构建失败机器：Zynga 之道

我想直接谈谈我发现最迷人的两页，它们相对。一页写着‘构建失败机器’，另一页写着‘甚至不要从想法开始’。我非常喜欢这两个概念。比如，我喜欢构建失败机器的概念。

I want to get right to what the two pages that I find the most fascinating and they're facing each other. One says build failure machines. The other says don't even start with an idea. And I love these two concepts a lot. Like I love the concept of building a failure machine.

是的，我也是。说起来容易做起来难，你知道。我本以为到现在有了 AI，每个人都会这么做，但我还不能说见过一个好例子。所以我认为这仍然是一座需要攻克的山。

Yeah, me too. It's easier said than done, you know. I think I would have thought by now with AI that everyone would do that and I can't say that I've even seen one good example yet. So I think that's still a hill to take.

为什么不给我们一个好例子呢？

Why don't you give us a good example?

我认为我们最好的创新来自于迫切的需求和无情的生存需要。在 Zynga 的最初，我们日复一日、周复一周地活着。如果 Facebook 平台或 MySpace 上有什么变化，我们就会被消灭。圣诞假期杀死了排名第一的游戏。它曾是 Facebook 上排名第一的游戏，叫《Jetpack Joy Ride》。人们从圣诞节回来时感受不到快乐了。它就这么消失了。所以我们意识到了这一点。这意味着每一天都很重要，我们不能犯错。所以我们会测试东西。我写了这些文本链接，放在扑克游戏的顶部，点击它们就给你一千个筹码。非常直接和俗气。我们只看点击量，不断测试直到某样东西获得大量点击。我们不会构建任何东西，直到它变得巨大。我们寻找的是大约 25%的玩家点击某样东西，然后我们才会去构建它。

I think our best innovations come out of dire need and ruthless need to survive. In the very beginning of Zynga, we were living day to day and week to week. Literally, if something changed on the Facebook platform or MySpace, we would have been wiped out of existence. Christmas vacation killed the number one game. It had been the number one game on Facebook, called Jetpack Joyride. People didn't feel the joy when they came back from Christmas. It was just gone. So we were aware of that. It meant that every day really mattered and we couldn't be wrong. So we would test things. I wrote these text links that we would put at the top of the poker game and we would offer you a thousand chips for clicking on them. It was very direct and cheesy. We just looked at the clicks and kept testing until we got a lot of clicks on something. We didn't build anything until it got huge. We were looking for on the order of 25% of players clicking on something before we were going to build it.

所以，有人曾经告诉我一个想法。你卖这些在线课程，但不知道先创建什么在线课程。所以你说，‘嘿，25 美元，这里有一个在线西班牙语课程，如果人们不点击它，你就不做这个在线西班牙语课程。’所以，这和那个想法类似吗？

So, someone once told me an idea. You sell these online courses, but you don't know what online courses to create first. So, you say, 'Hey, for $25, here's an online Spanish course, and if people don't click on it, you don't make the online Spanish course.' So, is it similar to that idea?

是的，100%。它就是一个 404 页面未找到。那就是你得到的。我们甚至没有时间放一个登陆页面，按理说我们应该放一个来登记你的兴趣之类的。只是你以为页面没加载，但它实际上完成了它该做的。

Yeah, 100%. It was a 404 page not found. That was what you got. We didn't even have time to put up a landing page, which arguably we should have, to register your interest or something. It was just you thought the page didn't load, but it actually did what it was supposed to.

你用那些测试了哪些例子？比如你放了什么文本链接？

What were example things you were testing out with those? Like what text links were you putting up there?

嗯，结果发现，这不是一个谦虚的吹嘘，我写了最成功的文本文案。我有四个姐妹和三个女儿。我似乎只是调动了我内心的少女，她似乎是当时 Facebook 和 MySpace 上参与度最高的人群。我会拼错单词。我会语法不好。我只是很快地输入一些东西，就像发短信一样。现在我想人们看到会觉得俗气，但它就像有人在告诉你一个新游戏，一个新单词游戏，或者像你的朋友一样。我总是说，我总是在寻找一条血管。每个人都如此厌倦，你在寻找一个‘OMFG’时刻，就像寻找一条血管。你只是放越来越疯狂的东西，直到有人开始点击它，然后它就像一个完美的交易胜利。你可以据此设定你的销售额，因为一旦 22.8%的人点击，22.8%的人总是点击它。它不会改变。这是消费者的美妙之处。只要有足够的数据，唯一改变的是洛杉矶选举结果，但通常消费者数据保持一致。

Well, it turned out, and this is a not so humble brag, that I wrote the most successful copy on the text. I have four sisters and three daughters. I just seemed to channel my inner teenage girl, who seemed to be a lot of the most engagement on Facebook at that time and MySpace. I would misspell things. I would have bad grammar. I would just kind of really quickly type something in like it was a text message. Now I think people see that and it's cheesy, but it would be like somebody telling you about a new game, a new word game, or like your friend was. I always say I'm always trying to find a vein. Everyone's so jaded that you're trying to find an OMFG moment, like trying to find a vein. You just put crazier and crazier until somebody starts clicking on it and then it's like a perfect trade win. You can set your sales by it because once 22.8% click, 22.8% always click on it. It doesn't change. That's a beautiful thing about consumer. With enough numbers, the only thing that changed was the LA election results, but normally consumer numbers stayed consistent.

那么，具体是什么文本例子呢？比如你具体说了什么？

Well, what was an example text though? Like what were you specifically saying?

我不确定我是否记得 2007 年的文本，但我会说我当时也发了这些非常成功的邮件。那是邮件还管用的时候。我为扑克游戏在愚人节做了一件事，我发了一封非常成功的邮件，内容大概是，很俗气，但就像‘你刚刚赢得了一万亿扑克筹码’。我不记得具体了，但我就是试着想出奇怪疯狂的不着边际的东西。我对消费者的理解之一是，要震惊和惊喜人们非常困难，因为我们太习惯每一条营销信息了。这就是为什么拼写错误管用。那些让人感觉背后有人的东西。它必须出乎意料。它必须看起来不对劲。我们在看模式，我们的大脑，我们看到一个预期的模式越多，我们就越忽略它。所以即使它没有意义，比如 101 号公路去机场的广告牌，如果你放一个不太有意义的营销口号，它会在人们脑海中盘旋。它曾经是一个好口号，你记得吗？现在它只是一个奇怪的口号。他们为什么要那样做？但我的观点是，失败机器的要点是建立一个非常一致的方式，你可以有效地、以最快最省力的方式，每天测试 100 个或更多的想法。我们的口号是，一周内测试的想法比整个行业一年测试的还多。这并不难，因为 2007 年唯一比与游戏行业竞争更好的事情，可能就是与 NASA、国防工业或政府竞争了。

I don't know if I can remember my texts from 2007, but I'll say that I also sent these really successful emails. That's when emails worked. One thing I did for poker on April Fool's Day, I sent a really successful email saying something like, it was cheesy, but it was like, 'You just won a billion poker chips.' I don't remember exactly, but I would just try to come up with weird crazy non sequitur things. Part of what I've understood with consumers is that it's really hard to shock and surprise people because we're just so used to every marketing message. That's why misspellings work. Things that made it feel like there was a human there. It just had to be unexpected. It had to be like something that doesn't seem right. We're looking at patterns, and our brains, the more we see a pattern that we expect, the more we ignore it. So even if it doesn't make sense, like a billboard on 101 to the airport, if you put up a marketing slogan that just doesn't quite make sense, it'll roll around in people's minds. It used to be a good slogan, you remember? And now it's one that is just weird. Why would they do that? But my point was that the failure machine point was to put up a really consistent way that you can just effectively, in the fastest least effort way, be testing 100 ideas a day or more. Our mantra was test more ideas in a week than the industry tests in a year. That wasn't hard because the only thing better than competing with the game industry in 2007 was probably like competing with NASA or the defense industry or the government.

是的，因为我觉得在 Zynga，你基本上可以一直测试所有游戏上的许多新功能。在 Facebook 上的巅峰时期，你同时有多少款游戏在运行？

Yeah, because I feel like with Zynga you were able to basically test lots of new features on all of your games all the time. How many games did you have going at any single moment at your peak on Facebook?

这是个好问题。在我们的巅峰时期，我们大概有 14 个全球工作室，每个工作室可能有一个主要系列，然后还有几个其他游戏。我们还在印度有一个工作室，那是我们送所有游戏去死的地方。

It's a good question. At our peak, we probably had 14 global studios and each studio probably had one major franchise and then a couple other games. We also had a studio in India, which is where we sent all of our games to die.

任何日收入低于 10 万美元的游戏，我们都会发到印度去。它们变得非常有创新性，实际上成了创新中心，因为当没人关心时，会发生一件很棒的事：你可以只管尝试，没人会阻止你。所以我们得到的意外收获是，我们把游戏发到印度去等死，结果它们突然开始增长。印度就是我们发现订阅模式有效的地方。它们真的成了创新中心，只是因为，一旦游戏在我们看来无关紧要，人们就可以随心所欲地折腾。而像蒂芙尼这样的品牌就很难创新，团队会害怕。

Any game that got below $100,000 a day in revenues, we shipped off to India. And they were amazingly innovative. And they actually became a center of innovation because there's a great thing that happens when nobody cares. You get to just try and nobody's going to stop you. And so the unexpected things we got were we'd ship these games off to India to die and all of a sudden they'd start growing. India is where we found out that subscriptions worked. They actually became a real center for innovation and it was just because, you know, once the game was, in our book, inconsequential. That's when people can just do whatever they want with it. When it's very hard to innovate on like your Tiffany brand, the team gets afraid of it.

是的。我觉得在游戏行业，你可以做那种测试，尤其是在线游戏，因为你可以实时更改并观察用户行为。你还看到哪些公司快速试错？也许不是每天几百个想法，但每年很多想法。

Yeah. And I feel like in the gaming industry, you can do that kind of testing, particularly online gaming, because you can make changes on the fly and see human user behavior. What other companies do you see kind of failing fast? Maybe not hundreds of ideas a day, but lots of ideas a year.

