空。我特别喜欢送他们上学，我们早上会在一起待很久。我尽量每天都去学校接他们。嗯，对，我们一起待着。所以说，一周里你确实有不少小时数可以用，我只是几乎不怎么花时间在别的事情上，除了

[33:32]

family and this and like exercise, you know. I mean, I interviewed uh Sahil who said like something very true which is like um like when like our most busy professional time is also when our kids need us the most. So you kind of have to remind yourself to take a step back sometimes. Yeah. You know, and I think for me, like I had working parents and every time uh I feel like I'm working a lot. One of the things to do is write down why it's great for my kids to see how much I love what I'm building and I built something myself and I'm having fun and I have interests and I vibe code with my seven-year-old. Um and my 5-year-old

家庭、这些工作，还有锻炼，你懂的。我之前采访过 Sahil，他说了一句特别真实的话——就是说，我们职业上最忙的那段时间，恰恰也是孩子最需要我们的时候。所以你有时候得提醒自己退一步、缓一缓。对。你知道吗，对我来说，我父母都是上班族，所以每次我觉得自己工作太多的时候，我会做的一件事就是写下来：为什么让孩子看到我有多热爱自己在做的事，是一件好事——我亲手做出了点东西，我乐在其中，我有自己的兴趣，我还会和我七岁的孩子一起 vibe coding。还有我五岁的孩子，

[34:13]

keeps me really humble. So, I think it's good to have a parent who really loves what they do um and inspires you to lean into your passions. And I hope, knock on wood, I mean, I ask my kids, I say, "I have too many jobs." And they all say, "But your number one job is being our mom." So, they they they know my priorities are straight and that's all that matters. Yeah. And and and like Yeah. I work the weekends, too. I don't I don't mind it because it's like you're building something you're building something great, right? Like something fun. So, Yeah. totally. Yeah. Um so then um let's talk about like product

总能让我特别保持谦卑。所以我觉得，有一个真正热爱自己工作、能激励你去拥抱自己热情的家长，是件好事。我也希望——但愿如此——我是说，我会问我的孩子，我说：「我的工作太多了。」他们都会说：「但你的头号工作是当我们的妈妈。」所以他们知道我的优先级摆得很正，这就足够了。对。还有还有，就像，对，我周末也工作。我一点都不介意，因为这就好比你在打造一个东西，你在打造一个很棒的东西，对吧？一个有趣的东西。所以，对。完全同意。对。那么接下来咱们聊聊产品

[34:46]

management a little bit. You know I I think you had a famous presentation saying product management is dead. I I I I think it's more like traditional product management is dead or or is dying, right? Is that fair to say? Maybe. I mean let's see. Let's play it. Let's play it forward and see. Um it certainly hasn't died yet in terms of like I'm still hiring a product manager. Great. At Launch Darkly we just hired a bunch of product managers. I just think the role, the expectations are really going to change in the role. Um, I think the tools and the hard skills you're going to need are going to change in the

管理这块吧。你知道，我记得你做过一个很出名的演讲，说「产品管理已死」。我觉得更准确的说法应该是，传统的产品管理已死，或者说正在消亡，对吧？这样说公平吗？也许吧。我是说，咱们走着瞧。让我们往后推演一下，看看会怎样。嗯，至少就目前而言它肯定还没死——因为我还在招产品经理啊。很好。在 LaunchDarkly 我们刚招了一批产品经理。我只是觉得这个角色、对它的预期会发生很大变化。嗯，我觉得你需要的工具和硬技能也会在这个角色里发生变化。

[35:21]

role. And I know lots of startups that have gone many, many years before hiring their first PM and kind of don't want to hire many. And so, yeah, you know, it's it's hard to say how existing companies and teams will shift in the short and midterm. I'm actually really curious about how new teams start to form and what the sort of like new persona of a building team will look like and whether or not what we call a product manager has any role on that team. I mean, it's happening already, right? Like they're like I I think the designer is becoming like a design engineer. Totally. And then maybe there's like someone who's

