OpenClaw, Claude Code, and the Future of Software | Peter Yang on The a16z Show
频道: a16z
视频: https://www.youtube.com/watch?v=UE8jx4dvlSQ
原文语言: en
统计: 共 74 轮 · Peter Yang 41 · 主持人 33
[0:00] Peter Yang
Interesting is that software will eat the world. I I feel like coding will eat all knowledge work, right? And we're kind of going in that direction already.
有意思的一点是，软件正在吞噬世界，而我感觉写代码会吞噬掉所有的知识工作。我们其实已经在朝这个方向走了。
[0:05] Peter Yang
The whole agent stack is emerging. Identity, payments, marketing, even CLI versus MCP, like all of these are really new things and I think a lot of the old playbook goes away. Yeah, it's a whole whole new world. I hope more companies will stay small. And I think the founders of this generation realize that. Like they want to stay as small as possible.
整个 agent 技术栈正在成形——身份认证、支付、营销，甚至 CLI 还是 MCP 这种选择，这些都是全新的东西。我觉得很多过去那套打法都不管用了。对，这是一个全新的世界。我希望会有更多公司选择保持小规模。我觉得这一代创始人是想明白了的，他们就是想尽可能地把团队保持得小一点。
[0:24] 主持人
Yeah. And instead of having like a 10-person product team, you have like a two or three-person product team and instead of a bunch of agents help help
对。与其养一个十人的产品团队，不如就两三个人的产品小队，剩下的靠一堆 agent 来帮你干活。
[0:30] 主持人
you. Yeah. Someone tweeted that like the job market is so bad that I can only pursue my dreams now or something like that. So like it's it's like, you know, it's like, yeah, so maybe you lost your job, but like now you can actually do your own thing. Yeah, 100% and have a shot at actually achieving it, you know. Yeah. All right, welcome everyone. I've got my friend Peter Yang here. Peter, welcome. Yeah, good to be here. It's good to see you again.
有人在推特上发过一句话，大意是「现在就业市场这么差，反倒让我只能去追自己的梦想了」。你想啊，也许你丢了工作，但现在你反而能真正去做自己的事情了。对，完全同意，而且是真有机会把它做成。好，欢迎各位。今天请到了我的朋友 Peter Yang。Peter，欢迎你。很高兴来到这里，也很高兴又见到你。
[0:54] 主持人
Yeah, yeah, it's great to see you. Peter and I worked together at Credit Karma for a brief stint and then we went our separate ways and you know, I rediscovered Peter from his prolific posts on X and your YouTube and you know, you've got a little bit of a a Clark Kent Superman thing going cuz you still got a day job, right?
是啊，很高兴见到你。我和 Peter 当年在 Credit Karma 一起共事过一小段时间，后来各奔东西。再后来我是从你在 X 上高产的发帖、还有你的 YouTube 频道重新发现了你。你身上有点那种 Clark Kent 和超人的双重身份的感觉，因为你现在还有一份正职，对吧？
[1:11] Peter Yang
That's right. I still have a day job. Yes. Yeah, can you share where? Uh yeah, I work at Roblox as a PM. Amazing, Roblox. Andreessen portfolio company. One of my favorites. Well, incredible man. Let's get right into it. Maybe I'll start with a softball fun question and then you know, we're going to talk about everything in the claw ecosystem. We're going to talk about coding agents. We're talking about a little bit about maybe what students should study, advice and some of the things that you've talked about online.
没错，我还有一份正职。对。能说说在哪儿吗？嗯，可以，我在 Roblox 做 PM。太棒了，Roblox，是 Andreessen 投资组合里的公司，也是我最喜欢的公司之一。真厉害啊老兄。咱们直接进入正题吧。我先抛个轻松的暖场问题，然后我们会聊整个 claw 生态、聊编程 agent，也会稍微聊聊学生该学点什么、一些建议，还有你在网上聊过的一些话题。
[1:33] 主持人
Yeah. Sure. Um maybe to start, what is the name of your Oh, how many claws do you have and tell me their names? Uh I only have one. I call her Zoe. Zoe.
好的，没问题。那先从这个开始吧——你那只 claw 叫什么名字？哦，你养了几只 claw，分别都叫什么？嗯，我只有一只，我叫她 Zoe。Zoe。
[1:41] Peter Yang
But I have like multiple conversations going with her. Okay.
不过我同时跟她开着好几个对话。好。
[1:44] 主持人
Yeah. And why Zoe? Um I I I have two girls and I was going to call my younger one Zoe and I did not. So I'm like I call my open claw Zoe instead. I see. I see. Yes. Yeah, this is your fallback plan. Yeah. tell me a little bit about, you know, Open Claw, how you discovered it, um how you're using it today, and and what you think the implications are. Yeah, I was lucky to interview Peter Steinberger before he became super famous and the whole thing blew up. And um then rather than interview him, I I like I set up the thing. It took forever to set up. It was super janky. Mhm. Um and and yeah, it does a lot of things for me. It like pulls analytics for me across um YouTube and like my Mercury bank account. Mhm. It um can update Google Docs for me. It can build little websites for me. But if I was honest with you, dude, like I mostly just talk to it through voice and get voice replies. And like every other day I ask it to give me like a pep pep talk. Or like, you know, give give me like like look look look through all your memory and like give me some like deep insights that I I don't know about. Okay. And and like it it gave me like like I remember I was on a walk and gave me like a 3-minute pep talk that was like really amazing. Really amazing. Like it some something about like uh like oh, you're like pointing me about your creator business and blah blah and like your your job, but like doesn't matter that your kids, you know, summer of four are going to grow up very soon and they're not going to want to spend time with you. Wow. So like you should really optimize for, you know, them instead. Yeah. It's really cool. And I mean, very cool, but also something that all of the language models could have done prior. So, what's the difference between this and in a use case like
嗯，那为什么叫 Zoe？我有两个女儿，本来打算给小女儿取名叫 Zoe，结果没用上。所以我就拿这个名字给我的 OpenClaw 用了。原来如此，这是你的备用方案。哈哈。跟我讲讲 OpenClaw 吧，你是怎么发现它的、现在怎么用它、以及你觉得它意味着什么。我挺幸运的，在 Peter Steinberger 大火、整个事情爆起来之前就采访过他。后来与其说采访，我干脆就自己把这套东西搭起来了。配置花了好久，特别糙。嗯。它现在帮我做很多事，比如帮我跨平台拉数据分析，包括 YouTube 和我的 Mercury 银行账户的数据。嗯。它还能帮我更新 Google Docs，帮我搭些小网站。但说实话老兄，我大部分时候就是用语音跟它聊天、听它语音回复。差不多每隔一天我就让它给我来段打气的话，或者说，让它「翻一遍你所有的记忆，然后给我一些我自己都没意识到的深刻洞察」。然后呢，我记得有一次我在散步，它给我来了段三分钟的鼓励，真的特别打动我，特别棒。它说的大概是「你一直在惦记你的创作者事业、你的工作，但这些都不重要——你那俩四岁的孩子很快就会长大，到时候就不愿意再花时间陪你了」。哇。「所以你真该把精力放在他们身上」。是吧，真的很酷。我是说，是很酷，但这些其实之前的语言模型也都能做到。那这个用例跟那些模型的区别到底在哪儿？
[3:11] Peter Yang
that? Yeah, that's a very good question. So, I don't know because I have it installed on Telegram, it just feels like more personal than using like Claude or ChatGPT. And and and just feels like something I can like text in bed. This is probably not very healthy, but like I text to it in bed. I I just talk to it during my commute. And it feels like it feels more like a personal like like actual human. Yeah. Yeah. So, then so how much for you is Open Claw about the kind of interface, like pushing it to messaging and you know, maybe helping to trick our brain into feeling like, "Hey, this is a person or a person-esque thing." Um versus all the other components of the stack, the self-modification, the skills directory, um every all the all the rest?