我要挑战你刚才说的。过去在游戏行业，你可以测试大量想法。但实际上我们倒退了，因为从网页游戏——对我来说那是游戏创新的真正复兴——转向了 Unity 上的移动游戏开发，这很像老游戏行业，因为它是编译代码。必须上传到应用商店，下载。你可以 A/B 测试文本和小部分内容，但功能测试在开发中已经大大放缓并倒退。心态上，甚至不值得再开每周路线图会议，因为团队现在一周内改变游戏太难了。所以改为每月路线图会议。从每周节奏变成了每月。这让我很难过。但还有谁在创新？我们看到 Facebook，或者抱歉 Meta，我还是叫他们 Facebook，他们雇了很多 Zynga 的人。他们雇了我们整个数据科学团队，所有产品经理，Meta 已经成为测试和数据驱动方法的卓越中心，但他们似乎更关注推动参与度和变现的货币化方面，而不是面向消费者的功能。我看到 Duolingo 做了一些很酷的测试和引入游戏机制，但我不确定。你看到谁了？我很难指出任何人。

I'll challenge what you just said. It used to be that in the gaming industry, you could test tons of ideas. We've actually gone backwards because now we've gone from web gaming, which to me was the real renaissance in game innovation, to mobile game development on Unity, which is a lot like the old game industry because it's compiled code. It has to be uploaded to the app store, downloaded. You can A/B test text and little bits of content, but features, the feature testing in development has really slowed down and moved backwards. And the mentality, it's not even worth having a weekly roadmap meeting anymore because it's too hard for a team to change a game in a week at this point. So you have a monthly roadmap meeting. It's moved from a weekly cadence to monthly. That's sad to me. But who else innovates? I mean, we do see Facebook, or sorry Meta, I still call them Facebook, they hired so many Zynga people. They hired our whole data science team, all of our PMs, and Meta has become a center of excellence for testing and data-centric approaches, but they seem to focus much more on monetization things that drive engagement and monetization and less on consumer-facing features. I've seen Duolingo do some cool stuff testing and bringing out game mechanics, but I don't know. Who have you seen? I have trouble pointing to anybody.

是的，我觉得 AI 会做到，而且我在质疑你之前的观点，你说你没看到 AI 做到这一点，但我觉得 AI，因为你可以在几小时内编程出应用，不是玩具应用，而是有意义的应用。用这些 AI 编码工具，你真的可以在几小时内编出有趣的东西。这应该是一个人们可以快速失败的地方。你可以每天发布一个应用，这在 30 年前可能是一个完整的业务。你可以用 AI 编码工具每天发布、测试并扼杀一个应用。但你说你没看到有人这么做，但这可能只是开始。

Yeah, I feel like AI will, and you know, I'm questioning your earlier point that AI, you haven't seen AI do this, but I feel like AI because you can program apps in a few hours, like significant apps, not like toy apps. You can really program interesting things in just a few hours using these AI coding tools. This should be a place where people fail very quickly. You can launch an app a day that could formulate that 30 years ago would have been a full business. You can launch, test, and squash an app a day potentially using AI coding tools. But you're saying you haven't seen people do that, but it's probably just the beginning.

嗯，这只是开始。但你说你可以每天发布、测试并失败一个应用，不是在应用商店里吧？我的意思是，你可以临时拼凑让它上 TestFlight，不是真正发布，但让它可用。所以这是可能的。你可以作为网页应用来做，但问题是你去哪里找人试用？这就是为什么很多人回到了 Discord，你可以在 Discord 社区里测试东西，我也做过一些。但我们真正需要的是一个分发沙盒。我的意思是，我希望看到 ChatGPT 或 Gemini。我希望看到某个大型 LLM 门户创建一个应用沙盒。我认为一旦我们看到那个，我们会再次看到测试和创新的爆发。

Well, it is just the beginning. And when you say you can launch, test, and fail an app a day, not in the app store, right? I mean, you could jerry-rig it to be on TestFlight and not exactly launch it, but have it turned on. So that's possible. You can do it as a web app, but there's a question of where are you going to get people to try it? And that's why a lot of people have gone back to Discord so you can kind of test things in Discord communities, and I've done some of that. But we really need, well you need like a distribution sandbox. I mean, I wish that we saw ChatGPT or Gemini. I wish we saw one of the big LLM portals create an app sandbox. And I think once we see that, we'll see an explosion in testing and innovation again.

但我觉得病毒式传播的本质是指数级的。所以你在 Twitter 上发一个链接，两个用户分别展示给两个朋友，如此反复，只需几次迭代你就有一百万用户，还是我抽风了？

But I feel like the nature of virality is that it's exponential. So you post a link on Twitter and okay two users show it to two friends each and on and on and just in a few iterations you have a million users, or am I smoking crack here?

你抽风了。

You're smoking crack.

不是第一次了。好吧，这算是好货。或者告诉我我漏了什么，也许我是在妄想。你能想到最近哪个应用病毒式爆发，或者成为应用商店前十甚至前二十五？嗯，不是应用商店，但像 ChatGPT，我觉得无论他们在哪里发那个链接，都会很快病毒式传播。

Not the first time. Or okay. Well, it's some good crack. Or I guess tell me what I'm missing because maybe I'm delusional. When's the last app that you can think of that blew up virally or that became the last time an app became a top 10 or even top 25 app in the app store? Well, not the app store, but like ChatGPT as an example where I think wherever they had posted that link, it would have gone viral pretty quickly.

好吧，但让我们想想那个例子。你只需要 100 亿美元和世界顶尖的 AI 工程师，还有八年的开发时间，然后你就能一夜之间病毒式成功，你可以把链接发到任何地方，它都会有效。我同意你。我很想上那趟车，但这对我们所有人来说都是一个很难效仿的例子。所以我不想泼冷水，但我认为目前消费者领域不可投资，已经死水一潭。我的意思是，现在为什么有人会做消费者领域，当他们显然可以构建企业级或至少是专业消费者级产品，获得一个较小的用户群，这些用户愿意付很多钱，而且他们已经有兴趣寻找？要进入大众市场非常难。我的意思是，他们不下载新应用。他们不在 Twitter 上找新应用试用。所以我不是说不可能，但我要告诉你，我的经验法则，当一个平台在消费者层面可投资时，我称之为 9010。我想我忘了把这个写进书里。我需要给书加个附录，因为我不断意识到……

Okay, but let's think about that one. All you need is $10 billion and the top AI engineers in the world and I don't know, eight years of development and then you could become an overnight viral success and you can post your link anywhere and it'll work. I'm with you on that. I would love to be on that train, but that's a tough example to point to for all of us to follow. So, I don't want to pour water on this, but I think at the moment consumer feels uninvestable and it's dead in the water. I mean, why would anyone do consumer now when they can clearly build enterprise or at worst prosumer and get a smaller base of users that are willing to pay a lot of money and they're already interested in looking? It's pretty hard to get to the mass market. I mean, they're not downloading new apps. They're not looking on Twitter for new apps to try. So, I'm not saying it's not possible, but I'll tell you this, my rule of thumb for when a platform was investable at a consumer level, I used to call it 9010. And I think I forgot to put this in my book. I need an addendum to the book because I keep realizing that...

是的，已经过时了。别买。别读。这是我的营销信息。所以我会说别买我的书。已经过时了。

Yeah, it's already out of date. Don't get it. Don't read it. That's my marketing message. So, I'll say don't buy my book. It's already out of date.

那会像什么？

That'll be like what?

我的经验法则，我称之为 9010 法则。如果一个平台能让一个应用在 90 天内达到 1000 万日活跃用户，那么它就是可投资的。所以我会看平台，无论是社交网络还是后来的移动平台，说当我们看到至少一个应用能在 90 天内达到 1000 万日活跃用户时，那就是可投资的。现在听起来我像在另一个星球。我的意思是，今天没有平台能做到这一点。这是真的，因为首先，像 Twitter 和 Facebook 这样的平台可能会限制病毒式传播。

My rule of thumb, I'd called it the 9010 rule. And it was if a platform enabled an app to get to 10 million DAUs in 90 days, it was investable. And so I'd look at platforms and say whether it was social networks or eventually mobile and say at the point that we can see at least one app that could get to 10 million DAUs in 90 days, that's investable. Now it sounds like I'm on another planet. I mean, there's no platform today that enables that. It's true because you see, first off, platforms like Twitter and Facebook might limit virality.

你知道，他们不一定希望事情病毒式传播，或者正如你指出的，应用商店太拥挤了，发现新应用真的很难。

You know, they don't want necessarily things to go viral or the app store is, as you point out, it's just too crowded and discovery is really hard in the app store.

想想我们的心态。我是说，你今天下载了多少新应用？昨天呢？

Think about our mentality. I mean, how many new apps did you download today? How many did you download yesterday?

零。

Zero.

是啊。上周你下载了多少？

Yeah. How many did you download last week?

不，你说得对。我有了所有我需要的应用。

No, it's a good point. I have all the apps I want.

没错。你不在发现模式。所以我们忘了这就像把巨石推上山，对吧？当我们最初几年有手机时，我们会说，‘哦，那是什么？一个新时钟。酷。’对吧？你在酒吧，朋友给你看个傻东西，你感兴趣。现在，如果他们问‘你看到这个新应用了吗？’你会觉得是个笑话，你会说，‘什么？’没人醒来会去找。事实上，我不确定这些数据完全准确。我希望有实时事实核查，就像 AI 在哪里？这个应用在哪里？但去年应用商店推出了大约 4 万个应用，零个进入前十，也许 4 万个中有一两个进入前 25，但也没能保持住。所以这对苹果来说是个问题。他们就是不明白。一个应用平台需要好的……他们需要两样东西。这就是为什么 Roblox 刚刚错过了收益，我认为这就是原因。要成为一个真正充满活力的应用平台，你需要两样东西。一是需要长期特许经营权。想想《黑道家族》对 HBO 有多重要。对于像 Netflix 这样的流媒体网络，如果 Netflix 没有我们在意的特许经营权，他们就不能保证我们会回来。没错。就像《黑道家族》、《权力的游戏》下一季让你回到 HBO。当他们不再拥有这些时，他们的观众就面临风险。他们还需要新颖刺激的东西。他们两者都需要。Roblox 去年表现很好。他们有一些应用达到了巨大的同时在线用户峰值。Roblox 的同时在线峰值从大约 500 万（与 Steam 相当）上升到大约 2500 万。这完全是因为一个叫《种花园》的游戏。所以他们有一个特许经营权。就像《农场小镇》。然后那个东西垮了。然后又有另一个特许经营权，但更小。另一个更小。现在他们有点麻烦，因为他们没有任何特许经营权，只有这些人们玩几天的迷因游戏。他们并非完全陷入困境，因为他们也算是一种社交网络。但我的观点是苹果——我不会说苹果有麻烦——但我认为他们因此更脆弱了。

Right. You're not in a discovery mode. So, we forget that it's like pushing a boulder up a hill, right? And when we all had our phones for the first couple years, we were like, 'Oh, what's that? That's a new clock. Cool.' Right? You're in a bar and your friend shows you some dumb thing and you're interested. Now, it would be a joke if they said, 'Oh, did you see this new app?' You'd be like, 'What?' Nobody wakes up looking. In fact, I don't know if I have these stats exactly right. I wish we had real-time fact-checking like where is AI? Where's this app? But last year there were like 40,000 apps launched in the app store and zero became top 10 and maybe one or two out of 40,000 made it to top 25 and then they didn't hold the position. So it's a problem for Apple. They just don't get it. An app platform needs good... they need two things. And this is why Roblox just missed their earnings and I think this is why. You need two things to be a really vibrant app platform. One is you need long-term franchises. Think about how important The Sopranos was for HBO. For streaming networks like Netflix, if Netflix had no franchises that we cared about, they don't have a guarantee we're coming back. Right. Like The Sopranos, the next season of Game of Thrones makes you come back to HBO. And when they stop having that, their viewers are at risk. And they need spicy new. They need both. Roblox was crushing it last year. They had these apps that got up to enormous peak concurrent users. Roblox saw their peak concurrency go from something like five million, on par with Steam, to something like 25 million. And it was all because of one game called Grow a Garden. So they had a franchise. It was like Farmville. And then that thing fell over. Then there was another franchise but that was smaller. And another one was smaller. Now they're kind of in trouble because they don't have any franchises and they just have these meme games that people play for a couple days. They're not totally in trouble because they're also kind of a social network. But my point is Apple—and I wouldn't say Apple's in trouble—but I think they're more vulnerable because of this.