我还知道很多创业公司熬了好多好多年才招第一个 PM，而且也不太想多招。所以，对，现有的公司和团队在短期、中期会怎么转变，还真不好说。我其实特别好奇的是，新团队会怎么开始组建，未来一个「构建型团队」的新画像会长什么样，以及我们所说的「产品经理」在那种团队里到底还有没有位置。我是说，这已经在发生了，对吧？比如设计师正在变成「设计工程师」。完全是这样。然后也许会有一个

[36:04]

like more backend focused who knows product too. Maybe that I mean at launch darkly we have this product called AI configs and lucky lucky team. I love this product so much that I'm PMing it right now and so I'm sort of the named product lead for this product. Guess what? I have an executive job. I have lots of other things to work on. So they get like a partial product manager. And what I'm seeing is our design lead, who's amazing, is writing PRDs. Our engineering manager, who's amazing, is writing PRDs. They're coming to me 80% formed. And I can say, "Oh, actually, I think this needs to tweak or this is how

更偏后端、但同时也懂产品的人。也许……我是说在 LaunchDarkly 我们有个产品叫 AI configs，团队特别棒。我太喜欢这个产品了，以至于我现在亲自在给它做 PM，算是这个产品挂名的产品负责人。结果呢？我有一份高管的工作，还有一大堆别的事要做。所以他们其实只能拿到一个「半个产品经理」。而我观察到的是，我们那位超级棒的设计负责人在写 PRD，我们那位超级棒的工程经理也在写 PRD。他们拿来给我的时候已经成型了 80%，我就能说：「哦，其实我觉得这里得调一下」，或者「这是

[36:43]

I measure it for the business or let's not forget this part, but they don't need a full-time me to make that that piece happen." Um, at least in in the short term. And so I do think people with the right hard skills and the right tools can step into step into the gap and and make something happen for a team that's, you know, flexible and adaptable. Yeah. And I think maybe the optimistic view of this thing is like the job will hopefully become more fun.

我从业务角度衡量它的方式」，或者「咱们别忘了这一部分」——但他们并不需要一个全职的我来推动那一块落地。至少短期内是这样。所以我确实认为，拥有合适硬技能和合适工具的人，能够补上这个空缺，为一个灵活、能适应变化的团队把事情真正做成。对。我觉得对这件事更乐观的看法是，这份工作有希望变得更好玩。

[37:09]

Like there's like a lot of parts of the job that's like not fun, right? Like like um I think so. I don't know. Going through a lot of re reviews or or like you know just having a lot of meetings all day. I totally think so. And and what's so funny to me is I don't say product management is dead because of doom and gloom. I say it's dead because one I think you're going to be in a world of pain if you're not prepared for a shift in the next 18 months. Like I just think not being prepared is the worst thing you can do for your career. And so if I can be a little bit incendiary on how I get you to prepare, I will I feel like I've done

这份工作里有很多部分其实并不好玩，对吧？比如——嗯我是这么觉得的，我也说不好——比如要过一堆评审，或者一整天开一堆会。我完全同意。让我觉得特别有意思的是，我说「产品管理已死」并不是出于什么悲观末日论。我说它「死了」，一是因为我觉得，如果你没为接下来 18 个月里的这场变革做好准备，你会非常痛苦。我就是觉得，没做好准备是你能对自己职业生涯做的最糟糕的事。所以如果用一点点「煽动性」的说法能让你动起来去准备，那我就用——我感觉我已经做到了

[37:44]

some good for folks careers. Yeah. And two, a lot of the job is not fun. Like a lot of the job is tedious. A lot of the job is not creative. A lot of the job is very disconnected from customer impact. And so if you can compress the parts of the job that are not high lever, not interesting, you can spend a lot more time in the space I like to spend, which is in being creative, in being generative, in building and collaborating with a team, in talking to customers. Those are the in competing in the market. Those are the things that I think are fun about the role. Yeah. And I just want to spend more time on them.