这个问题问得很好。我也说不太清，可能因为我把它装在了 Telegram 上，感觉就是比用 Claude 或 ChatGPT 更私人、更亲近，就像是个我可以躺床上发消息聊天的对象。这么说可能不太健康，但我确实会躺床上跟它发消息，通勤路上也跟它聊，感觉它更像一个真实的人。那对你来说，OpenClaw 的价值有多少是来自这种交互方式——把它塞进消息软件里，某种程度上「骗」过我们的大脑，让我们觉得「嘿，这是个人，或者类似人的东西」？相比之下，技术栈里其他那些组件——自我修改、skills 目录这些——又占多少？
[3:49] Peter Yang
I I think it's probably like 70-80% just like the per per per personal part of it part of it cuz I mostly just talk to it and like, you know, through voice. But I also think like is is something First First First of all, it is pretty janky. It tends to forget things things a lot.
我觉得大概有百分之七八十就是那种「私人感」，因为我基本就是用语音跟它聊。不过我也觉得，首先得承认它确实挺糙的，老是忘事。
[4:01] Peter Yang
Yeah. You have to keep reminding it. But like any kind of zany idea that I have, I just have to talk to it and it can probably just do. Like Like it's kind of just like Um like the other day I I was doing voice replies with it. I was like, "Hey, can we just have a live phone call instead?" Mhm. And and then it was like, "Okay, you you got to connect Twilio You got to do all this stuff." And then okay, fine. I I went off and did it. Yeah.
对，你得不停提醒它。但只要我有什么稀奇古怪的点子，我跟它一说，它基本就能搞定。比如前几天我在跟它用语音对话，我说「嘿，咱们能不能直接打个真的电话？」嗯。它就说「行啊，那你得连上 Twilio，得做这做那」。然后好吧，我就真去把这些都弄好了。对。
[4:20] Peter Yang
we had a phone call. It It called my phone.
然后我们就真打上电话了，它直接打到了我手机上。
[4:22] 主持人
You have that set up? I've been dying to set that up. Okay.
你把这个搭好了？我一直特别想搭这个。厉害。
[4:25] Peter Yang
not very good though. Like the latency is bad. But like the fact that I was able to get it going is like pretty impressive. So it's kind of like any kind of crazy idea I have, it can kind of kind of do. And then in practice, how are you doing that? Are you asking it to write a skill on the fly or are you discovering a skill? How much of the code gen are you actually using? Um I mean, I talk to it in like a super casual way with like just just like a friend. So I'm like, "Hey hey hey, you know, hey so you can can like can we have a phone call?" Yeah. And it's like, "Okay, you got to do all this stuff." I was like, "Okay, fine. I I'll I'll open my computer. I'll do all this stuff." And then it was like, "Give me a call." And then we trouble shoot a little bit and then it works. So like I with Claude I have like very fancy prompts. Uh like very long prompts. But with open Claude, I I just kind of text it. Yeah. Really interesting. So we sort of touched on a couple things actually. So one, there's mobile messaging. There's the memory system. Um there's the sort of code generation component. Um how much do you think the memory system like is it innovative because it's file based? You said that it forgets things, but so do language models. Like do you think the memory system is well done? Does it hold it back or does it enable it? I I I think the default memory system is actually not that great.
不过效果不算好，延迟挺大的。但能把这事跑起来本身就挺让人惊讶的。所以基本上我有任何疯狂的点子，它都能多少帮我实现。那实际操作的时候你是怎么做的？是让它现场写一个 skill，还是去找一个现成的 skill？你实际用到的代码生成有多少？嗯，我就是用特别随意的方式跟它聊，就跟跟朋友说话一样。我会说「嘿嘿嘿，那个，咱能打个电话吗？」对。它就说「行，那你得做这一堆事」。我就说「好吧，行，我打开电脑，把这些都弄了」。然后它说「给我打个电话吧」。我们再一起调试调试，然后就跑通了。所以在用 Claude 的时候我会写那种很花哨、很长的 prompt，但用 OpenClaw，我就随手发消息给它。对。真有意思。所以我们刚才其实顺带提到了几件事。一是移动端的消息交互，二是记忆系统，三是代码生成这部分。那你觉得这个记忆系统怎么样——它之所以创新是因为它是基于文件的吗？你说它会忘事，可语言模型也会忘啊。你觉得这个记忆系统做得好吗？是拖了它的后腿，还是成就了它？我觉得默认的那套记忆系统其实做得不怎么样。
[5:27] Peter Yang
Okay. Like the way I understand it works is like just like a memory.md text file. Yes.
嗯。据我理解，它的工作方式就是一个 memory.md 文本文件。对。
[5:31] Peter Yang
And then every day Per day, right? And then every day it updates. And it tends to forget things a lot. Yeah.
然后每天——是按天来的，对吧？每天它会更新一次。它经常忘事。对。
[5:36] Peter Yang
So so I actually installed this like three-layer memory system that to be honest, I I don't fully understand. But That's fancy. It has like Toby's QMD search tool. Okay. Uh so I installed that and and then I installed like a 2 GB thing. And and and then it got a little bit better. Okay.
所以我后来装了一套三层的记忆系统，老实说我也没完全搞懂它的原理。那挺高级的。它用了 Toby 的那个 QMD 搜索工具。好。我把那个装上了，然后又装了个 2GB 的东西，装完之后效果稍微好了点。好。
[5:52] Peter Yang
Um, but I I still have to remind it like I have to I have to put it into the agents.md like hey like before you answer any question for me like go through all your memory and like check everything. Yeah. And they also tend to forget that it can do stuff. Like you know, Right.
不过我还是得提醒它，我得在 agents.md 里写上「嘿，在回答我任何问题之前，先把你所有的记忆过一遍，每样都查一下」。对。而且它还老忘记自己能干什么。比如说，对吧。
[6:04] Peter Yang
like can you update my Google Doc? It's like oh I can't do that. Yeah, yes you can. It's it's in your it's in your file. Yes, yeah, yeah.
比如「你能帮我更新一下 Google Doc 吗？」它就说「哦，这我做不了」。我就说「不，你能，这功能就在你的文件里写着呢」。对，对对。
[6:09] Peter Yang
So, you have to remind it. Yeah.
所以你得提醒它。对。
[6:10] 主持人
Yeah. Really interesting. Well well maybe let's get into a little bit of the controversy. Sure. You know, you'd said that apps will die cause going to be everything and everywhere. I mean talk us through that that point of view. Yeah, well first of all I I I I tweet all all kinds of random crap that's like not super well thought out. We take it all as fact.
嗯，真有意思。那我们或许可以聊点有争议的话题。当然。你说过 app 会消亡，因为一切都会无处不在、随处可得。跟我们讲讲这个观点。好，首先得说明，我在推特上发的全是各种没怎么深思熟虑的破玩意儿。我们可都把它当真理对待呢。
[6:27] Peter Yang
Yeah. Um, but I do think like um, like ever since I set up all these apps like Mercury, MCP, and all this kind of crap on my open claw, like I I don't actually open those apps much anymore, you know? But I do agree with you. Like I I I think the ones that are going to die first or like maybe get less usage first is like apps that you're just opening to try to complete a task. Like you actually try to do something, you know? Like apps that you're opening to like get entertainment probably survive a little bit longer. But like apps that I'm opening to complete a task like it's just way easier to text my agent to do it for me. Yeah. It's like it's like you have a really good admin just to do stuff for it for you. Yeah.
哈哈。不过我确实觉得，自从我把 Mercury、MCP 这一堆东西都接到我的 OpenClaw 上之后，我就真的不怎么打开那些 app 了。但我也同意你说的。我觉得最先消亡、或者说最先开始被用得少的，是那些你纯粹为了完成某个任务才打开的 app——就是你真的要去办成一件事的那种。而那些你打开是为了消遣娱乐的 app，大概还能多活一阵子。但那些我打开只是为了完成任务的 app，直接发消息让我的 agent 帮我做实在太方便了。对。就好像你有个特别靠谱的助理替你把事都办了。对。
[7:01] 主持人
Yeah. And so how much are you finding it has this like reduced your smartphone usage outside of modulo open claw? Yeah. Um, no because like I'm I'm like a Twitter addict. So I still use the phone way way too much. But yeah, in terms of using those apps it's definitely re- reduced it. Yeah. Yeah. Like yeah cause you're not going to ask Zoe like hey read my X for me and tell me what's interesting. I mean it it it sends me like a morning briefing with like the top two tweets and stuff that like it trends but but yeah I I I still open X. I look through it. Yeah. Yeah. Yeah, you know it's interesting cause I've always had this theory that people open apps on their phone because they want to feel a feeling. Yeah. You know? And I think of course there's some like functional set of needs which is why you open calendar or something but I also think that you know, WhatsApp is you want to feel connected and Slack is you want to feel productive and of course you know TikTok is you want to feel entertained. So, I do wonder with just one agent, how do you sort of do the context switching of like when are you flirting, when are you getting done? I mean, there you know, in a sense app gives you a nice it sort of gives you a nice division of the intents Yeah, that that's the part.