这很有趣。所以问题似乎就像你说的分发问题，一切都变得小众化。所以每个人都对自己的东西感兴趣。没有一个地方像电影院那样，全镇的人都去那里。

This is very interesting. So it seems like the problem is like you were saying with distribution, like everything is niched down. So everybody's interested in their own thing. There's no one place where here's the movie theater. Everybody in town goes to this.

我称之为鸡尾酒会，对吧？这也在我的书里，你不应该买——或者你应该偷。偷自阿比·霍夫曼的《偷这本书》。我不在乎，但我的出版商可能在乎。所以，这是一种长期的本能脉络。在鸡尾酒会上。我认为它始于 Napster，那是我们第一次在网络上自我聚合。第一次，没有巴里·迪勒或某个付费专业数据库在中间，我们只是彼此连接。就像海盗流氓的东西，有点像互联网上的火人节。很酷。你感觉到有别的什么事在发生。我现在连接着 450 万台其他电脑，中间没有人。对我来说，这是整个社交媒体和这个人与人（而非页面）网络的开始。它并没有真正聚集；没有中心，但它是一个新版本的公告板，只是通过音乐文件连接我们所有人。在我看来，这导致了 Friendster、LinkedIn、Facebook 等等。但那些是我们聚在一起的鸡尾酒会。然后手机是鸡尾酒会吗？不完全是。我们有点碎片化了，不再真正有鸡尾酒会。现在就像我们在哪里闲逛？我们在 GPT 和 Claude 上闲逛，但那是单人游戏，对吧？

I call that the cocktail party, right? This is also in my book, which you shouldn't buy—or you should steal it. Steal from Abbie Hoffman's 'Steal This Book'. I don't care, but my publisher might. So, this is a long-term instinct vein. In the cocktail party. I think it started with Napster, which was the first time that we all just self-aggregated on the web. The first time that we all, without Barry Diller or some paid professional database in the middle, just connected to each other. It was like this pirate rogue thing, kind of like Burning Man for the internet. And it was cool. You had a sense of something else going on. I'm connected to 4.5 million other computers right now with nobody in the middle. To me that was the beginning of this whole social media thing and this network of people, not pages. It didn't really congregate; there was no center to it, but it was this new version of a bulletin board just connecting all of us around music files. That in my mind led to Friendster, LinkedIn, Facebook, all that. But those were cocktail parties that we came together. Then the mobile phone was a cocktail party? Not exactly. We've kind of fragmented and we don't really have a cocktail party. Now it's like where are we hanging out now? We're hanging out on GPT and Claude, but that's a single-player game, right?

他们还没搞明白社交。他们甚至有点在阻止社交。

They haven't figured out social. They're even kind of stopping social.

我不知道为什么，但我猜是因为他们不需要。现在他们的重点是更多算力，然后更多，你知道，追求编程机会，这很清楚明显。如果他们不这样做并赢得胜利，就像一系列预赛。如果他们不赢得那场冲刺，他们就无法进入下一场。所以即使他们做到了，他们也需要担心消费者参与度，但这并不重要。他们需要赢得这场算力扩展和收入战争，然后才能考虑抓住我们的心和思想。

And I don't know why, but I guess because they don't need it. Right now their game is more compute and then more, you know, go after the coding opportunity and that's pretty clear and obvious. If they don't do that and win that, it's like a series of heats. If they don't win that sprint, they don't get to be in the next one. So even if they get that, they're going to need to worry about consumer engagement, it just won't matter. They need to win this computing scaling and revenue war before they get to even think about holding our hearts and minds.

听着，大多数现有的客服机器人都是 15 年前的。它们很糟糕。你总是试图联系真人。我可以从个人经验告诉你，Decagon 不是那样的。听着，如果你的团队要处理更多的支持工单、更多的渠道和更高的客户期望，你可能考虑过使用 AI 进行客户服务，但大多数现有选项要么太有限，要么设置时间太长，要么就是老派。它们效果不好。这就是为什么我很兴奋地告诉你关于 Decagon。去看看吧。如果你有公司并且需要客户服务，就去看看。Decagon 帮助公司通过 AI 智能体在聊天、电子邮件、语音和短信中创建个性化的礼宾式客户体验。它们全天候可用。它们交谈起来很自然，可以自行解决客户请求，这样企业就能跟上请求而不失去个人触感。Decagon 直接插入你现有的系统，所以智能体可以快速上线。而且因为工作流可以用自然语言更新，团队可以自己进行更改，无需漫长的工程周期。这是关键。如果你在交互中发现什么不对，你只需告诉它，‘不，不，不，这样说。’因为它使用的技术与所有这些 AI 编码工具相同。就像你在实时编码客户服务。这意味着花更少的时间管理工作流，更多的时间关注客户，你将在几周内看到投资回报率。

Look, most of the customer service bots out there are 15 years old. They're horrible. You're constantly trying to reach a real person. I can tell you from personal experience that Decagon is not like that. And look, if your team is dealing with more support tickets, more channels, and higher expectations from customers, you've probably thought about using AI for customer service, but most of the options out there are either too limited or take way too long to set up or are just old school. They don't work that good. And that's why I'm excited to tell you about Decagon. Just check it out. If you have a company and you need customer service, just check this out. Decagon helps companies create personalized concierge-style customer experiences with AI agents across chat, email, voice, and SMS. They're available 24/7. They feel natural to talk to and can resolve customer requests on their own so businesses can keep up with requests without losing their personal touch. Decagon plugs directly into your existing systems so agents can get live quickly. And because workflows can be updated using natural language, teams can make changes themselves without long engineering cycles. This is key. If you notice something in the interaction that's wrong, you just tell it, 'No, no, no, say it this way.' Because it's using the same technology as all these AI coding tools. It's like you're coding the customer service in real time. That means less time spent managing workflows, more time focused on customers, and you'll see the ROI in weeks.

说得好。当然，如果你天马行空地想象，AI 平台上有很多社交媒体的机会，但它们不需要玩那个游戏。而且应该提到，你当时就在那里。你就像社交媒体教父一样。比如 Tribe，我认为 Tribe 是第二个社交媒体平台。GeoCities 可能是第一个，由你的朋友 Fred Wilson 投资，以数十亿美元卖给雅虎，然后你在 MySpace 之前创办了 Tribe，当然 Facebook 后来爆发了。你觉得是什么让 Tribe 没有成功？它一开始算是成功的，每个人都听说过 Tribe。比如我听说过，但我没用过。为什么没有……

It's a good point because there is, of course, if you let your imagination run wild, there are a lot of opportunities for social media on these AI platforms, but they don't need to play that game. And it should be mentioned you were there. You're like the godfather of social media. Like Tribe, I would argue Tribe was the second social media platform. GeoCities maybe the first one funded by your buddy Fred Wilson and sold to Yahoo for billions and then you started Tribe right before like MySpace and then Facebook of course blew up. What do you think made Tribe not a success? It was sort of a success at first like everybody had heard of Tribe. Like I had heard of it, but I wasn't using it. Like why didn't...

问题就在这儿。问题就在这儿。

That's the problem. That's the problem.

嗯，为什么我没有……我甚至没怎么用过 MySpace，但我在能用 Facebook 的时候就用得很猛。

Well, why didn't I... I didn't even really use MySpace, but I did use Facebook very aggressively right when I was able to.

Tribe 是一个重要的反面案例。它完美诠释了我的理论：相信你获胜的本能，但不要相信你的自负，要像穿白大褂的科学家一样做经过验证的新事物。因为 Tribe 本应非常成功。我的意思是，今天的人很难想象。那时候你得刻意让自己失败。不只是 Friendster、Facebook、MySpace。还有 Tagged、Bebo。每个社交网络都挺管用。人们就是喜欢。你记得你收到一封邮件说‘Mark 想成为你在 Tagged 上的朋友’之类的。你会想‘哦，酷，那是什么？’然后点进去。这些邮件的打开率有 80%，点击率有 50%。今天的人无法想象的数字。他们不会相信的，对吧？所以，对于 Tribe，我有三个获胜的本能和一个失败的想法。而我就像英雄一样，坚忍地坚持着那个失败的想法。我忽略了指标。它就像一艘正在下沉的快艇。它超级病毒式传播，增长比 LinkedIn 快得多。就像龟兔赛跑。Reed 有 LinkedIn，LinkedIn 只是慢慢爬升，没有参与度。而 Tribe 是一个喧闹的派对，是火人节。它每天获得的安装量多得多，但没有留存。最终只剩下一小群热爱它的人。而我在信任上搞错了。Friendster、Facebook 和 LinkedIn 都做对了信任。我没有意识到人们只是把自己放在公共网络上，尤其是主流人群，尤其是女性，对陌生人危险、对不认识的人联系他们感到不舒服。而我让 Tribe 成为一个开放、喧闹的平台。我这么做的原因是我专注于 Craigslist 和分类广告业务。我需要人们能够互相联系，买卖沙发、找室友和工作——至少我是这么认为的。我如此专注于这个分类广告业务和 Craigslist。我想，好吧，我通过社交网络和病毒式传播获取用户，然后把他们转移到工作和分类广告业务中。我忽略了用户真正喜欢的东西，也就是部落。所以，城市部落和通过松散联系的群体进行社交的想法是金矿，而我完全忽略了它。有巨大的参与度和热度。我几乎没给他们什么功能，它却自己起飞了。很多人有故事说通过它遇到了妻子，或者通过他们的 Mission Bay 小组去了巴西的奇妙旅行。我没有走 Paul Graham 的路线：先找到 100 个开心的用户，然后继续构建他们想要的东西。我有所有这些开心的用户，而我却忽略了他们。而且他们也是外向的人，发了很多不雅照，你知道，外向的人就是这样。