对大家的职业发展有好处。对，第二点是，这份工作里有很多事其实并不有趣，很多都是琐碎的活儿，没什么创造性，跟客户的实际影响也离得很远。所以如果你能把那些不高杠杆、不有意思的部分压缩掉，你就能把更多时间花在我喜欢的地方——也就是搞创意、做产出、动手去做、跟团队协作、跟客户聊天，还有在市场上竞争。我觉得这些才是这个角色里好玩的事，我想多花点时间在这些上面。

[38:25]

Yeah. Same same here. I think every PM wants to spend more time on this stuff. Yeah. Yeah. Totally. Real quick, I'm I'm curious uh at launch darkly, have you compressed the planning cycle of of of the pro road map or like you know because a lot of times what takes away from this customer iteration stuff is like you know making OKRs and like planning you know me and my OKR memes. Um yeah, which one of us has not been victimized by by an OKR? So this is how we do product planning. We do heavyish product planning about every six months. And so we set aside about three days to do product planning. It's really product

对，我也一样。我觉得每个 PM 都想多花点时间在这些事上。对对，完全是。顺便问一下，我挺好奇——在 LaunchDarkly，你们有没有把产品路线图的规划周期压缩一下？因为很多时候真正占用客户迭代时间的，恰恰是定 OKR、做规划这些东西，就跟我那些 OKR 段子说的一样。对啊，咱们当中谁还没被 OKR 折磨过呢？所以我们是这么做产品规划的：我们大概每六个月做一次比较重的产品规划，会专门留出三天来做这件事。它其实更偏产品……

[39:04]

strategy and it starts with what are our business goals for the next six months. And so it starts with what are we going to sell? How are we going to sell it? How is this going to acrue value to the business? What are our actual goals? Do we agree those goals are achievable and important? We spend a lot of time with that. And then we go into product strategy. And we define product strategy as a combination of what and why we're going to build things and how we're going to commercialize those and bring them to market. So it is a much more comprehensive view of planning and then we spend probably like less than 10% of

……战略，而且一上来就是问：未来这六个月我们的业务目标是什么？所以会从这些问起——我们要卖什么？怎么卖？这件事怎么给公司带来价值？我们真正的目标是什么？我们是否一致认为这些目标既可实现又重要？我们会在这上面花很多时间。然后才进入产品战略。我们对产品战略的定义是这么一个组合：我们要做什么、为什么做，以及怎么把它商业化、推向市场。所以这是一个对规划更全面的视角。之后我们大概只花不到 10% 的……

[39:42]

the time on the actual roadmap items. like the roadmap items cascade down and do we believe that they're the right ones but really we spend the time on the product and commercial strategic planning about every six months and then between those times it really is I just think an always on roadmap is the best way to go. I know where to go inside the company to find everything every single team's working on. And we know from a technology perspective and from a business perspective what are like the five most important things to achieve in any quarter and we look at those on a regular basis. So we try to keep it

……时间在具体的路线图条目上。路线图条目就是顺着层层往下展开，我们再看一眼这些是不是对的事。但真正花时间的，是每六个月做一次的产品和商业战略规划。而在这两次规划之间，我真心觉得保持一份「永远在线」的路线图是最好的方式。公司内部要找任何信息我都知道去哪儿找，每个团队在做什么我都清楚。我们从技术和业务两个角度都很明确——任何一个季度里最重要的五件事是什么，我们会定期去看这些。所以我们尽量把它保持得……

[40:18]

really pretty simple. Um, I have I I along with the rest of my leadership team fight the good fight on like dumping a bunch of time into planning meetings and actually spend more time on tracking are we getting stuff done and being more reactive in the moment when we need to address something. Makes sense. It sounds like you have the planning pretty compress compressed so it's like intensive three days but like it doesn't drag on too far beyond that. It probably takes teams you know two weeks to prep for it. They're not prepping for it full-time, but they do have to get get together with the team, make sure technical initiatives