对。那你发现，除了 OpenClaw 之外，它有没有让你刷手机的时间整体变少了？嗯，没有，因为我是个推特成瘾患者，所以手机还是用得太多太多了。但就用那些 app 而言，确实是减少了。对。因为你不会去问 Zoe「嘿，帮我读读我的 X，告诉我有啥有意思的」。其实它确实会给我发一份早间简报，列上当下热门的、最值得看的两三条推文之类的，但我自己还是会打开 X 一条条翻。对。你知道吗，这挺有意思的，因为我一直有个理论：人们打开手机上的 app，是因为想感受某种情绪。对吧？当然其中有一部分是功能性的需求，所以你会打开日历之类的。但我也觉得，比如 WhatsApp 是你想要那种被连接的感觉，Slack 是你想要那种高效的感觉，而 TikTok 当然是你想要娱乐的感觉。所以我就好奇，只用一个 agent 的话，你怎么去做这种情境切换——什么时候是在调情，什么时候是在干正事？某种意义上，app 给了你一个很好的、对意图的划分。对，问题就在这儿。
[7:59] Peter Yang
get with Zoe. That's a good point, but I I do have multiple channels set up with Zoe in Telegram. Like one is just to random voice replies, and then the other one is we're actually working on a project together. Oh.
在 Zoe 这边确实会有这个问题。这点说得好，不过我在 Telegram 里确实给 Zoe 开了好几个频道。一个就纯粹用来随便发语音聊天，另一个是我们俩在一起搞一个项目的。哦。
[8:08] Peter Yang
And then the other one I have like a public channel where like I'm giving dev demos. I don't want to reveal private information. Yes. So, I have like multiple channels. And is that um implemented as sub-agents or No, it's just some janky setup I I found online like you just you can set up the multiple Telegram channels and then I'm not sure if it actually remembers across context across the channels, but like you can have separate conversations at least. Got it. Yeah. And how how, you know, transparent are you with your agent? Like does it does it see your personal email, your Uh I'm I'm like super transparent. Well, I I did it by the Mac mini I set up its own email. Okay. And but I gave it like read access to my email and like calendar. And uh I also gave it like write access to some docs. Yeah. But it can't like scroll my entire drive or something, you know? So.
还有一个是公开频道，我在那儿做开发演示，不想泄露隐私信息。对。所以我有好几个频道。那这个是用 sub-agent 实现的，还是？不是，就是我在网上找到的某种很糙的搭法，你可以建好几个 Telegram 频道。我不太确定它跨频道之间是不是真能记住上下文，但至少你能分开来对话。明白了。对。那你跟你的 agent 之间有多透明？比如它能不能看你的私人邮件、你的——我超级透明的。我在那台 Mac mini 上给它单独建了个邮箱。好。然后我给了它读取我邮件和日历的权限，也给了它对一些文档的写权限。对。但它没法把我整个云盘都翻一遍，懂吧。所以。
[8:52] 主持人
How do you imagine Open Claw, which it's sort of an architectural primitive, how does it get productized, packaged for the world? I mean, I think that's what Peter Steinberger is working on at OpenAI, right? He's probably going to build something to ChatGPT which everybody uses. Uh so that ChatGPT can actually get stuff done for you and like maybe it feels more human. Yeah. Dude, like let me rant about ChatGPT for a little For some reason for some reason they they trained the model so that like at the end of every conversation it's always like "If you want, I can also do X and Y." Yeah. And dude, I I I got so annoyed by it that that kind of turned me off from ChatGPT. Really?
你觉得 OpenClaw——它某种意义上是个架构层面的原始组件——它会怎么被产品化、被打包推向大众？我觉得 Peter Steinberger 现在在 OpenAI 做的就是这个，对吧？他大概会做点东西接进大家都在用的 ChatGPT 里，让 ChatGPT 真能帮你把事办了，可能还更有人情味。对。老兄，让我吐槽一下 ChatGPT。不知道为什么，他们把模型训练成每段对话结尾总要来一句「如果你愿意，我还可以帮你做 X 和 Y」。对。老兄，我被这玩意儿烦得不行，这点直接让我对 ChatGPT 失去了兴趣。真的假的？
[9:26] Peter Yang
Yeah. So, so it probably increases their metrics, but like it's just like super annoying. It's like why don't you just do it in the first place? It's like are you a Claude guy now? Yeah, I'm I'm Claude guy now. But but I I do use Codex uh to code. Yeah, yeah. You like Codex, you prefer it to Cloud Code, or you use both?
对。所以这大概是提升了他们的指标，但实在是太烦人了。就是——你干嘛不一开始就直接做了呢？那你现在是 Claude 党了？对，我现在是 Claude 党了。不过我写代码确实会用 Codex。嗯嗯。你喜欢 Codex，你觉得它比 Claude Code 更好用，还是两个都用？
[9:41] Peter Yang
Um Codex is when I want to try to build something real, and Cloud Code is when I'm just like vibing vibing. It's like very very vibing, you know. Well, it's interesting. I think they live at different points in you know, there's a sort of space of tradeoffs. Yeah. Whereas I find um Cloud Code and Opus 46, it's a little more chatty, it makes more assumptions, but it can be more pleasant for a synchronous experience.
嗯，想认真做点真东西的时候我用 Codex，纯粹想随便玩玩、vibe coding 的时候就用 Claude Code，就是那种特别随性的状态，懂吧。挺有意思的，我觉得它俩处在不同的位置上，毕竟有一整套权衡取舍的空间。对。我的感受是 Claude Code 加 Opus 46 话稍微多一点、会做更多假设，但作为一种同步交互的体验它会更舒服。
[9:59] 主持人
Yeah. Whereas Codex, it really thinks hard, and it's more often accurate, but sometimes it's sort of like being in a conversation where the other person pauses for like 3 minutes to think. Yeah. Yeah, you don't have to flow flow state, right? It's hard to get in a flow. Like Cloud Code, dude, I I tweeted the other day that Cloud Code is almost like uh like a slot machine. It's like it's like it has different things each time. It's just like 100% dude. It's like a slot machine. Look, I I do think that if you think Remember we were talking about in the old social networking era, it was variable scheduled rewards, right? That was the whole magic of it. Like you open your Facebook feed, and you know, once in a while it's like boring boring, oh my god, this is so exciting.
对。而 Codex 则是真的会很努力地深思，准确率往往更高，但有时候那感觉就像跟一个会突然停下来想三分钟的人聊天。对。你没法进入心流状态，对吧？很难进入心流。Claude Code 这东西，老兄，我前几天发推说它简直像台老虎机。每次出来的东西都不一样，简直百分之百就是台老虎机。你看，我确实觉得——还记得我们以前聊过老社交网络时代吗，那时候靠的是「可变间隔奖励」，对吧？这就是它全部的魔力所在。你打开 Facebook 信息流，时不时刷到的是无聊、无聊、然后突然「我的天，这也太精彩了」。
[10:32] 主持人
And I the coding agents have the exact same property. Also, the time is variable. So, sometimes you get something in a second, sometimes it takes 5 minutes, so up to a certain point I actually think that both of those things give it that casino like feeling. Yeah. And and the other thing that's very different about the product strategy, or maybe it's just the way it works, is like Cloud Code is kind of like self-explanatory. Yeah. And Cloud Code, you have all this crazy You have like hooks and like skills and Like you have to you have to Plugins.
我觉得编程 agent 有着一模一样的特性。而且连耗时也是不确定的——有时候一秒就给你结果，有时候要等五分钟。所以在一定程度上，我真觉得这两点加在一起给了它那种像赌场一样的感觉。对。还有一点很不一样的，关于产品策略，或者说也许这就是它本来的运作方式——OpenClaw 算是某种程度上不言自明、开箱即懂的。对。而 Claude Code，你有一大堆疯狂的——你有 hooks、有 skills，还有——你必须得——还有 plugins。
[10:57] Peter Yang
Like if if you're if you're not following Yeah, if you're not following X, you have no idea how to customize this thing. Yeah.