Tribe is a major case study in what not to do. And it's a perfect case study in these theories I have of trust your winning instincts but not your ego, and do proven better new like a scientist with a white lab coat. Because Tribe should have been huge. I mean, it's hard for people to imagine today. You had to will yourself into failing at that point. Like, there wasn't just Friendster, Facebook, MySpace. There was Tagged, Bebo. Every social network kind of worked. People were just into it. You remember you got an email saying, 'Mark wants to be your friend on Tagged or whatever.' And you're like, 'Oh, cool. What's that?' And you'd click on it. These had 80% open rates, 50% click-through rates. Numbers that people can't imagine today. They wouldn't believe it, right? So, with Tribe, I had these three winning instincts and one losing idea. And I just heroically, stoically stuck with it. And I was ignoring my metrics. It was a sinking speedboat. It was super viral. It grew way faster than LinkedIn. Like it was the tortoise and the hare. Reed had LinkedIn and LinkedIn just kept slowly ticking up and had no engagement. And Tribe was this rowdy party. It was Burning Man. It was getting way more installs a day, but it had no retention. It boiled down to a very small group of people who loved it. And I got trust wrong. Friendster and Facebook and LinkedIn all got trust right. I didn't get that people were just putting themselves out on the public web and especially mainstream people, especially women were not comfortable with stranger danger, with people they didn't know contacting them. And I made Tribe this open, rowdy platform. The reason I did it was because I was focused on Craigslist and the listings business. And I needed people to be able to contact each other to buy each other's couches and find roommates and jobs, or I believed I did. And I was so focused on this listings business and Craigslist. And I thought, okay, I'm going to acquire my audience through social networking and virality, but then I'm going to move them into this jobs and listings business. And I ignored what my users were into, which was the tribes. So the idea of the urban tribes and networking through your loose ties through groups that you're loosely a part of, that was gold and I completely ignored it. There was huge engagement and heat. I barely gave them any features and it just took off. And so many people have stories about meeting their wife through that or this amazing trip to Brazil through their Mission Bay group. And I was not on the Paul Graham train of get a hundred happy users and then keep building what they want. I had all these happy users and I was ignoring them. And they were also extroverts posting a lot of dick pics, you know, as extroverts do.

是的。

Yes.

以前没有这样的地方……嗯，现在有了，我想现在有 OnlyFans，而且围绕它有一个不错的生意。但这些没人想要的社区最终出现在 Reddit 和其他地方，它们最终找到了如何把它变成生意的方法。但社区在互联网上总是超额交付，因为人们很孤独，如果他们能在网上找到新朋友，他们就会，然后他们会紧紧抓住他们。这在所有 Zynga 游戏中都是如此，在 Tribe 中也是如此。但它与我的商业模式不完全一致，也与主流人群的留存不一致。所以，我写进书里的许多理论，以及我的产品管理、产品思维方法，都来自 Tribe 漫长而痛苦的失败。那是我的学习场。太棒了。

There's no place on the... well now there is, I guess now there's like OnlyFans and there's a good business around it. But these communities that nobody wants ended up on Reddit and other places, and they ended up figuring out how to turn that into a business. But community always overdelivers when you enable it on the internet because people are lonely and if they can find new friends on the internet, they will and then they'll hang on to them. And that was true in all Zynga games and that was true with Tribe. But it didn't align perfectly with my business model and it didn't align with the retention of mainstream people. So, many of my theories that I built into my book and my approach to product management, product thinking came from the long painful failure of Tribe. It was my learning grounds. It was amazing.

关于信任这一点很有意思。比如 Facebook 当然有.edu 域名，所以你知道，嘿，这些人跟你一样在上大学，哦，这些人也在哈佛，跟你一样。所以你可能会遇到他们。所以当有信任时，违反礼貌的情况就少了。

It's interesting about the trust. Like Facebook had, of course, the .edu so you knew, hey, these people are going to college just like you are and oh these people are also at Harvard just like you are. So you might run into them. So there's less violation of politeness when there's trust.

嗯，那是你的部落，对吧？他们在同一所学校。他们超越了一个最低界限，有真实身份。他们是这个更广泛社区的一部分。事实证明这有很大的不同。多年后，我发现了 Raya，我和 Reed 一起，我们为 Raya 投入了所有资金。Raya 也在其他约会应用忽略的方面做对了信任，那就是人工策展级别，一个像 Soho House 那样的委员会必须审核你。你必须有一个真实的 Instagram 账号，由人工检查并接受或拒绝你。所以它让……我喜欢说约会应用是 1/10 的体验，Raya 是 3/10，它只是糟糕的程度少了三倍。它消除了陌生人危险或那种 creepy 的感觉。另一件我想回到的事情是，这与我们今天思考消费者如何被重塑、社交如何被重塑非常相关。这一切的开始，当 Reed 和我在 2002 年花了很多时间头脑风暴他所谓的 Web 2.0 时，就是潜在客户生成。好吧，这听起来可能违反直觉。

Well, it's your tribe, right? They're at your same school. They're beyond some minimum boundary that they have a real identity. They're part of this broader community. And that turns out to make a big difference. Years later, I discovered Raya and I, along with Reed, we put up all the money for Raya. And Raya got trust right also in a way that the rest of the dating apps missed, which was this human curation level that a committee just like Soho House had to vet you. You had to have a real Instagram account and humans checked it and accepted you or didn't. And so it made... I like to say that dating apps were a 1 out of 10 experience and Raya was a 3 out of 10 and it just sucked three times less. It just took the stranger danger or that creepiness out of it. And the other thing I want to circle back to because this is so relevant today as we think about how is consumer going to be reinvented? How is social going to be reinvented? The beginning of all this when Reed and I really spent all this time in 2002 brainstorming what he called Web 2.0 was lead generation. Okay, this may sound counterintuitive.

这不像你想象中 Instagram 或鸡尾酒派对的样子，但它的实用性就在那里。你当然能获得娱乐，这也是你来这里的原因。但一个好的鸡尾酒派对带来的价值往往是潜在客户开发。这种社交偶然性会带来旅行推荐、工作机会或约会——你无法预知会得到什么，但你周围是一群有趣的人，有些认识，有些不认识，但经过精心策划，它带来了更高的潜在客户开发信号，这正是 Reed 在 LinkedIn 上做得很好的地方。当你思考 LinkedIn 将如何被重塑？Instagram 将如何被重塑？它必须改变这种关于信噪比和时间回报的范式。在这个 AI 智能体时代，它也有巨大的机会，因为几乎就像重新发明 Napster 与 LinkedIn 的结合。我们的智能体可以一直参加鸡尾酒派对，进行社交，并穿越信任、访问和信息的膜，从而为我们带回潜在客户。

It's not the way you think about Instagram or a cocktail party, but it's the utility. You get entertainment for sure and that's the reason you come there. But the value you get out of a good cocktail party is often lead generation. It's this social serendipity that leads to a recommendation on a trip you end up going on or a job lead or a date or you know you don't know what you're going to get but you're around a bunch of interesting people some you know some you don't but it's really well curated and it leads to much higher signal for lead generation and that's what Reed nailed with LinkedIn. When you think about how will LinkedIn be reinvented? How will Instagram be reinvented? It's got to change this paradigm on signal to noise and the return on your time. In this AI agentic world, it has a huge opportunity too because it's almost like reinventing Napster meets LinkedIn. Our agents could be going to cocktail parties all the time, networking and navigating this membrane of trust and access and information in order to bring us back leads.

有趣的是，这归结为潜在客户。我的意思是，同样地，现在你提到应用市场是一个东西不会病毒式传播的地方。有一个非常引人入胜的统计数据：去年推出了 4 万个应用，但没有一个进入前十。如果每个人都去自己的人工智能策划的应用市场呢？不一定只是移动应用，而是网络应用、移动应用、iPhone 应用等等。但这就是我的人工智能策划的应用商店。

That's interesting that it boils down to leads. I mean, you can similarly though right now you've mentioned like the app marketplace is a place where things aren't going viral. It's really fascinating statistic that 40,000 apps launched last year and none made it to the top 10. What about if everybody went to their own AI curated app marketplace? Not necessarily just mobile apps but web apps, mobile apps, iPhone apps and so on. But it's just this is my AI curated app store.

嗯，首先你得对应用有需求，对吧？

Well, first you got to have a desire for apps, right?

是的，没错。

Yeah, that's true.

没错。所以，这很有趣，因为 Facebook 就像是你不知道自己想要的应用。我没有意识到，哦，内心深处我想和二年级的朋友保持联系，看看他们的孩子昨晚的足球比赛怎么样，而我已经 30 年没和他们说话了。但现实是，你有这些生活的考古层，Facebook 让你能够深入挖掘，但这些现在已经不存在了。就像没有人真正再用 Facebook 了。

Right. So, it's funny because Facebook is like the app you didn't know you wanted. I didn't realize, oh, deep down I wanted to stay in touch with my friends from second grade and see how their kids did in their soccer game last night when I hadn't spoken to them in 30 years. But the reality is you had these archaeological layers of your life that Facebook allowed you to drill down on which doesn't really exist anymore. Like nobody really uses Facebook anymore.

不，我的意思是，我甚至一个月上一次 Instagram，而我交谈的大多数不使用 Instagram 的人都会带着一种自豪感说出来，就像有一种奇怪的东西你可以追踪。你可以看看净推荐值（NPS）分数。你知道净推荐值吗？

No, I mean I don't even get on Instagram once a month and most people I talk to who don't use Instagram say it with a level of pride and it's like there's this weird thing that you can track. You can look at the NPS scores. Do you know about net promoter score?

不知道。

No.