……相当简单。我和我整个领导团队一起在打一场「正义之战」，就是抵制往规划会议里砸一大堆时间，而是把更多时间花在跟踪「事情有没有真的做完」上，需要处理什么的时候就当下更灵活地去应对。有道理。听起来你们的规划压缩得挺到位——就是那种三天的高强度集中，但不会拖很久。大概每个团队要花两周来准备，但他们不是全职在准备，只是确实得跟团队凑到一起，把技术上的事项……

[40:52]

are accounted for, all that kind of stuff. Okay, great. So, so when I hear um you got to get ready for like this new new era, like a lot of the advice is just like, hey, you got to go play with these AI tools and you got to get your hands dirty, right? But but like and I think that's good advice, but like how do you go beyond that? Like um like you know, when you evaluate PMs, this AI stuff, how do you what skills do you look for? Yeah, I mean I just think again I think it's pretty shortsighted to say vibe coding will be the differentiator or like prototyping skills will be the differentiator um

……都梳理清楚，诸如此类。好的，太好了。那我听到的是——你得为这个新时代做好准备。很多建议无非就是「嘿，你得去玩玩这些 AI 工具，得亲自上手」，对吧？我觉得这建议是好的，但怎么才能更进一步呢？比如你在评估 PM 的时候，对 AI 这块，你会看重哪些技能？嗯，我还是那句话，我觉得说「vibe coding 会成为胜负手」或者「做原型的能力会成为胜负手」是相当短视的，因为……

[41:26]

because that only takes you so far. So what I really think about is how to cultivate attributes that futurep proof your career in this world of AI, not AI hard skills. And so the attributes I think about cultivating are scrainess, how do I get something done with very little? If I need to solve a problem and I have zero resources, how do I get that done? Can I prove I can get things done in a very lean environment? Perhaps using a set of tools, but perhaps not. I think scrappiness is one one piece that I assess for. The second thing I assess for is the ability to learn. So again, you could show that you're capable of

……那些只能带你走这么远。所以我真正在想的是，怎么去培养那些能让你的职业在这个 AI 时代「面向未来」的素质，而不是 AI 的硬技能。我想培养的素质，第一个是「scrappy」——也就是用极少的资源把事情搞定的能力。如果我得解决一个问题，手里又一点资源都没有，我怎么把它搞定？我能不能证明自己能在一个非常精简的环境里把事情做成？也许会借助一套工具，也许不用。我觉得这种「不挑条件、硬是把事干成」的劲头是我会考察的一点。第二点我会考察的是学习能力。同样，你可以证明你有能力……

[42:08]

learning tools or capable of learning new skills. That's one way to demonstrate learning. You can also demonstrate that you're capable of learning the market or capable of learning finance, any of those things. But I really think that ability to learn at this moment is a really important attribute for um for getting stuff done. And then honestly I think the ability to clearly communicate is so critical right now. And this is where I think maybe like classic product managers have have the edge. Um us us yappers h have the edge which is if you look at something like Devon it is a language-based

……学会某个工具、学会新技能，这是展示学习能力的一种方式。你也可以展示你有能力去搞懂市场、搞懂财务，任何这类东西都行。但我真心觉得，在当下这个时刻，学习能力是把事情做成的一个非常重要的素质。然后老实说，我觉得现在「清晰沟通」的能力极其关键。这可能正是那些传统产品经理有优势的地方——我们这些「话痨」反而占了便宜。因为你看像 Devon 这样的东西，它就是一个基于语言的……