对，如果你不刷 X，你根本不知道这玩意儿该怎么自定义。对。
[11:01] Peter Yang
But once you customize it, you kind of feel like it's part of you. So, it's kind of it's kind of hard to
但你一旦把它定制好了，就会觉得它成了你的一部分。所以很难——
[11:05] 主持人
turn. It's interesting with uh so, I've customized mine cuz also I read the long thing that Boris put up.
很难弃用。挺有意思的，我自己也定制过我的，因为我把 Boris 发的那篇长文也读了。
[11:10] 主持人
Yeah. But I will say that I think that you know, Cloud Code, a lot of the reasons that I enjoy it are just harness features.
对。不过我得说，我喜欢 Claude Code 的很多原因，其实都是 harness 层面的功能。
[11:17] 主持人
Mhm. You know, like for example, if you cut an image, you have to paste it into a file before and then paste that file into Codex. Okay. You can't just take a subset of the screen, screenshot it, and then paste it directly into Codex the same way you can with Cloud Code.
嗯。比如说，在 Codex 里如果你截了张图，你得先把它存成一个文件，再把那个文件粘进去。明白。你没法像在 Claude Code 里那样，直接框选屏幕的一部分、截图，然后直接粘进 Codex。
[11:31] Peter Yang
Already, okay, okay. So, just like little things like that, you know, Claude code added voice. It's a little bit janky right now, but it's going in the right direction. So, they've just got a bunch of quality of life things.
了解，好的。就是诸如此类的小细节。还有 Claude Code 加了语音功能，现在虽然有点糙，但方向是对的。所以他们就是攒了一堆提升体验的小功能。
[11:40] 主持人
Yeah. You know, Claude code speaks to Claude in Chrome. Okay. And Codex doesn't speak to Atlas. Got it. So, I think these are all things that OpenAI will fix. Yeah, I think Codex is actually a much better model, um, but they don't exist today. Yeah, yeah, they they need to fix it. I mean, they're they're going to go all all in on Codex, I'm sure. Yeah. Talk to me about coding agents. Like, what's your general view? You know, do you think it's the end of SaaS? Do you think these are just a toy? Uh, well, first of all, I'm I'm I'm like I'm not an engineer, so I'm like a novice. But, I do hear that, um, like I was talking to some folks, uh, the other day and like like a AI-native star startup. And they're basically try trying to they have a bunch of vibe coders. Mhm. And all the vibe coders are just trying to build internal tools that replace their SaaS. That they're paying for.
对。你看，Claude Code 能跟 Claude in Chrome 联动，而 Codex 没法跟 Atlas 联动。明白。所以我觉得这些都是 OpenAI 早晚会补上的。对，我其实觉得 Codex 的模型本身要强得多，只是这些功能现在还不存在。对对，他们得把这些补上。我相信他们肯定会全力押注在 Codex 上。对。跟我聊聊编程 agent 吧。你大体上怎么看？你觉得这是 SaaS 的终结吗？还是说这些就是个玩具？嗯，首先得说，我不是工程师，所以我算个外行。但我确实听说——前几天我跟一些人聊，是一家 AI 原生的初创公司。他们基本上是有一帮 vibe coder，嗯，这帮 vibe coder 都在忙着搭内部工具，用来替代他们正在付费的那些 SaaS。
[12:22] 主持人
Really? Yeah. So, this is an actual company that's doing this.
真的吗？对。所以是有一家实实在在的公司在这么干。
[12:24] Peter Yang
This is an actual company, yeah. It's an AI-native company. It's like one of the vibe coding companies, like one of the more popular ones. Interesting. Yeah. It's interesting Oh, oh, I see. So, they're not actually an app gen company. They're app gen company and they they're paying for a bunch of SaaS, like, uh, and they they want to get rid of the payment. They want to just buy buy app coding internal tools. By using Okay, so, in that case, that that they might be the most extreme form of adopter cuz their own product is app gen, so they should use app gen for everything.
是一家真实存在的公司，对。是家 AI 原生公司，是那种做 vibe coding 的公司，还是其中比较有名的一家。有意思。对。挺有意思的——哦，我明白了。所以他们本身不是做 app 生成的公司？他们就是做 app 生成的公司，而他们在付费用一堆 SaaS，嗯，他们想把这些付费砍掉，想干脆全用自己 app 生成出来的内部工具。那这么说的话，他们大概是最极端的那种采用者，因为他们自己的产品就是 app 生成，所以他们本该把一切都用 app 生成来搞定。
[12:46] Peter Yang
Yeah. I guess, is your prediction though that the average company will churn off of Slack or Deel or, you know, Um, I don't think I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I Uh, but some of the other ones, they are pretty complicated, you know? So, like, it's going to be be hard to vibe code like that stuff. But, but I feel like if you have like an app like maybe Calendly or like something more simple,
对。那我想问，你的预测是不是说，普通公司会从 Slack、Deel 之类的服务上流失掉？嗯，我不觉得——我不觉得。不过其他有些服务确实挺复杂的，懂吧。所以那种东西要 vibe code 出来会很难。但我感觉，如果是像 Calendly 那种、或者更简单一点的 app，那是有可能的。
[13:11] Peter Yang
Yeah. then why why why should I why should I pay for it? Like, I just Why should I pay for it? Well, the counterpoint is that it's not that expensive, and do you really want to maintain your own Calendly thing? Yeah. You know, versus pay 20 bucks a month, it always gets updated, it's always up, Yeah, yeah. you know, cuz there's just like a fixed amount of capacity that anyone in the organization is going to have for all this stuff. Yeah, that that that that's true. Unless you hire like dedicated by quarters like the startup doesn't make you just want to buy stuff. But then it's like, you know, the cost benefit versus just paying for Calendly. Yeah, yeah, yeah. Yeah. Yeah. It's interesting about like for example like a lot of people are tweeting about Figma. Yeah. Yeah. Uh like like the stock is down like you know are are are you going to survive? Yeah. And um I feel like the jury's out there. Like it's kind of hard to say. Yeah. I I I feel like all the designers are still on Figma but like as a designer you kind of need to learn how to buy code otherwise you're going to like if you if you only know how to do Figma like you're probably going to be like out of date. In couple years.
对啊，那我为什么要付费？我干嘛要为它掏钱？——不过反过来想，它也没那么贵，你真愿意自己去维护一个 Calendly 那样的东西吗？是啊。比起一个月花 20 块，它一直在更新、一直在线。对对。因为组织里任何人能投在这些杂事上的精力总归是有限的。这倒是真的。除非你专门为此招人，但创业公司不会让你光想着买东西。可这又得算成本收益账，跟直接付钱用 Calendly 比划算不划算。是是是。说到这个挺有意思的，比如最近好多人在推上聊 Figma，对吧。股价跌了不少，大家在问它还能不能撑下去。我感觉现在还说不准，挺难下结论的。设计师们目前确实还都在用 Figma，但作为设计师你多少得学会 vibe coding（氛围编程），不然要是你只会用 Figma，那过几年你大概就跟不上趟了。
[14:04] 主持人
Yeah. My counterpoint to that is that I think that I've thought a lot about the sort of thinking tools versus making tools, right? The IDE was historically a making tool. It's a place for execution. I think it's migrating away from that. And now with execution going to zero I think these sort of like multi-agent, you know, next gen IDEs a lot of them are about trying things and using the trial and error as a way to inform your thinking. Like a lot of times I'll just build a feature in a really naive way and I'll hammer the coding agent until it works. Then I'll say, "Hey, write all the things that you would have done differently." And I'll go back to the initial point and redo it. So I wonder if and I think Figma actually does both. I think it's a place for design execution but it's also an important place for design thinking. And I think that's their opportunity to be highly relevant in the new stack. Yeah, I totally agree. I totally agree. But I I I I think A16Z has like you guys investing pencil. Or something? Pencil.dev. Yeah.