好的。这在消费者领域曾经是一件大事，但净推荐值是一个衡量品牌力量的非常简单指标。某人向他人推荐你的服务的可能性有多大？eBay 在巅峰时期和亚马逊达到了 80，所以 80%的人感觉积极并会推荐。你可以有正面或负面。负面的 NPS 意味着人们积极告诉别人不要使用它。对吧？就像有线电视公司、民主党、也许共和党、可能任何政治家、国税局。对吧？你会看到一个奇怪的现象，我们在 Farmville 上看到过，然后在 Facebook 和 Instagram 上又看到了。Farmville 在人们玩的时候大约有正面 40 的 NPS。但之后他们停止玩了，就像戒了烟一样，我们得到了负面 35 的 NPS。

Okay. It was a big deal for a while in consumer but the net promoter score is a very simple measure of brand power. What's the likelihood that someone would recommend your service to someone else? eBay at its height and Amazon have hit like 80, so 80% of people feel positive and would recommend it. You can have a positive or a negative. Negative NPS means people actively tell people not to use it. Right? That's the cable company. That's the Democratic party. That's maybe the Republican party. That's probably any politician. That's the IRS. Right? There's a weird thing that you'll see, which is, and we saw this with Farmville, and then we saw it again with Facebook and Instagram. Farmville had about a positive 40 NPS when people were playing it. But then after they stopped, it was like they'd given up cigarettes and we were like a negative 35 NPS.

我们也看到了同样的事情。我向你保证，Instagram 也是如此，你可能在你的朋友身上看到过，对吧？当有人这样时，就像他们戒了烟。他们变得像狂热分子，对吧？如果有人不再使用 Instagram，他们会带着自豪感告诉你。我不再用 Instagram 了，对吧？我戒掉了那种毒，对吧？我不在上面浪费时间。他们不会说他们想念它。他们会说，“啊，我希望，你知道，我只是没时间用 Instagram。我的家庭妨碍了。生活妨碍了。我真的想重新用 Instagram。”你知道，没人那么说。

And we saw the same thing. This I promise you the same is true with Instagram and you probably seen it with your friends, right? When somebody has it's like they've quit smoking. They become like a zealot, right? If someone has got is no longer using Instagram, they will say it to you with a level of pride. I don't use Instagram anymore, right? I'm off of that drug, right? I don't waste my time on that. They don't say it like they miss it. They're like, "Ah, I wish, you know, I just don't have time for Instagram. My family got in the way. Life got in the way. I'm really trying to get back to Instagram." You know, nobody says that.

不过，有趣的是，最近 Instagram 取代了我的电视。比如，我每天上 Instagram 找娱乐。

You know, it's funny though because lately Instagram has replaced TV for me. Like, I go on Instagram every day for entertainment.

但为什么呢？我的意思是，你的内容是什么？比如，什么？带我们回顾一下昨天。比如，你的 Instagram 是什么？

But why? I mean, what is your content? Like, what? Take us through yesterday. Like, what was your Instagram?

它来回切换。比如我看到很多历史内容或不寻常的事实，你可以在 Instagram Reels 中深入挖掘。然后还有 Clavicular，一个 20 岁的孩子，谈论外貌打蜡，已经火了。所以就这样来回切换。

It goes back and forth. Like I see a lot of like history stuff or unusual facts and you get to dive down into them in the Instagram reels. And then there's Clavicular who's like this 20-year-old kid who talks about looks waxing and it's gone viral. So it goes back and forth.

我同意你，因为那是 Twitter。抱歉，我还没改口叫 X，但对我来说那就是 X。我上瘾了，它是我的社交网络，我的新闻。我得到股票想法，学习历史。我有一整个社区，里面的人我只知道他们的网名。是的，我明白。我确实明白。但话说回来，现在大部分内容，我不会说大部分因为我不太确定。但在 Twitter 和 Instagram 上，我觉得有很多 AI 生成的内容。比如给我 14 个关于内战的不寻常事实，然后我就有了一个完整的 Instagram Reel，生成图片，搜集所有事实。

I'm with you because that's Twitter. Sorry, I haven't gotten over calling it X, but that's X for me. I'm addicted to it and it's my social network. It's my news. I get stock ideas. I learn about history. I have a whole community there of people I only know by their screen names. Yeah, I get that. I do. But then again though, most of the content now, I wouldn't say most because I don't really know. But on Twitter and Instagram, I feel like it's a lot of AI generated content. Like give me 14 unusual facts about the Civil War and now I have a whole Instagram reel, generate images, source all the facts.

我不知道你怎么想，但我发现 Reels 很阴险，因为我被这些 Reels 吸引，它们开始一个接一个循环，然后我就陷进去了，我不喜欢这样，这很糟糕，很邪恶。

I don't know about you, but the reels I find insidious because I get pulled into these reels and they start looping one into another and it just gets me and I don't like it and it's like that's bad, that's evil.

但你会称之为病毒式传播吗？有些有百万观看量。

Would you call that viral though? Some of those are getting like a million views.

说什么是病毒式传播很有趣。它是由算法呈现给你的，对吧？就像 TikTok。所以这是一种病毒式传播，但这是算法制造的病毒式传播，我们不太了解或理解。我想你可以逆向工程。这是一种病毒式传播。就像机器驱动的病毒式传播，或者说是人和机器驱动的病毒式传播。它不同于人们想要主动互相分享。它就在那里，我确信有人作为影响者、播客、视频等爆红。也许我们在说应用时想反了。也许有一种尚未发明的新型服务会通过那种机制传播，但它必须是我们可以参与的东西。我一直认为它是帮助我们转向生成而非消费的东西，因为我不知道价值。就个人而言，我看不到消费的价值，只是让你看东西并消费它。对我来说，我想给你一些能帮助你生成你认为有价值的东西。

It's funny to say what's viral. It's being surfaced to you by an algorithm, right? Just like TikTok. And so it is a kind of virality, but it's an algorithmic manufactured virality that we kind of don't know or understand. I guess you could reverse engineer it. It is a virality. It's like a machine-driven virality or it's a human and machine-driven virality. It's different than people wanting to actively share with each other. It's out there and I'm sure there are people blowing up as influencers and podcasters and videos and stuff like that. Maybe we're thinking backwards when we say apps. Maybe there's a new kind of service that's not invented yet that will spread through that kind of mechanism but it's got to be something we engage with. And I keep thinking it's something that helps us flip to being generative and not consumptive because I don't know the value. Personally, I see little value in consumption and just getting you to watch something and consume it. To me, I want to give you something that is going to help you generate something you think is valuable.

也许我们只是怀念过去人少的时代，而现在人很多。例如，20 年前，互联网上的人更少。

Maybe we're just missing times when there were fewer and now there's a lot. So for instance, 20 years ago, there were just fewer people on the internet.

比如你创办 Zynga 的时候，那是我们第一次达到十亿用户，现在听起来不算多，但现在互联网用户有五十亿了。是啊。比如我刚开始做这个播客的时候，只有几百个播客，现在有 500 万个播客，所以更难了。好吧，去年有 4 万个应用进入商店。也许在 2009 年或者应用商店刚推出的时候，只有一千个应用。我不知道第一个……不是一千个。我的意思是，老实说，如果我们回头看，大概有 4 万个。但区别在于，尽管用户更少，但有很多人，他们安装得更多。他们下载和安装得更多。然后开发爆发了。可能达到了 50 万个应用，现在又大幅下降，因为对任何人来说，它都不再是一个很好的分发机制了。

Like when you started Zynga, there was basically that was the first time we hit a billion users, which doesn't sound like a few, but now there's like five billion users of the internet. Yeah. When I started this podcast, for instance, there was just a couple hundred podcasts and now there's 5 million podcasts, so it's harder. Okay, there's 40,000 apps going into the store last year. Maybe in 2009 or whenever they launched the app store, there was a thousand apps. I don't know what the first... not a thousand. I mean, honestly, if we look back, there was probably around 40,000. But the difference is that there were many more people even though there were less users, they were installing a lot more. They were downloading and installing more. And then development exploded. It probably got up to 500,000 apps and now it's crashed way down because it's not a very good distribution mechanism for anyone anymore.

是啊。那么，企业家该怎么做呢？同样，AI 编程似乎是一个可以轻松创建有用应用的地方。如果某样东西非常有用，那么我们应该谈谈你的“Proven Better New”方法，这是一个很好的模型，就像一个过滤器，用来决定该构建什么应用。哦，它被验证过吗？它是行业的改进吗？它比之前存在的更好吗？有没有什么新东西你可以测试？这似乎是一个很好的过滤器，用来寻找你可以构建的 AI 应用，这些应用可能不会病毒式传播，但足够有用，人们会为之付费，或者围绕它会有某种商业模式。

Yeah. So, what does the entrepreneur do? Like again, AI coding seems like a place to create useful apps easily. And if something's super useful, so and we should talk about your proven better new approach which is a great model for kind of a great filter for saying what app should I build? Oh, is it proven? Is it an improvement industry? Is it better than what existed before? And is there something new that you could kind of test? It seems like a great filter for finding AI apps that you could build that might not be viral but useful enough that people will pay for it or there will be some business model around it.

事实上，我们根据第一章和整本书制作了一个 GPT 和一个技能文件，它在“Proven”方面做得非常出色。它真的能加速你的进程。我的意思是，你不应该在不从它开始的情况下做任何事情。我们会在某个地方发布它，比如 GitHub 之类的。

Well, in fact, we made a GPT and a skill file out of first that chapter and then the whole book that does an amazing job on proven. It really accelerates you. I mean, it's like you shouldn't do anything without starting with that. It's just you should just we'll post it somewhere, GitHub or something.

好吧，也许描述一下“Proven Better New”，因为它真的很迷人，我也喜欢你把它应用到娱乐领域。比如在开发电视剧时，它非常有效。你可以在每一部电视剧中看到它。你知道，哦，这个类型是经过验证的，比如西部片或者机制。

Well, maybe describe the proven better new because it's really fascinating and I also like how you apply it to entertainment. It really works well when developing a TV show for instance. You can see it in every TV show out there. You know, oh, this genre is proven like a western or the mechanics.