[42:49]

interface. You are telling a system what to do. your ability to clearly articulate what to do in a way that a system can appropriately interpret it, access its set of tools and get the job done correctly is actually a really important skill. And so you can call that prompting, you can call that clear communication, but the ability to clearly uh articulate a set of outcomes and get a system whether that system is human or otherwise to do things I think is very very very important. So those are some of the things that I look like from an attribute perspective really like scrappiness, ability to learn and

……交互界面。你是在告诉一个系统去做什么。你能不能把要做的事讲清楚，讲到这个系统能正确理解、能调用它那套工具、把活儿正确干完——这其实是个非常重要的技能。你可以管它叫 prompt，也可以管它叫清晰沟通，但那种能把一组想要的结果清楚地表达出来、并让一个系统（不管这个系统是人还是别的什么）去把事做成的能力，我觉得真的非常非常重要。所以从素质这个角度，我看重的一些东西就是：scrappy 的劲头、学习能力，还有……

[43:27]

ability to communicate. You can demonstrate those through through these skills, but if you have those kind of attributes, you'll be successful in kind of any role. Yeah, that's a really good point. I I think um every time you're prompting, you're basically improving your written communication skills. Yeah, totally. You're kind of for force, too. Yeah. Um Okay. And uh do do you like uh evaluate like in like for promotions and stuff do you evaluate PM separated from design or like you know you talk about this AI power triple threat. Do you try to like people who can blend roles together? Is that something you try to

……沟通能力。你可以通过这些具体技能把它们展示出来，但只要你具备这类素质，你在几乎任何角色里都会做得很好。对，这点说得真好。我觉得每次你在写 prompt 的时候，其实都是在锻炼你的书面表达能力。对，完全是。也算是一种倒逼吧。对。好。那你在做晋升评估这类事的时候，会把 PM 和设计分开来评吗？还是说——你之前讲过那个「AI 加持的三料全能」，你会不会去……

[43:59]

encourage? Yeah. Um so we try to encourage it from a cultural perspective. So maybe the fourth attribute I would throw on that list is low ego. Like you're just not going to survive in new orgs if you get territorial. Like what a what a waste of of effort and potential outcome for people to block progress based on being territorial. So you know a designer that says no one but me can design this or an engineer that says no one but an engineer can push code. I just think that mindset is very limiting in the face of what is pretty radical change.

……鼓励那种能融合多个角色的人？会的。我们会从文化层面去鼓励这件事。所以我想往那个清单上再加的第四个素质就是「低自我」（low ego）。如果你很有地盘意识，在这些新型组织里你是活不下去的。为了护着自己那一亩三分地而去挡别人的进展，对个人也好、对最终成果也好，都是极大的浪费。比如一个设计师说「除了我谁都不能设计这个」，或者一个工程师说「除了工程师谁都不能提交代码」——面对这种相当激进的变革，我觉得这种心态是非常局限的。

[44:32]

And so what we look like look for in promotions is one I think you do need AI fluency and this is probably even more impactful in the engineering track. I run the engineering organ as well organization as well. Um fluency with these tools matters. You need to figure out how to use AI powered idees. You need to figure out how it's appropriate to use something like Devon inside your organization. You need to have an opinion on what models work for what.

所以我们在晋升里看重的：第一，你确实需要有 AI 的熟练度，而这一点在工程线上影响可能更大。我同时也管着工程组织。对这些工具的熟练度很重要。你得搞明白怎么用 AI 加持的 IDE，得搞明白在你们组织内部怎样用 Devon 这类东西才合适，还得对「什么模型适合干什么」有自己的判断。

[45:00]

you need to have the ability to architect uh a repo to be easier to use with these tools. All those things are hard skills that should be part of a promotion ladder or career path for software engineers and you can you can imagine the parallel on the product management or design side. So I think there's like tool fluency is one piece that's pretty important and then it's just leverage right as you become more and more senior in a product role. the expectation you have a higher impact on the business and my expectation is that you use all the tools available to you to have that impact. And so we certainly