对。我对这点的看法是，我琢磨了很久所谓「思考工具」和「执行工具」的区别。IDE 在历史上一直是个执行工具，是干活的地方。但我觉得它正在脱离这个定位。现在执行成本趋近于零，这些新一代的多 agent IDE，很多其实是在帮你做尝试，把试错本身变成一种厘清思路的方式。很多时候我会先用最笨的办法把一个功能堆出来，然后死磕这个编码 agent 直到它跑通，再跟它说「把你本来应该换种方式做的地方都列出来」，然后我回到最初那一步重做一遍。所以我在想——我觉得 Figma 其实两头都占。它既是做设计执行的地方，也是做设计思考的重要场所。我觉得这正是它能在新技术栈里保持高度相关性的机会。是啊，我完全同意。不过 a16z 你们好像投了 pencil 之类的？Pencil.dev，对。
[14:57] Peter Yang
run it and then like um Yeah, Figma needs to like level up his AI tooling cuz you know like watching these AI agents collaborate with you and like do stuff is like very very interesting. I know it's top of mind for them. What do you think are the most under discussed capabilities of coding agents? You know, what's what's under hyped and maybe what's over hyped as well. This is probably not under hyped but I I but you know like you know I I I feel like interesting that software will eat the world. I I feel like coding will eat all knowledge work, right? And we're kind of going in that direction already. Like I I think lovable recently launched like today that they can support everything and can make decks. Yeah, so So yeah, so I and I feel like everyone is chasing this. Like I don't know who's probably in the lead. Yeah. But like, you know, like I I don't want to use PowerPoint anymore. I don't want to like write a Like I hate writing Google Docs, dude. Like I lost my entire life. Yeah, yeah, yeah. So so like but but but the other day I was writing my blog post and instead of typing it out, I was like, "Hey, let's let me just use Claude code and like, you know, let me give you a bunch of feedback and you you write it for me." Yeah. And then you just keep improving it. Um it it did the first 80%. The last 20% I still had to manually like go in there like tweak tweak tweak stuff. Yeah. But like that that's the way I work now. I I never start from zero. Like I always get the first 80% from AI. Right. Yeah. Yeah, it's interesting, you know, if you look at there there are also like historical analogies of this. I think Satya said this, which is that Excel is the most powerful or most popular programming language in the world.
跑一下，然后……对，Figma 得把它的 AI 工具升级一下，因为看着这些 AI agent 跟你协作、帮你干活，真的特别有意思。我知道这事在他们那边也是重中之重。——你觉得编码 agent 里哪些能力是被讨论得最不够的？什么被低估了，可能还有什么被过度吹捧了？这个大概算不上被低估，但我觉得挺有意思的：以前说软件吞噬世界，我现在感觉编码会吞噬所有知识工作，对吧？我们其实已经在往那个方向走了。比如 Lovable 最近——我记得就在今天——发布说他们什么都能支持，连 PPT 都能做。所以大家都在追这个，我也不知道谁现在领先。但你知道，我已经不想再用 PowerPoint 了，也不想写……老兄，我特讨厌写 Google Docs，简直耗掉我半条命。是是是。可是前几天我写博客的时候，没有自己一个字一个字敲，而是想「干脆用 Claude Code，我给你一堆反馈，你帮我写」。然后就这么不停打磨。它头 80% 写得挺好，最后那 20% 我还得手动进去一点点改。对。但我现在就是这么干活的，我再也不从零开始，永远先让 AI 把头 80% 搞定。——对，这挺有意思，你看历史上也有类似的类比。我记得 Satya 说过，Excel 是世界上最强大、最流行的编程语言。
[16:21] 主持人
Yeah. And that it's sort of a programming language that millions and millions, I mean 100 million plus people must know, maybe even more. Um and yet we don't think of it that way. It's a way to sort of describe and solve problems. Yeah. And I think coding agents are going to be that, of course, times a thousand. Yeah. Where even things that feel subjective, like writing Google Docs, can be represented in the coding domain in such a way that it's more satisfying, more productive, more high leverage to use agents to do it. Yeah, cuz Excel was like popular because it's super approachable, right?
对。它算是一种编程语言，全世界有上亿、可能远不止一亿人会用，可我们从来没把它当成编程语言看。它本质上是一种描述和解决问题的方式。而我觉得编码 agent 也会成为这样的东西，当然量级要再放大一千倍。哪怕是那些看起来很主观的事，比如写 Google Docs，都能在编码这个领域里被表达出来，而且用 agent 去做会更有成就感、更高产、更高杠杆。对，因为 Excel 之所以流行，就是因为它特别好上手，对吧？
[16:50] Peter Yang
Yes. And like coding is just the code is basically gone. Like it's like abstract away. You're just talking to some some agent and getting it to do do stuff. So
没错。而编码这件事，代码本身基本上已经消失了，被抽象掉了。你就是跟某个 agent 对话，让它去把活干了。所以——
[16:57] 主持人
Yeah. Yeah. Exactly. It's going to be huge yeah. What do you think the future company looks like? Is it just a bunch of agents with a CEO? Is the CEO an agent? I mean, what is the role for people in a company in the future? Okay, well, I have some hot takes. So we were we both worked at some companies together and um
对对，正是这样。这会非常了不起。——你觉得未来的公司会是什么样？是不是就一堆 agent 加一个 CEO？CEO 本身会不会也是个 agent？在未来的公司里，人到底扮演什么角色？好，我有几个比较大胆的观点。我们俩之前在几家公司共事过，那个——
[17:12] Peter Yang
Uh-huh. Uh let me give you a hot take, man. Maybe hope this out. But like I feel like as a company bigger, it tends to get it tends to become like a shitty shittier place to work, dude.
嗯哼。我先给你抛个暴论啊，希望别翻车。我感觉公司越大，往往就越容易变成一个更糟心的工作场所，老兄。
[17:21] Peter Yang
Yeah. Like like cuz there's like a lot of people you have to align.
对。因为有一大堆人需要去对齐。
[17:23] 主持人
that's axiomatic, yeah.
这是公理级别的真理，没错。
[17:24] Peter Yang
Right? And and I don't I don't remember at you know, maybe shouldn't mention this company, but I remember you know, our company there used to have one of these like OKR meetings and like everyone's in a room for like 3 hours talking about OKRs. I'm just like, dude, this is like a waste of my life. Yeah, so what where I'm going with this is I hope more companies will stay small. And and I I think the founders of this generation realize that. Like they want to stay as small as possible. Yeah. Um and instead of having like a 10-person prod team, you have like a two or three-person prod team and instead of having a bunch of agents help help you. Yeah.
对吧？我也不记得……也许不该点名这家公司，但我记得当年我们那家公司开那种 OKR 会，所有人挤在一间屋里花三个小时聊 OKR，我心想，老兄，这简直是在浪费我的生命。所以我想说的是，我希望以后会有更多公司选择保持小规模。我觉得这一代的创始人意识到了这一点，他们想尽可能保持精简。与其养一个 10 人的产品团队，不如只要两三个人，然后让一堆 agent 来帮你。
[17:50] Peter Yang
Cuz I I I think it's way easier to cross launch and align with the agents than with with humans. Yeah, well, actually in a in a sense the agents actually because it takes the emotion out of it, too. Like you can imagine if I sent my agent, you sent your agent to go negotiate something and they came out with some conclusion, it's not emotional. It's not
因为我觉得跟 agent 之间做跨团队协调、对齐，要比跟人之间容易太多了。——对，其实某种意义上，agent 的好处还在于它把情绪因素剔除掉了。你想，如果我派我的 agent、你派你的 agent 去谈某件事，它们谈出个结论来，这事一点都不带情绪，不会——
[18:07] 主持人
for either of us, you know? It's very objective, yeah. Yeah, yeah, exactly. Yeah, it's funny, you know, one of the things that we've been talking a bunch about is like what is the pro case for AI at work in terms of employee experience? And I think it's what you're describing, right? Like how do you increase the NPS of work? Yeah. So if we like go all the way back or even broadly the NPS of the human experience, right? Like think of the NPS of the day-to-day human in 10,000 BC when it's like just don't get eaten by the lion and that's like a good day, right? Or you know, maybe 100 years ago it's like, okay, don't get killed at the factory. Mhm. Um crushed by the the steam press or whatever else and um and now a lot of it is like, like just don't get sucked into some high emotion, you know, sort of negotiation with another VP's subordinate. Yeah, like a 50-message Slack thread going back and forth.