我的意思是，这不仅仅适用于游戏，虽然从游戏开始学习这个概念很好，因为它更容易理解，但它实际上适用于一切，甚至适用于企业产品和模拟产品。想法是这样的。首先，你有一些直觉，识别并隔离你内心感受到的直觉并写下来非常重要。与那种直觉接触，然后意识到把它与你正在实例化的东西分开，也就是你产品化那种直觉的方式，然后分开它，因为这将极大地增加你成功的几率，或者减少你因错误原因失败的几率。然后走出去说：‘好的，哪些最接近的产品已经做了我兴奋的部分或大部分，我可以在同一平台上为同一用户看到，我要真正成为那个产品的博士。我要深入到像素级别。我的意思是，我不会让体验中的任何一秒、一次点击或一个像素被忽视。我要用如此精细的显微镜观察它，然后重建它，思考机制，从首次用户体验开始的一切。我要逐个屏幕、逐个点击地检查一切，每个功能，无论你是否认为它是一个功能。如果有通讯录上传，任何构成体验的东西，你都要捕捉并观察它，并试图理解为什么它是经过验证的？为什么人们在该市场中最喜欢那个产品？如果它非常成功并且对你有吸引力，你可以认为它是经过验证的，然后你说：‘我能让那个产品明显更好的是什么？’这不是你认为更好的东西，因为那是新的。十个现有用户中有十个毫无疑问会说‘操，是的’。而且它现在是免费的。没有下载。你知道，价格减半。它是明显更好的东西。人们在产品中关心的东西，而且明显更好。它可能是一个非常非常小的微调，比如我们的游戏《与朋友一起玩单词》，你可以说那只是拼字游戏。人们说：‘哦，那只是拼字游戏的复制品。’但它肯定不仅仅是复制品，因为它击败了拼字游戏。我们达到了 1400 万日活跃用户，而拼字游戏只有 200 万日活跃用户，那还是在几年后。所以肯定有其他我们做对了的东西，更好。所以有时可能是打磨，你知道，在《农场小镇》中，我们有更好的作物美术和更好的作物数学。所以可能是一些非常小的东西。我们点击次数更少。新的是你的新颖想法。那是你的创新区域。那可能是你最兴奋的东西。也是最可能失败的东西。我在 Zynga 走廊里仍然被重复的一句格言是‘所有新的都会失败’。这听起来可能像打击。如果你以这句格言开始，你不会失望，因为有很多很多统计证据表明所有新的都会失败，比如去年应用商店中 100%的新应用都失败了，对吧？我的意思是，从统计上看，你可以看到所有新的都会失败。这并不意味着没有新产品突破，也不意味着你不应该去尝试。但这确实意味着你必须谦虚和好奇，并假设你的这个新想法可能会失败，并隔离它。所以这部分是不要因为错误的原因失败。如果你改变了首次用户体验中的一切，而它并没有更好。十个中有十个人不会感谢你改变它，而且它显然不是你的新创新区域，那么你就是一个糟糕的产品制造者。你知道，初级产品制造者会犯这个错误。他们只想成为他们认为的毕加索，但他们不是真正的毕加索，因为真正的毕加索首先是一个临摹和复制他人艺术的大师，然后才去创作自己的艺术。我想说，如果我们真正尊重我们的用户，如果我们真的非常有野心，那么我们将杀死我们的自我，我们将说，我定义成功不是通过同行的尊重，不是通过我的自我想要被喂养的东西，而是通过印第安纳州一位护士的眼睛来定义创新。她是否只是更喜欢我的产品而不是其他产品？她甚至说不出为什么。因为我的创新如此微小和不引人注目。那就是《与朋友一起玩单词》。

I mean, and it's not just in games when it's great to start learning this concept around games because it's easier to get your head around it, but it actually applies to everything even to enterprise products and even analog products. The idea is this. It's that there's well first of all you have some instinct and it's really important for you to identify and isolate what is the instinct that I'm feeling in my gut and write about it. Get in touch with that instinct and then realize separate that from what you're instantiating, the way that you're productizing that instinct and just separate it because that is going to massively increase your odds of success or reduce your odds of failing for the wrong reasons. And then go out and say, 'Okay, what are the closest products that have done some or most of what I'm excited about that I can see on the same platform for the same users and I'm going to really be the get my PhD in that product. I'm going to get down to the pixel level. I mean, I'm going to not let any second or click or pixel of that experience go unnoticed. I'm just going to have such a fine microscope on it and I'm going to kind of rebuild that and think of the mechanics, what are everything from the first time user experience. I'm going to go screen by screen, click by click into everything, every feature whether you think of it as a feature or not. If there's an address book upload, like anything that makes up that experience, you want to capture and look at that and try to understand why is that proven? Why do people engage the most with that product in that market? If it's very successful and it speaks to you, you can consider that proven, then you say, 'What could I make obviously better about that product?' And it's not what you think is better because that's new. What 10 out of 10 of the existing users without question would say, 'Fuck yes.' And that's it's now free. There's no download. You know, it's half the price. It's something noticeably better. Something people care about in the product and it's noticeably better. And it could be very very small fine-tuning that if you think of our game Words with Friends, you could say that was just Scrabble. People said, 'Oh, that's just a copy of Scrabble.' But it must have been more than a copy because it trounced Scrabble. Like we got up to 14 million daily active users and Scrabble got up to like 2 million daily active users and that was even years later. So there must have been something else that we got right that was better. And so sometimes it can be the polish, you know, and in Farmville, we had better artwork around the crops and better math around the crops. So it can be something very small. We had less clicks. New is your novel idea. That's your innovation zone. That's what maybe you're most excited about. It's also what's most likely to fail. And one of my mantras that's still repeated in the hallways at Zynga is all new fails. And that might sound like a beatdown. If you start with that mantra, you won't be disappointed because I mean there's lots and lots of statistical proof that all new fails like the app store last year where 100% of new apps failed, right? I mean statistically you can see that all new fails. It doesn't mean no new products break through and it doesn't mean you shouldn't go for things. But it does mean that you have to be humble and curious and assume that this one new idea of yours probably will fail and isolate that. So part of this is don't fail for the wrong reasons. If you changed everything in the first time user experience and that wasn't better. So 10 out of 10 people don't thank you for changing that and it wasn't obviously your new innovation zone, then you're a shitty product maker. You know, junior product makers make that mistake. They just want to be what they think of as a Picasso, but they're not a real Picasso because the real Picasso actually was a first he was a master at tracing and copying other people's art before he ever went and did his own art. And I like to say, you know, if we really respect our users and if we're really really ambitious, then we're going to kill our egos and we're going to say, I'm going to define success not by respect from my peers, not by what my ego wants to be fed, but I'm going to define innovation in the eyes of a nurse in Indiana. And does she just vibe more with my product than some other product? And she can't even say why. And because my innovations were so little and unnoticeable. And that was Words with Friends.

它只是打磨得更好，更流畅，而且即时社交。那就是 Zynga Poker。我们完全复制了领先的扑克游戏。你找不到任何不同之处，但不需要下载。所以，他们都在我们之前上了 Facebook。有十个扑克游戏。它们都需要下载。我们不需要。所以，他们在第一次点击时就流失了 50% 到 80% 的用户，因为你说下载会流失 80% 的用户，让他们点击会流失 50%。所以他们已经落后于我。然后他们的游戏被证明更好，因为没有下载，而且新想法在我第一次尝试时就成功了。当你终于准备好做“Proven Better New”时，事情就是这样。你的第一个想法就奏效了。这不是承诺，但这是我身上发生的怪事。我的新想法是来自 Facebook 的真实用户，25% 的情况下人们会加入朋友的牌桌。所以这第一次尝试就成功了。但我不明白为什么人们不利用这个框架来创建 GPT 或技能文件。至少它可以为你做“Proven”和基准测试，几秒钟内完成。它可以做到 A 级水平，和最好的 Zynga 产品经理一样好。它在“Proven”方面很出色。那是 AI 擅长的。它在“Better”方面是 B 级，在“New”方面是 F 级。这对我们来说是好事，因为这意味着 AI 不会很快取代我们。

It was just such better polish and it was smooth and it was instantly social. And that was Zynga Poker. We completely copied the leading poker games. There was not a single thing you could detect that was different, but there was no download. So, they were all on Facebook before us. There were 10 poker games. They all required downloads. We didn't. So, they were losing 50 to 80% of their users at the first click because you lose 80% of your users when you say download, 50% when you ask them to click. So, they were already behind me. And then their game was proven better because there was no download and the new idea happened to work on my first try. That's the way it goes when you're finally ready to do proven better new. It turns out your first idea works. That's not a promise, but that's a weird thing that happened to me. And my new idea was real people from Facebook, from your work, and 25% of the time people were joining their friends at the table. So that happened to work on the first try, of course. But I don't know why people wouldn't use this framework and create a GPT or a skill file. At least it can do proven for you and benchmarking in seconds. It can go and do that at an A level, as good as the best Zynga PM. It's amazing at proven. That's what AI is great at. It's a B at better or worse and it was an F at new. So that's good news for us because it means that AI isn't going to just replace us anytime soon.

你知道，Jay，我前几天在想，我记得在 90 年代末和 2000 年代初，我在交易、管理资金、创办企业。到处都是电子表格，墙上的便利贴。我用了大量互不通信的软件程序。每次有人问一个简单的问题，比如‘嘿，我们现在到底有多少现金？’需要三个人以上才能搞清楚。快进到今天，我仍然看到创始人们经历同样的噩梦。只不过现在世界以光速运转。他们说你的企业每迟一天使用 AI，就会落后两天。竞争对手不会等待。但好消息是，实际上有一种方法可以赶上并大幅领先。它叫 Netsuite。Next，你可能已经知道 Netsuite。它是一个 AI 驱动的业务管理工具。将所有数据连接到一个地方。我说的是财务、库存、商务、人力资源、CRM 等等。超过 43,000 家公司使用它。但 Netsuite Next 是他们的下一个巨大飞跃。AI 不仅仅是附加的。它内置于你所做的一切中。它会在你的一天中自动呈现定制洞察。你甚至不知道自己需要看到的东西。就像 AI 智能体与你并肩工作，解决问题，消除通常占用大量时间的例行工作。最棒的是，你可以询问关于你业务的问题。你可以问，‘哦，我们上个月的增长是多少？我们需要招聘新员工吗？我需要跟进哪些销售联系人？’等等。它拥有所有这些数据。它不是某种通用的一刀切的东西。Netsuite 几乎可以为你能想到的每个行业定制。所以无论你是经营电子商务品牌还是制造企业，都没关系。你的公司赚几百万还是几亿，都没关系。你的企业终于可以使用 AI 了。你不必自己想办法编码。只需立即访问 netsuite.ai/james。netsuite.ai/james。为每个行业构建。为每个董事会准备就绪。netsuite.ai/james。去做吧。你以后会感谢我的。

You know, Jay, I was thinking the other day, I remember back in the late 90s and early 2000s, I was trading, managing money, building businesses. There were spreadsheets everywhere, post-it notes on the wall. I was using a ton of different software programs that just didn't talk to each other. And every time someone asked a simple question like, 'Hey, how much cash do we actually have right now?' It took three people and more to figure it out. Fast forward to today and I still see founders living that same nightmare. Except now, of course, the world moves at light speed. They say that every single day your business is late to AI. You fall two days behind. The competition isn't waiting. But here's the good news. There's actually a way to catch up and then pull way ahead. It's called Netsuite. Next, look, you probably already know Netsuite. It's an AI powered business management tool. Connects all of your data in one place. I'm talking financials, inventory, commerce, HR, CRM, whatever. Over 43,000 companies use it. But Netsuite Next, this is their next huge leap. AI isn't just bolted on. It's built into everything you do. It automatically surfaces custom insights throughout your day. Stuff you didn't even know you needed to see. Like AI agents are working right alongside you solving problems, knocking out all the routine work that usually eats up a lot of time. And the best part is you can ask questions about your business. You can ask it like, 'Oh, what was our growth last month? Do we need to hire new employees? Which sales contacts do I need to follow up on?' And on and on. It has all this data. It's not some generic one-size-fits-all thing. Netsuite gets customized for pretty much every industry you can think of. So whether you're running an e-commerce brand, a manufacturing business, it doesn't even matter. Your company's doing millions, hundreds of millions, it doesn't matter. Your business finally can use AI. You don't have to figure out how to code it yourself. Just go to netsuite.ai/james right now. netsuite.ai/james. Build for every industry. Ready for every boardroom. netsuite.ai/james. Do it. You'll thank me later.