你得有能力把一个代码仓库架构成更容易配合这些工具来用的样子。所有这些都属于硬技能，应该成为软件工程师晋升阶梯或职业路径的一部分；你也可以想象在产品管理或设计那一侧有完全对应的东西。所以我觉得工具熟练度是挺重要的一块。然后另一块就是杠杆——对吧，随着你在产品岗位上越来越资深，大家对你的期望就是你对业务有更高的影响力，而我的期望是，你会动用一切可用的工具来产生这种影响。所以我们当然……

[45:39]

do look at leverage um especially in senior promotions of this individual is having a high impact on our business and they could do that a lot of different ways but often they're they're they're leveraging all the tools available to them. Yeah. It's kind of like a no-brainer like you basically have like a AI team help you. Why why would you not use this stuff? Yeah, totally. I tell I tell all my engineering managers when they use when they uh you know use Devon, I say, "Congratulations, I just gave you 10 interns." Like, no one no you know no engineering manager turns their nose up at headcount. So I'm

……会去看「杠杆」这件事，尤其是在资深岗位的晋升上——这个人是不是对我们的业务产生了很高的影响。他们可以用很多不同的方式做到，但通常都是因为他们把一切可用的工具都用上了。对。这基本上是个不用想的事——你等于有一支 AI 团队帮你干活，你凭什么不用呢？对，完全是。我跟我所有的工程经理都这么说：当他们用 Devon 的时候，我会说「恭喜，我刚给了你 10 个实习生」。没有哪个工程经理会对增加人头嗤之以鼻。所以我就……

[46:14]

trying to position it like congratulations you got overnight headcount. Is that your response when they ask for more like human headcount? You're like okay I'm gonna give you that. So I I I do when people ask for headcount I say is your team maximally efficient just like you know, is your team maximally efficient? Yeah. Um, and have they experimented to know where they could get efficiencies? A lot of times the answer is yes, and I have a good sense of that. And sometimes the answer is no. And then they need to spend some time getting there. Okay. So, um, last question and related to what we talked about like how do you drive uh,

……试着把它包装成「恭喜你一夜之间多了几个人头」。那当他们来跟你要真人的人头时，你也是这么回的吗？你会说「行，我给你」？我确实会这样——当有人来要人头时，我会反问：你的团队效率拉满了吗？就是那句，你的团队效率真的拉满了吗？对。他们有没有做过实验，知道自己在哪些地方还能提效？很多时候答案是「有，而且我心里很有数」。有时候答案是「没有」，那他们就得花点时间先把这块搞清楚。好。那么，最后一个问题，跟我们刚聊的也有关——你是怎么推动……

[46:48]

organizational adoption of AI? Like a lot of this stuff is just limited based on like security or like you know people are like like I a lot of companies you can't even use all the tools that you want to. I know it's so sad. Um, so I think a couple things you need to do. You need to figure out the finance and security framework for your company to enable experimentation. So luckily we're relatively small company with a you know good mature security team and program and a good mature uh finance program. Finance is not a problem. I can make the uh case for high ROI of these tools all day and all

……整个组织去采用 AI 的？这方面很多事都卡在安全合规上，或者就是——你知道的，很多公司里你连那些你想用的工具都用不了。我知道，太让人难受了。所以我觉得有几件事是你必须做的。你得为你公司搞清楚那套财务和安全的框架，好让大家能去做实验。我们算是运气好，相对来说是家小公司，有一个比较成熟的安全团队和安全项目，也有一个比较成熟的财务体系。财务不是问题，我可以白天黑夜地为这些工具的高 ROI 来论证……

[47:25]

night. ROI of the the spend is non-issue and then infosc we just had to build a framework by which they could quickly evaluate thumbs up and thumbs down use of tools in a particular context to make sure that we were meeting our requirements and obligations to our customers as well as internal security principles. Once we had that framework we could move very fast in terms of onboarding tools. The other thing that you have to do from a cultural perspective is normalize AI tool use and give the happy path to adopting it because otherwise from a security perspective you're stuck with a bunch of shadow IT where people like hide the