对我们俩谁来说都不掺杂感情，你懂吧？非常客观。对对，正是如此。说来有意思，我们最近一直在聊的一个话题是：从员工体验的角度看，AI 在职场的「正向价值主张」到底是什么？我觉得就是你刚才描述的那个——怎么提升工作的 NPS（净推荐值）。咱们干脆往回追溯，或者放大到整个人类体验的 NPS：想想公元前一万年普通人每天的 NPS，那时候只要别被狮子吃掉就算好日子了，对吧？或者一百年前，差不多是「别在工厂里被那台蒸汽压机压死」之类的。而到了现在，很大一部分变成了「别被卷进跟另一个 VP 下属之间那种情绪高涨的拉锯谈判里」。对，就像一条来回扯了 50 条消息的 Slack 长串。
[18:51] 主持人
Yeah, which exactly. Then eventually everyone's like, I don't want to tell the CEO and eventually it goes there and it's just terrible. Um so maybe the future of this is that a lot of that emotional subjection of work gets handled and we're sort of guiding the process but not in the middle of it in a way that just doesn't suit us as humans. Yeah, like I you know, I I I think that the world of a PM creator and like I feel like all the PMs actually just want to create products. They just want to create products and and
对，正是。然后最后所有人都「我可不想去跟 CEO 告状」，可结果还是闹到 CEO 那儿，那场面就特别难看。所以也许这件事的未来是：大量这种情绪化的工作折磨被自动接管掉，我们只负责引导流程，而不必再卡在中间那个对人类来说根本不舒服的位置上。对，我觉得 PM 作为创造者的世界——我感觉所有 PM 其实就是想做产品，他们就想做产品，然后——
[19:15] 主持人
why we all got into it, you know, it's it's so interesting I mean, Nikel talks about it all the time, but like every PM's sort of view view of the ideal PM is the innovator, you know? Like I came up with the new thing and it's like I sort of like had the big insight and it unlocked the product.
这正是我们当初入行的原因，对吧，特别有意思。我是说，Nikhil 老把这事挂在嘴边：几乎每个 PM 心目中理想的 PM，都是那个「创新者」形象——我想出了个新东西，我有了那个重大洞察，然后它把整个产品给打开了。
[19:29] Peter Yang
Yeah. I think the black pill is I don't think most PMs know how to do that. In fact, many companies have zero people that know how to do that at all in any function. Yeah. So, nonetheless, I think PMs aspire to be able to do that and they should either do it and either be successful or maybe not successful and move to a different function. I also feel like my hot take is like like basically all the PMs I know are trying to write code at nights and weekends. Yeah. And I feel like my hot take is that like I feel like if you're actually unemployed, like you probably have more time to be a builder and like to be innovative. Yeah.
对。但残酷的真相（black pill）是，我觉得大多数 PM 根本不知道该怎么做到这一点。事实上很多公司里，任何职能岗上能做到这点的人都是零。可即便如此，PM 们还是渴望能做到，而他们也应该去试——要么做成了，要么没做成、然后转去别的职能。我还有个暴论：基本上我认识的 PM 都在利用晚上和周末偷偷写代码。对。而我的暴论是，你要是真失业了，可能反而有更多时间去当一个 builder、去做创新。
[19:59] Peter Yang
actually like play all this stuff and like learn all this stuff. A lot of PMs are just trying to Or maybe be an engineer in the team, you know? I used to be an engineer and I got sort of I don't know if I got forced to be a PM, but maybe I also perceived
真去把这些东西玩一遍、学一遍。很多 PM 不过是在试图……或者干脆去团队里当个工程师，你懂吧？我以前就是工程师，后来不知怎么就——我不确定算不算被逼着转成了 PM，但也许我自己也觉得——
[20:10] Peter Yang
PM is like being a little more high status.
PM 这个岗位听起来地位高一点。
[20:11] 主持人
Yeah, yeah. When I joined Google, but then eventually you come around the other side you're like this is terrible, you know? Like you don't never really get the satisfaction Yeah.
对对。我刚进 Google 那会儿也这么想，但到后来你绕到另一头才发现，这岗位太难受了，你懂吧？你几乎从来得不到那种满足感——
[20:19] 主持人
of actually shipping other than like, you know, once a quarter when you ship. I mean the PM skills of like talking to users and like trying to figure out what to do like what's the problem to solve. Like those are all very important still. Yeah. And like but yeah, you got to wear multiple hats too. You got you got to like go go build a thing yourself, go prototype it and get some feedback and then maybe bring an engineer along. How much do you think that everyone has to go as fast as, you know, I mean like Gary was talking about stimmies and skipping sleep and Gary Tan like, you know, G stock. I mean is Hey, I mean is that like the default way that we all need to work or do you think there's a trade-off for thoughtfulness? I I I think it's very easy now with all these AI tools to just go in like 10 different directions at once. Yeah. So, sometimes you do have to slow down and try to figure out where you want to go. Yeah. But I also believe that like the traditional process where you like do annual planning and like do all this bull Like I I I feel like that doesn't really work anymore. You know?
——真正把东西发布出去的满足感，除了每季度发一次版的时候。当然 PM 那套技能，比如跟用户聊、搞清楚该做什么、要解决什么问题，这些到现在都还非常重要。对。但你也得身兼数职，得自己上手把东西做出来，先做个原型拿去收集反馈，然后也许再拉个工程师一起干。——你觉得是不是每个人都得拼到那种程度？我是说，Gary Tan 老讲什么嗑提神药、不睡觉、G 系列那套狠劲。喂，那种是不是我们所有人都得默认采用的工作方式，还是说为了想得更周全，得做点取舍？我觉得现在有了这么多 AI 工具，太容易一下子朝十个方向乱冲了。所以有时候你确实得慢下来，想清楚自己到底要去哪。但我同时也相信，那种你做年度规划、搞一大堆形式主义的传统流程——我感觉那套现在真的不灵了，你懂吧？
[21:08] 主持人
Yeah. Yeah, for the record, I love Gary and I like think the world of him. Yeah. So here's my view on that cuz I thought a bunch about it. You know who asked me this the other day? Hiten, um who's really really impressive. Yeah, yeah, in the show, yeah. So we're talking about this like sort of productivity porn and everybody's got 20 agents running and 20 monitors and blah blah blah. And like I do think when it comes to um fully realizing a local a sort of local maxima Mhm. you should go very fast, right? Mhm. So let's say you kind of hill climb, you get to the bottom of a new local maxima. I think with agents you should be able to get to the top of that hill extremely fast, right? Yes. You have a new insight, build everything around the insight so it's fully expressed. But then I think to get to the next, you know, the next sort of hill Mhm. you've got to probably slow down and almost stop and go touch grass and do whatever. Yeah. So I think there's this combination of like fast and slow that's probably the future way. Yeah, I think so. And and like you got to go on that random walk to try to find your product market fit, which take take us a while, right? So it's not like, yeah. So we were talking before we started recording about some of the business in a box platforms. Have you looked at them? Do you have a view? I've looked at Post Post that we talked about.
对。先声明一下，我很喜欢 Gary，我特别敬重他。所以这是我对这事的看法，因为我琢磨过不少。你知道前几天谁问过我这个问题吗？Hiten，这人真的非常厉害。对对，上过这个节目。我们当时在聊那种「生产力炫技」——人人都开着 20 个 agent、摆着 20 块显示器之类的。我确实觉得，当你要把一个局部最优解（local maxima）彻底吃透的时候，你应该跑得非常快。比方说你在爬山，刚到一座新山丘的山脚，我觉得借助 agent 你应该能极快地冲上这座山顶，对吧？你有了个新洞察，就围着这个洞察把所有东西都搭出来，让它充分展开。但接下来要去够下一座山丘的时候，你大概就得慢下来、几乎停下，出去走走、接接地气什么的。所以我觉得这种快慢结合，大概才是未来的方式。对，我也这么想。而且你还得来一段随机漫步去摸索产品市场契合度（PMF），这是要花点时间的。——对了，我们开录之前聊到一些「开箱即用做生意」（business in a box）的平台，你研究过吗？有看法没？我看过 Post，就咱们提过的那个 Post。
[22:14] Peter Yang
Uh-huh. Like I don't know if the guy like intentionally made it the opposite of AI slop or or is it kind of a He intended it to be I think so. Yes. Yes. Yes. Yes. That's funny. Well, I mean I I have a pretty big public presence, right? So I connect all my to to it and then um I mean it's it definitely gives a good peek into what's possible, but like right now it's probably still pretty like uh early stage. Like it's telling me to run like Facebook Facebook ads. Yeah. Why am I running Facebook ads? Yeah. You know? I don't know. Yeah, yeah. I I mean I'm very excited about it cuz it does feel like it's a path for more people to build companies. Yeah. Even if they're single, one-person companies. You know, like if you think about how competitive it is to build a billion-dollar business, like the markets that support it, the number of people trying versus 100 million versus 10 million versus $100,000 TAM. Yeah. Like maybe there are these pockets all over the country, all over the world, where there are opportunities for $100,000 TAM products.