我喜欢这个‘Proven 等于复制过去’的概念。就像你在书中提到的，毕加索如何临摹大师的作品来学习他们的风格，甚至模仿、复制或窃取。这是一个隐藏的优势，人们在心理上通常不认同。他们很像……你在书中还举了另一个很好的例子，Slack 对比 HipChat。是的，Slack 基本上是一样的东西。它们只是像你提到的其他产品一样，打磨得更平滑。它是由一家游戏公司制作的。而 HipChat 是一家无聊的企业公司。所以 Slack 更有游戏感。它第一天就有大约 8,000 次安装。

I love this concept of proven equals copy the past. Like you mentioned in the book, how Picasso would literally trace the old masters in order to learn their style and maybe even emulate or copy or steal. And this is such a hidden advantage that people psychologically don't normally subscribe to. They're a lot like... and you can give another great example in the book of Slack versus I guess it was HipChat. Yeah, where Slack basically was the same thing. They just had, like you mentioned with these other products, they had smoother polish. It was made by a game company. So, and HipChat was like an enterprise boring enterprise company. So, Slack just had a more of a game-like feel in there. It had what about 8,000 installations the first day.

是的，没错。这实际上是一个彩蛋，很有趣。如果你能制作一个更优秀的企业应用，并在其中加入一点乐趣或游戏机制，你会感到震惊。我的意思是，这太愚蠢了。我和 OpenAI 的一些团队聊过，他们说这种事经常发生。每一个大型超大规模公司都有过这种经历。这不是他们想要的，也不是他们构建的目的，但却是他们得到的。他们说，‘哦，天哪，有人在 GPT 上放了一个愚蠢的宠物游戏，一个小 AI 宠物游戏，它火了，OpenAI 内部每个人都在玩，这太奇怪了，我们不知道为什么。’Facebook 也是如此。当游戏流行起来时，他们很震惊。这不是他们构建或意图的。那是在消费领域。你知道，Slack 证明了人们在企业领域也想要乐趣。我们与美国运通合作发行了一张联名信用卡，我想叫它‘乐趣卡’。他们给它起了一个糟糕的名字。我想它叫 Discover 卡，但每次……人们喜欢赌博，对吧？我们喜欢赌博。我们的生活很无聊，对吧？用信用卡购物很无聊。如果你每次拿出信用卡都能赢点什么，你会选择它而不是其他吗？我想是的。还有 20,000 个注册 Discover 卡的孩子也这么想，对吧？每次他们用它买东西，都会赢点什么。偶尔还会赢大东西。他们在我们的游戏中赢了东西。事实证明这很有效。你可以在企业应用中使用这些游戏参与机制。你可以奖励人们每小时回来一次，我是说每天。你可以创建排行榜。所有这些都有效。这是经过验证的。Slack 在企业中证明了这一点。没有人费心去复制他们。我只想对你们的用户，你们的听众，这个播客的用户说。这是 Peter Thiel 意义上的道德套利。他喜欢好的道德套利。这个就摆在我们所有人面前。如果你愿意放下自我，好吧，如果你愿意，而且如果你做得很好，就像 Stewart 做 Slack 那样，你的用户甚至不会指责你抄袭，对吧？这才是真正的大师课。

Yeah. And exactly. And this is actually an Easter egg for people, which is fun. If you can go and make a proumer enterprise app and you actually put in little doses of fun or game mechanics in it, you'll be shocked. I mean how dumb it is. I mean I was talking to some of the team at OpenAI and they said and this always happens. Every one of these big hyperscalers has had this experience. It's not what they want. It's not what they build for but it's what they get. They said, 'Oh my god, someone put a dumb pet game, a little AI pet game out on GPT and it's gone viral and everyone inside OpenAI is playing it and it's so weird and we don't really know why.' That was the same thing at Facebook. They were shocked when games took off. It's not what they built for or intended. And that's in a consumer space. You know, Slack proved that people want fun in the enterprise space. You know, we did a joint credit card with American Express that I wanted to call the fun card. They called it some terrible name. I think it was called the Discover Card, but every time it... People love gambling, right? We love gambling. We have boring lives, right? It is boring to buy with your credit card. What if you could win something every time you brought your credit card out? Would you pick that over not? I think yes. And so did 20,000 other kids who signed up for the Discover card, right? Every time they bought something with it, they won something. And once in a while, it was something big. They won something in our games. That turns out it worked. You can use these game engagement mechanics in an enterprise app. You could reward people for coming back once an hour. I mean, every day. You could create leaderboards. I mean, all this works. It's proven. And it's Slack proved it in the enterprise. Nobody bothered copying them. And I'll just say this for your users, for your listeners, the users of this podcast. It's a moral arbitrage in the Peter Thielian sense of a moral arbitrage. He loves a good moral arbitrage. This one is just sitting there for all of us. If you were willing to kill your ego, okay, if you're willing, and by the way, if you do this really well, the way Stewart did with Slack, your users, no one even accuses you of copying, right? That's the real master class of this.

如果你能完美复制一个已被证明成功的应用，而且没人觉得它是抄袭，那就是魔法般的技巧。就像你说，‘看这边，各位，看看我进门时的合十礼和可爱的外观。别往那边看，那边只是 100%的 Hip Chat。’我的意思是，除了我之外没人真的指责他们抄袭 Hip Chat，而这是尊重。当我看到那种水平时，那就是大师课，那是天才的产品制造者。

If you can perfectly copy a proven successful app and no one thinks it's a copy, that is the magic trick. Like you just say, 'Look over here folks, look at my namaste when you first enter and my cuddly look and feel. Don't look over here where it's just 100% Hip Chat.' I mean, no one really accused them of copying Hip Chat except me, and it was respect. When I see that kind of level, that is a masterclass, that is a brilliant product maker.

是的。不，我同意。我喜欢的是，你似乎在告诉人们，你有权回到那些价值数十亿美元的行业，重新做一遍，并乐在其中。

Yeah. No, I agree. And what I like in this is you're sort of telling people you have permission to go back to the multi-billion dollar industries that are out there and just do it again and have fun with it.

这是另一件我忘了写进书里的事，但这是我的人生哲学之一。

This is another thing I think I forgot to put in the book, but one of my life philosophies.

你已经像是在做第二部分了，甚至不是附录。你需要一个续集。

You already made like part two, not even an addendum. You need a sequel.

是的。我会把它发在 Medium 或 Substack 上。抱歉，我不想听起来那么过时。我是 Substack 的投资者。我的意思是 Substack。我的一位合著者开始称这些为‘马克主义’：变化越多，越保持不变。所以即使在这样一个感觉什么都没被证明的环境里——你可以说在消费级 AI 中，很少有被证明的东西——如果你称之为平台，这个技术平台加上这个受众，那么你就要去看之前被证明过的东西，因为它很可能会被重新发明。社交网络上的第一批病毒式应用，几乎每次这些消费平台爆发时，都是——我不知道你是否记得那些‘真相盒’或‘诚实盒’或‘暗恋’应用。它们从高中生开始。你会收到一条消息：‘你认识的某个人回答了关于你的问题。找出他们对你的看法。找出三个朋友匿名对你的评价。’你点击，它说：‘你需要先回答关于三个朋友的问题，我们才会显示给你。’你输入答案，然后他们发短信给他们。这些是超级病毒式的，而且总是有效。我的意思是，Gas 和 tbh（老实说）以及 Nikita 在移动端做过这些。现在我知道有几家初创公司正在用 AI 的方式做这些，让 AI 作为中间人，更智能、更对话化。你会看到这些可能再次成功，带着 AI 的 twist。这是‘已被证明的更好新事物’的一个变体。

Yeah. I'll just post it somewhere on Medium or something or a Substack. Sorry, I don't want to sound so outdated. I'm an investor in Substack. I mean Substack. One of my co-authors started calling these 'Marcisms': the more things change, the more they stay the same. So even in an environment like this where it feels like nothing is yet proven—you could argue in consumer AI very little has been proven—if you call it a platform, this technology platform with this audience, then you go look at what was proven before because it will probably be reinvented. The very first viral apps on the social networks, pretty much every time one of these consumer platforms blows up, are the—I don't know if you remember these truth boxes or honesty box or crush apps. This is where they start with high schoolers. It's where you get a message: 'Someone you know has answered these questions about you. Find out what they think of you. Find out what three friends all said about you anonymously.' And you click and it says, 'You need to answer questions about three friends before we'll show it to you.' And you put them in, and then they text them. And these are hyperviral and they always work. I mean, it's what Gas and tbh (to be honest) and Nikita did these on mobile. And now I know a couple of startups that are doing these with an AI twist, where you have the AI in the middle that's being the intermediary and is a little more smart and conversational. And you're going to see those probably work again with an AI twist. And that is a variant of proven better new.