……这笔花费的 ROI 根本不是问题。然后在信息安全这块，我们只需要搭一套框架，让他们能在某个具体场景下快速地对一个工具的使用「点赞」或「点踩」，以确保我们既满足了对客户的要求和义务，也守住了内部的安全原则。一旦有了这套框架，我们在工具的接入上就能跑得非常快。另一件你必须从文化层面去做的事，是把使用 AI 工具这件事「常态化」，给大家一条顺畅的采用路径；否则从安全角度看，你就会被一堆「影子 IT」困住——人们会偷偷……

[48:03]

fact that they're using AI tools. They do it on like personal accounts. Yeah. They do it on uncontrolled licenses and that is actually the worst situation from a security perspective. So you kind of have to accept like people are going to do this so let's give them the clean path to do it and let's put it out in the open versus not. So that's sort of the admin side of things. I think the real uh needle mover however is cultural and so what we've done is we have at launch darkly a slack channel called project building with AI and we throw you have to request AI tool access in that channel. So everybody who has AI

……瞒着用 AI 工具。他们会用个人账号来用，对。他们会在不受管控的授权许可下用，而那其实是从安全角度看最糟糕的情况。所以你某种程度上得接受「人们就是会去用这个」，那不如给他们一条干净的路径去用，把它摆到台面上，而不是藏着掖着。这算是管理这一侧的事。不过我觉得真正起作用的「胜负手」是文化层面的。所以我们在 LaunchDarkly 做的是，建了一个 Slack 频道，叫「用 AI 搞建设」（project building with AI）。你要申请 AI 工具权限，就得在那个频道里申请。所以每个拿到 AI……

[48:40]

tool access is in the same channel which is good. Okay. And then everybody shows off stuff that worked, stuff that didn't, stuff they tried, stuff they've learned in one single channel. It creates a lot of momentum. And so people will post their cursor wins or people will co post their dev and fails or they'll post new prototyping tools. And so it gives it gives socialization to a lot of use cases. And then in technology we've named a single engineer Zach. He's amazing as sort of like the responsible party for AI engineering adoption. And so what Zach does is he goes and evaluates tools for us. He makes sure

……工具权限的人都在同一个频道里，这一点很好。然后大家都在这一个频道里展示什么管用、什么不管用、自己试了什么、学到了什么。这制造出了很强的势头。所以有人会发他们用 Cursor 的成功案例，有人会发用 Devon 翻车的经历，或者发现的新原型工具。这就把大量的用例「社交化」了。然后在技术这边，我们指定了一位工程师，叫 Zach——他特别厉害——让他当那个负责推动 AI 在工程上落地的「负责人」。Zach 干的事就是去帮我们评估各种工具。他会确保……

[49:16]

that those tools are set up with our codebase and our team to be built into our systems and processes, you know, making recommendations on, for example, cursor rules or Devon environment set up. And then he runs, which he's running right now, I'm going to do it after this podcast. He runs a Friday AI power hour where it's like a Twitch stream of of Zach using AI. So, um he'll sit, he'll show off some stuff, he'll invite others to show off stuff and um we socialize it live as well. And people show up for this. Oh, I had like 60 70 people every week. Wow. Like it's fun. Yeah, it's really fun. Yeah. Um, any any uh closing

……那些工具能跟我们的代码库、我们的团队对接好，被内建进我们的系统和流程里，比如就 Cursor 的规则（rules）或者 Devon 的环境配置给出建议。然后他还会跑——他现在正在跑——这个播客录完我也要去——他会主持一个周五的「AI power hour」，有点像 Zach 用 AI 的一个 Twitch 直播。他会坐那儿，展示一些东西，也邀请别人来展示，我们就这么实时地把它社交化。而且大家真的会来看。哦，我每周大概有 60、70 个人来。哇。挺好玩的。对，真的挺有意思。对。那，还有没有什么收尾的……

[49:58]

words of advice for for folks or any hot like you're you're you're the queen of hot takes. Give me some give me some closing hot takes. Yeah, I think this is the year of the agent. I know it's been said, but I think this is the year where you go from sort of like co-pilot singleplayer experiences in AI to multiplayer agentic experiences that feel more like your colleague. Again, get good at it. Like, get good at working with agents. Figure out which ones are out there that add value.