嗯哼。我不太确定那哥们是故意把它做成「AI 流水线垃圾」（AI slop）的反面，还是说它本身就——他是有意为之的，我觉得是。对对对，是的。挺逗的。这么说吧，我有挺大的公开影响力对吧，所以我把我所有账号都接进去，然后——它确实让你窥见到一些可能性，但现在大概还处在很早期的阶段。比如它让我去投 Facebook 广告，我干嘛要投 Facebook 广告？对啊，我也不知道。不过我对它还是挺兴奋的，因为它确实像是一条让更多人能创业的路。哪怕是单打独斗的一人公司。你想想，要做一门十亿美元级别的生意有多卷——能撑起这种规模的市场、扎进去的人那么多；可对比一下，一亿、一千万，再到一个 10 万美元 TAM（潜在市场规模）的市场。也许全国、全世界各个角落，都藏着这种能做 10 万美元 TAM 产品的机会。
[23:05] 主持人
Yeah. And that would change somebody's life. Now, that's not an enterprise venture-backed company, but that's okay. Yeah. So, I hope that whole thesis works, cuz I do think it's a way to get more people to participate, you know? Yeah, that that that's my plan for my kids, dude. Like, I I want them to just build like um bootstrapped businesses in high school. Yeah. And they can skip the whole college and like uh core corporate life. Yeah. Well, dude, I think this is like, you know, for 10 years there's this moral panic about the kids wanting to be YouTubers. Yeah. You know, you're a YouTuber. Yeah, yeah, I am. Um you know, you know, in the in the vein of Mr. Beast. And and I think the like pro case for that actually is that the kids wanted to be entrepreneurs or have agency, and the only channel for people if they weren't programmers was creating YouTube videos, at least online. Yeah. So, if you're like an online native generation, you want to create something, you're not a programmer, you make a YouTube show. And now you can make a lot more than that. Yeah, you can you can build whatever you want.
对。而这种生意就足以改变某个人的人生了。当然，那不是企业级、拿风投的公司，但没关系。所以我希望整套设想能成立，因为我确实觉得这是一条让更多人参与进来的路。对，老兄，这就是我给我孩子定的计划。我想让他们在高中就去搞那种自力更生（bootstrapped）的小生意。对。这样他们就能跳过整个大学、跳过那套朝九晚五的公司生活。是啊。老兄，我觉得这事就好比——过去十年大家一直对「孩子都想当 YouTuber」这事道德恐慌，对吧。你自己就是个 YouTuber。对对，我是。就 Mr. Beast 那一路的。但我觉得这件事的正面解读其实是：孩子们想当的是创业者、想要掌控感（agency），而对那些不会编程的人来说，唯一的出口——至少在线上——就是拍 YouTube 视频。所以如果你是在网络上原生长大的一代，又想创造点什么，自己又不是程序员，那你就去做个 YouTube 节目。而现在，你能做的远不止于此了。对，你想造什么就能造什么。
[23:53] Peter Yang
Exactly. So, Exactly. It'd be very exciting. Yeah. Any other hot takes for us? I'm curious about your thoughts about this, actually. Like, so, I feel like a lot of people are saying like, "Agents will interact with your product first." Right. And and then you see all these great companies like uh building like APIs and MCPs. Mhm. But like, how do you how do you think about like, you know, you being a customer for a while? So, like, the customer's like, you got to get the user to come back and use your product, right? Yeah. But but now, like, the user's like, "Hey, go send the agent to use it." So, how do you think about retention and all this like basic stuff? Like, how do you you know? Or or even like brand equity, cuz the agent just like points to some API. Like, you know. I don't Yeah, I don't know. Okay, so, I I think one of it's a I don't know is the Yeah. is the truth, but um there's I have a few thoughts. So, one, I think that a lot of the sophistication sophistication that happened in consumer happened because we had to have indirect monetization. Mhm. Okay, like, we just can we're never charging consumers directly for these products, which is why you got ads and stuff. ads and just large-scale networks, and we all obsessed with retention and engagement and whales and all of these things really, really mattered because we didn't simply charge people for products. Mhm. So, I think one big thing that's actually really helped in the AI era with that is that consumers are now excited to try new things, are willing to pay, they're willing to pay a really high price point. Um there's also consumption revenue in consumer for the first time. Like tokens and stuff.
正是。所以——没错，这会非常令人兴奋。对。还有别的暴论要分享吗？我其实挺好奇你怎么看这个：我感觉很多人在说「以后是 agent 先跟你的产品打交道」，对吧。然后你就看到一堆很棒的公司在拼命做 API、做 MCP。可问题是，你怎么看待——你也当了这么久甲方——以前是「你得让用户回来用你的产品」，对吧。但现在变成用户说「喂，派 agent 去用就行了」。那留存这些事你怎么想？还有更基础的那些东西。甚至连品牌资产都成问题，因为 agent 就只是指向某个 API 而已，你懂吧。我不——对，我也说不好。好，我觉得其中一点是——「我不知道」恐怕才是真实答案，但我有几点想法。第一，我认为消费互联网（consumer）之所以发展出那么多复杂玩法，是因为我们当时只能靠间接变现。对吧，我们基本永远没法直接向消费者收这些产品的钱，所以才有了广告这些东西——广告加上大规模网络，于是我们全都痴迷于留存、活跃、大 R（whales）这一套，这些之所以那么要命，就是因为我们没法直接向用户收产品的钱。所以 AI 时代有一个特别大的帮助，就是消费者现在愿意尝试新东西、愿意付费，而且愿意接受相当高的价位。消费端还头一回有了「按用量计费」的收入，比如 token 之类的。
[25:11] 主持人
Yeah, yeah, for like tokens. You have your subscription plus your tokens. So, and then the actual like the sort of blessing in disguise is that there are real costs as well. You have these inference costs. So, you're like, "Wow, we have to charge our customer on day one." So, I think one thing is that like the business model simplification, I think will really help with a lot of what you're describing. Two, I think that a lot of the products will have a sort of, you know, it'll have an API interface for your agents to interact with or for, you know, for transactional sort of wrote things.
对对，比如 token。你有订阅费，再加上 token 费用。然后还有个「因祸得福」的地方，就是这里确实存在真实成本——你有这些推理（inference）成本。于是你会想「哇，我们第一天就得向客户收钱了」。所以我觉得其中一点是，这种商业模式的简化会对你刚才描述的那一堆问题帮上大忙。第二，我觉得很多产品都会有那么一层——它会有个 API 接口，供你的 agent 来对接，或者用来处理那些交易型、机械重复的事。
[25:39] 主持人
Yeah, yeah. And then it'll have like a consumption-based interface as well. So, you you can also imagine a like a mobile app where there's like the feed, but then you can kind of turn it over to where the wires are. And you can just ask for things to get done or you can just see the log of the things that got done. Yeah, and maybe people will do both, right?
对对。然后它还会有一个基于「消费/浏览」的界面。所以你也可以想象一个 App，里面有信息流（feed），但你随时能把它翻过来，看到底层的「线路」在哪。你可以直接吩咐让事情办好，也可以只看那些已经办完的事的日志。对，而且也许两种人们都会用，对吧？
[25:54] Peter Yang
I mean, you can imagine Credit Karma where we work, you know? Like once you want to just take a look at your score history and a few other maybe credit card offers, I don't know. I mean Yeah, yeah. If I get my score, I'll have all the kind of credit card offers, I'll definitely do that.