给大家一些建议，或者说点犀利的观点——你可是 hot take 女王。给我来几个收尾的犀利观点吧。好啊。我觉得今年就是 AI agent 元年。我知道这话已经被说烂了，但我真觉得今年是个转折点：AI 从那种 co-pilot 式的单人体验，转向了更像有同事在身边的多人协作式 agent 体验。所以还是那句话，把这件事玩熟。学会怎么跟 AI agent 协作，搞清楚现在市面上哪些 agent 是真正有价值的。

[50:32]

Figure out how you prompt them correctly. Figure out how you orchestrate them together. I think that's a hard skill that's going to really be um important in the next year. And so, um, I think I think this is the year where you get out of this like single player mode and you start getting your your team of AI AI interns. So, so you mean let's let's define what agent is real quick. Like agent is like someone who goes off and does work for you and collaborate with other AI, right? Is that what it is? Yeah. Yep.

搞清楚怎么把 prompt 写对，搞清楚怎么把多个 agent 编排到一起协同工作。我觉得这是一项很难但接下来这一年会变得特别重要的技能。所以我觉得今年就是你从单人模式里走出来、开始拥有一支「AI 实习生团队」的一年。那我们先快速定义一下什么叫 agent 吧。agent 就是那种会自己跑去帮你干活、还能跟其他 AI 协作的东西，对吧？是这个意思吗？对，没错。

[51:02]

Yeah, it makes sense. Yeah. Just just like what you demonstrated with Devon. Like you don't have to babysit it with prompts all the time. They can just get it to do do stuff. Yep. Cool. All right. Well, um, let's talk about where can people find How I AI podcast. So, you can find How I AI. I recommend Spotify or YouTube because our guests get on and screen share their AI use cases. It's also available wherever you want uh to get your podcasts, although I recommend the video aspects.

嗯，说得通。对。就像你刚才用 Devin 演示的那样，你不用一直用 prompt 盯着它、给它喂指令，它自己就能把活儿干了。没错。挺酷的。好，那我们聊聊大家在哪儿能找到《How I AI》这个播客吧。你可以这样找到《How I AI》：我推荐 Spotify 或者 YouTube，因为我们的嘉宾会上来分享屏幕、演示他们的 AI 用例。当然你常用的各大播客平台上也都有，不过我还是更推荐看视频版。

[51:32]

And then you can find me on X at Clarvo or I might revive the Tik Tok. So, the chief product officer Tik Tok might be coming back. Yeah, Tik Tok is perfect for your hot takes. You should definitely revive it. I just I can't take like a fifth or sixth job and Tik Tok feels like a fifth or sixth job. Yeah. Yeah. True, true. Yeah. Cool. Well, thanks so much, Claire. This is another amazing conversation. I definitely learned a lot. Oh, I really appreciate it. It was super fun. Yeah.

另外你可以在 X 上找到我，账号是 @Clarvo，我也可能会重新开始玩 TikTok。所以「首席产品官的 TikTok」说不定要回归了。是啊，TikTok 太适合你发犀利观点了，你真应该重启它。我就是没法再接第五份、第六份工作了，而玩 TikTok 感觉就像第五、第六份工作。是啊，是啊，确实确实。嗯，酷。好了，太感谢你了 Claire，这又是一次特别精彩的对话，我真的学到了好多。哦，我也特别感谢你，聊得超开心。是啊。