我是说，你可以拿我们待过的 Credit Karma 来想象，你懂吧？比如有时候你就想看看自己的信用分历史，再瞅几张可能合适的信用卡推荐，我也说不好。我是说——对对。要是它把我的分数算出来，再给我配上各种信用卡推荐，那我肯定会去看。
[26:04] 主持人
Yeah, 100%. Exactly. On the other hand, like sometimes you want to just be like, "Yo, like can you just fix all my stuff or like what stuff did you fix this week? How much money did I save?" You know? Got it. Got it. Yeah. It's definitely interesting, yeah. But look, I also just think the whole agent stack is emerging. Yeah. Identity, payments, marketing, we don't even even CLI versus MCP, like Yeah. all of these are really new things and I think a lot of the old playbook goes away. Yeah, it's a whole whole new world. And like in 2025, I thought agents was overhyped, but now I think it's really kind of coming coming coming. Me too. I know. It's just the word is frustrating cuz it gets so overloaded.
对，百分之百。正是如此。但另一方面，有时候你就只想说「喂，你能不能直接把我那堆事全给办了」，或者「你这周帮我处理了哪些事？给我省了多少钱？」你懂吧。明白明白。对，这确实挺有意思的。不过你看，我也觉得整个 agent 技术栈都还在萌芽期。身份、支付、营销，甚至连 CLI 还是 MCP——这些都是全新的东西，我觉得很多老套路都要作废了。对，是个全新的世界。还有，2025 年的时候我觉得 agent 被过度炒作了，但现在我觉得它是真的要来了。我也是。我知道。只是「agent」这个词挺让人头疼的，因为它被塞进了太多含义。
[26:38] Peter Yang
Yeah, there's like workflows, like all this kind of
对，又是工作流（workflow）啊，各种这类的——
[26:40] 主持人
Totally. I've been trying to just say like can we just say like model in a loop? Yeah, exactly. Model that uses tools in a loop. That's the best definition. Yeah, yeah. But nobody likes to hear that. It's a agents is much flashier, you know?
完全同意。我一直试着干脆把它叫做「循环里的模型」（model in a loop）。对，没错。「在循环里调用工具的模型」，这才是最好的定义。对对。但没人爱听这个，「agents」听着洋气多了，你懂的。
[26:49] Peter Yang
Yeah, it's flashier. Yeah, yeah. My my hope is that um all all this stuff like a lot of people I think like we're going to lose our jobs which is probably what would happen at some point but like I hope all this stuff makes just makes like human work more fun like your jobs more fun. Yeah. Dude, I don't think we're all going to lose our jobs. Like I really think that and we see this a lot of companies, you know, so we look at a ton of companies and we've seen two different buckets. So one bucket is hey we dramatically increase productivity for a person or a team. We see this in like recruiting. But we couldn't do 100% of the job. So we could do the phone screen but we couldn't obviously you know show the candidate around the office. Or we could do the phone screen and we could like answer all the questions about the company and we could even do the like comp negotiation but we couldn't do the onboarding. Yeah. The other style of company which we see which is maybe a decagon right or a happy robot is hey we did 100% of a job like customer support. Okay. You know the the customer called in they had a question we hopefully resolved their query and then that's it. Mhm. And that is 100% automated. I'd say that that second group where you have 100% automation of a job function is really rare. Mhm. Almost every AI product AI native X or Y we see is able to provide dramatic lift but it's not able to do 100%. So the last 5% is done by humans. Yeah it's still it's today anyway it's still humans that do that stuff. And it's interesting too cuz the buyer looks at that as software as expensive software whereas in the case of something like a happy robot decagon Sierra they look at it as like cheap labor. Mhm. So I do think there's a different buyer mindset but because
对，它更花哨。对对。我的希望是，所有这些东西……很多人觉得我们会丢掉工作，这某种程度上可能确实会发生，但我希望所有这些东西能让人类的工作更有意思，让你的工作更有意思。对啊。哥们儿，我不觉得我们都会丢掉工作。我是真这么想的，而且我们在很多公司里都看到了这一点。我们接触过大量公司，发现存在两种不同的情况。第一种是：嘿，我们大幅提升了某个人或某个团队的生产力。在招聘这一块我们就看到这种情况，但我们没法把这份工作做到 100%。我们能做电话初筛，但显然没法带候选人参观办公室。或者我们能做电话初筛、回答关于公司的所有问题，甚至能搞定薪酬谈判，但我们没法做入职引导。对。我们看到的另一类公司——可能是 Decagon、或者 Happy Robot 这种——是：嘿，我们把某项工作做到了 100%，比如客户支持。好。你懂的，客户打电话进来有个问题，我们但愿能解决他们的疑问，然后就完事了。嗯。这是 100% 自动化的。我得说，第二种、也就是能把一项职能 100% 自动化的，其实非常罕见。嗯。我们看到的几乎每一个 AI 产品、AI 原生的这个那个，都能带来巨大的提升，但做不到 100%。所以最后那 5% 还是由人来完成。对，至少到今天为止，那部分活儿还是人在干。而且有意思的地方在于，买家会把这看成软件、看成很贵的软件；而像 Happy Robot、Decagon、Sierra 这种，买家则把它看成廉价的劳动力。嗯。所以我确实觉得这是两种不同的买家心态，但因为
[28:14] Peter Yang
there's been this difficulty of getting to 100% automation I think a lot of the efficiency gain shows up in just a different way probably not less jobs maybe we get like the European style four-day work week maybe companies get like twice as productive. I have no idea. Yeah. But but you don't think that like I I I feel like there's going to be a transition from like these like 10,000 plus people companies laying a lot of people off to hopefully like more smaller companies like solopreneurs and stuff like that. Yeah. I I think yes I think that they'll be the the sort of shape of the economy is going to change like the amount of concentration but I just don't think there's going to be less jobs. I think human ambition has no ceiling. That's right. You know, human desire has no ceiling. And just read any mildly interesting science fiction book. Like there's no way this is the peak expression of all the stuff that we want and we need and we're going to convince ourselves and you know, all the new things that you read about every day is these luxuries, peptides and you know, everybody is going to have all of that stuff and they'll want even more. You know, dude, I I I saw a really good tweet about this. Like someone tweeted that like um the the job market is so bad that I can only pursue my dreams now or something Yes. So like it's it's like you know, it's like yeah. So maybe you lost your job, but like now you can actually do your own thing. Yeah, 100% and have a shot at actually achieving it, you know.
一直以来做到 100% 自动化都很难，我觉得很多效率提升其实是以另一种方式体现出来的——大概率不是岗位变少，也许我们会迎来欧洲那种一周四天工作制，也许公司的生产力翻倍。我也说不准。对。但是……你不觉得会有这样一个转变吗：从那些上万人的大公司大规模裁员，转向但愿是更多的小公司，比如独立创业者那种？对。嗯，我觉得会的，我认为经济的整体形态会改变，比如集中度会变。但我就是不觉得岗位会变少。我认为人类的野心是没有上限的。没错。人类的欲望是没有上限的。随便读一本稍微有点意思的科幻小说就知道——绝不可能我们现在就是人类一切所求所需的巅峰表达。我们会不断说服自己，还有你每天读到的那些新东西，那些奢侈品、各种肽（peptides）什么的，每个人都会拥有这些东西，然后想要更多。哥们儿，我看到过一条特别好的推文。有人发推说，现在就业市场太差了，差到我现在只能去追逐自己的梦想之类的。对。所以，你懂的，就好像——是啊，也许你丢了工作，但现在你反而真能去做自己的事了。对，完全同意，而且真有机会把它做成，你懂的。
[29:22] 主持人
Yeah. Yeah. Yeah.
对。对。对。
[29:23] 主持人
Cool. Well, awesome, man. Maybe that's a good positive note to end on. Yeah, that's a good note. Yeah, cool. Good seeing you, dude. Thanks, Peter. Thanks for having me.
酷。嗯，太棒了，哥们儿。也许就用这个积极的调子收尾正好。对，这个收尾不错。嗯，酷。很高兴见到你，哥们儿。谢谢你，Peter。谢谢你请我来。