这也让我想起：你记得 Genie 吗？它也是存在的。它是‘建立你的家谱，顺便说一句，如果他们还在世，把他们的电子邮件地址放在这里，我们会发给他们去建立家谱。’它一周内就达到了一亿用户。我觉得 Genie 没有完全成功，但他们在内部使用 Yammer 来讨论公司。他们发明了 Yammer 作为讨论自己公司的工具，后来以数十亿美元卖给了微软。我不知道这个。我知道 Ancestry.com 变得很大，我认为那是另一个很好的领域。你可以——我会鼓励听众，不要发明新业务。去追求那些庞大而无聊的现有业务，比如招聘、约会、家谱。找到那些已经有很多用户和金钱的东西，看看你是否能加入 AI 的 twist，或者通过你提到的那种病毒式传播找到方法。因为有视频会被机器算法捕捉到。然后如果你要复制它们，就像大师一样复制，不要只做一个变体，做 20 个。而且不要先构建这 20 个，而是从测试什么能吸引点击开始，或者从哪些视频会病毒式传播开始，然后从那里构建你的产品。我的意思是，这有点像漏斗顶部思维。‘好的，我想找到关键，然后我再往下走。’是的，我同意你之前说的，AI 可以让我们快速构建原型或测试事物、创建视频、快速测试。不幸的是，现在人们似乎以不同的方式使用 AI。他们只想构建他们的原型，并对那一个原型感到兴奋，然后一直做下去。所以我没有看到人们测试和失败的很好例子。

It also reminds me somehow: do you remember Genie, which was also around? It was 'build your family tree, and by the way, if they're alive, put their email address here, we'll send them to build their family tree.' And it went to 100 million users in a week. And I think Genie didn't quite work out, but they were using internally Yammer to discuss the company. They invented Yammer as a tool to just discuss their own company, and that sold to Microsoft for billions of dollars. I didn't know that. I know Ancestry.com got pretty big, and I think that's another great area. You could—and I would encourage listeners to, you know, don't invent a new business. Go after big boring existing businesses like jobs, dating, ancestry. Find things that already have lots of users and money, and see if you can add an AI twist to it, or come up with a way to get viral through the kind of virality you were talking about. Because there's videos that the machine algorithm picks up. And then if you're going to copy them, copy them like a master, and don't just do one variant, do 20. And don't build the 20 like first start with testing what clicks, or which start with what videos get viral, and then build your product from there. I mean, this is kind of thinking top of the funnel. 'Okay, I want to find the key, then I'm going to go lower.' And yes, I agree with what you said a while ago that AI can enable us to build prototypes or test things, create the videos, test things quickly. Unfortunately, it seems like people are using AI differently now. They're wanting to just build their prototype and they're excited about that one prototype, and then they just keep going with it. So I'm not seeing great examples of people testing and failing.

是的，我想没错。你自己有没有用这些编码工具编程过任何 AI 应用？

Yeah, I guess that's right. Have you programmed any AI apps yourself using any of these coding tools?

我试过，而且我做得非常糟糕，太难了。它还没准备好迎接像我这样的傻瓜的黄金时代。比如我试过 Replit，想为我伴侣 Hillary 做一个网站。而 Midjourney 是神奇的。我的意思是，Midjourney 从我第一次使用起就很神奇。通过 Discord 的用户界面极其困难且糟糕。是的。我有点讨厌 Discord 因为 Midjourney。但不知怎的，我不会改变它。那是经过验证的。如果我要和他们竞争，我会做同样奇怪的东西，因为那就是他们做的，而且有效。但 Midjourney 真的让我看起来比实际更有创意。我的意思是，Midjourney 的输出难以置信。所以我给她做了一个标志。她觉得世界的本质对她说话，告诉她她的使命是做一个垃圾收集者。她的使命是找到不同的方式收集世界上的垃圾，像一只秃鹫。所以我在 Midjourney 上为她做了一个非常棒的标志，一只秃鹫飞过城市寻找垃圾。然后我试图用 Replit 为她做一个网站，只是一个网站。我花了一周时间，但无法让 Replit 移动标志，它总是遮住网站标题。我的意思是，整整一周。是的，我很糟糕。我很烂。这就是为什么我对消费者有好的头脑，因为我代表傻瓜。我不愿意学习如何使用 Replit。我不愿意让别人教我。我就是不够在乎。然后我用了 Claude Code，试图做一个 AI 助手，花了两个星期，Google 连接器每天都坏。所以我不得不创建另一个健康检查应用，每天检查那个的健康状况，然后尝试修复。然后它会卡在这些‘允许’上。然后我试图让 Claude 同事调度帮我点击‘允许’，然后它开始拒绝这样做。所以这是一次很好的学习经历。我确实在大约一小时内做了一个第一人称射击游戏，玩了一次。我发现我的高度是 Claude Code。我觉得神奇的是调度技能文件代理。我创建了代理，持久代理，并以我的家人命名。所以 Carmen，我的女儿，是我的产品经理。

I have, and it's amazing how badly I did, how hard it was. It's not ready for what I'd say the prime time of dumb like me. Like I tried Replit and I tried to make a website for my partner Hillary. And Midjourney is magical. I mean, Midjourney has been magical since the first time I used it. The user interface through Discord is insanely hard and terrible. Yes. I kind of hate Discord because of Midjourney. But somehow I wouldn't change that. That's proven. Like if I was trying to compete with them, I'd make the same weird ass thing because that's what they did and it's working. But Midjourney literally makes me look more creative than I am. I mean, it's unbelievable the output of Midjourney. So I made her a logo. She thinks the nature of the world spoke to her and told her her mission is to be a garbage collector. Her mission is to find different ways to collect garbage in the world and be like a vulture. And so I made her this really amazing logo on Midjourney with a vulture flying over a city looking for garbage. And then I tried to make her a website with Replit, just a website. And I spent a week on it, and I couldn't get Replit to move the logo, and it kept covering the title of the website. I mean, literally a week. And yes, I am terrible. I suck. That's why I have a good mind for consumers, because I represent the dumb. I'm not willing to learn how to use Replit. I'm not willing to have someone teach me. I just don't care enough. And then I used Claude Code and I tried to make an AI assistant, and I spent two weeks on that, and the Google connector broke every day. And so then I had to create another health check app that would check the health of that every day and then try to fix it. And then it would get stuck on these 'allows'. And then I tried to have Claude coworker dispatch try to click 'allow' for me, and then it started to refuse to do that. So it was a good learning experience. I did make a first-person shooter game in like an hour that I played one time. I found my altitude was Claude Code. What I find magical is dispatch skill files agents. I've created agents, persistent agents, and I name them after my family. So Carmen, my daughter, is my product manager.

那么，你的智能体都做些什么？我自己在编码方面其实更成功，但智能体方面我不太确定。你拿智能体做什么？

And so, what do your agents do? I'm actually having more success with the coding stuff on my own, but the agent stuff, I'm not sure. What do you use the agents for?

我喜欢这些智能体。我觉得这很神奇。所以，我四处走动，和我的所有智能体进行长谈，我们还有团队辩论。我的设计主管是希拉里，总经理是乔治娅，工程主管是怀亚特，大多数事情从卡门开始，因为她是我的项目经理，她负责做更好的新基准测试。她有点愤世嫉俗。她经常回来给我的想法泼冷水，告诉我这已经有人做过了，或者这违反了某些规定。她几乎成了否决部门。我需要稍微重新编程她。然后怀亚特是负责构建的人，希拉里负责设计。所以我想出一个点子，然后他们辩论如何在 24 小时内做出一个原型，而且是一次性的。我们如何做出一个可视化的原型，或者能实现我说的功能，但只能我自己用？我们不能把它发布到网站上，也不能让它靠近任何谷歌或任何会出问题的地方。我吸取了教训，不用 API。但这非常有效，最有效的是四处走动、头脑风暴，帮助我思考，真正加快了我思考想法的速度。

I love the agents. I find that magical. So, I walk around and have long conversations with all of my agents and we have team debates. I have Hillary as my head of design, Georgia as my general manager, Wyatt as my head of engineering, and most things start with Carmen because she's my PM and she does proven better new benchmarking. She's somewhat cynical. She constantly comes back to me pouring water on my ideas and telling me this has already been done and this is a toss violation. She's kind of become the no department. I need to reprogram her a little bit. And then Wyatt is the build-it guy, and Hillary is design. So I come up with an idea and then they debate how to get it to be a prototype in 24 hours, and it's throwaway. How do we make a prototype that is visual or just can do the function I'm talking about, but in a way that only I can use it? We can't publish it to a website or it doesn't go near any Google or anything that's going to break. I've learned my lessons, no APIs. But that has been really effective, and the most effective is just really walking around and brainstorming and helping me think through, really accelerating the pace that I can think through ideas.

那么哪些想法通过了测试？你在做什么？

And what ideas are passing the test? Like what are you working on?

我刚刚终止了一个为期四年的项目，我在书里提到过。20 年来，我一直在研究这个关于地球的愿景。再说一次，我的直觉是元宇宙。这是以游戏速度进行的生活。这不仅仅是书名。这个想法是，我们都将像埃隆那样生活。就像我们能够向宇宙讲述一个想法，然后神奇地，它会带回人、智能体和能帮助我们把它变成现实并测试它的人。一个测试机器会为我们弹出。我的意思是，它会消除所有枯燥的工作。这就是我这本书的真正氛围。但我终止了正在构建的一个基于浏览器的游戏渲染引擎，叫做 Stem Studio，我们刚刚把它开源了。所以我认为它是最复杂的 3GS 渲染引擎。短期内没有商业机会。我说过，我知道我的直觉是对的。但这个想法行不通，这很痛苦。这是一个激情项目。我终止了它。但它构建的目的是为了做更好的新东西，因为基本上你可以构建一个飞行模拟器，或者拿别人的飞行模拟器。它全是 stem，可以修改，你可以改变它，用 AI 让它变成多人游戏，或者说我想拿那个，现在你可以降落并进行坦克大战。所以无论如何，这些都是很酷的想法，但它不会成功，或者我得去筹集很多钱。

I just pulled the plug on a four-year project I talk about in the book. For 20 years, I've been working on this vision of Earth. And again, my instinct is the metaverse. It's life at the speed of play. It's not just the book title. This idea that we're all going to live the way Elon lives. Like we're going to be able to tell an idea to the universe and magically it's going to bring back people, agents and people that can help us turn that into something real, test it. A testing machine will pop up for us. I mean, it's going to take all the drudgery out of this. And that's really the vibe that my book is about. And so I pull the plug though on I was building a web-based browser-based game rendering engine called Stem Studio and we just open sourced it. So I think it's the most sophisticated 3GS rendering engine. There's no commercial opportunity for it anytime soon. And I said this the instincts I know are right. This idea is not going to hunt and it's painful. It's a passion project. I'm pulling the plug. But it's built to do proven better new because basically you could build a flight simulator or you could take someone else's flight simulator. It's all stems and mod it and you could change it and use AI to make it multiplayer or say I want to take that and now you can land and have tank battles. So anyway, it's really cool ideas, but it wasn't going to work or I was going to have to go raise a lot of money.

我正在做的是，我一直在思考我所说的社会膜。就像在这个智能体式 AI 世界里，当我们的智能体是我们的化身，我们的智能体在我们不在的时候漫游世界，它们与其他人类建立信任，为我们寻找线索，我们该如何建立网络？第一个测试将在未来几周内启动。我和几个黑客一起构建了一个专门为智能体设计的网络。所以我说，智能体关心什么？我们的智能体关心 token 和有效性，以及 token 有效性。所以我们启动的第一件事，我们称之为 slashwork，第一件事将叫做 slashchallenge。你的智能体——如果你是一个大 token 用户——你将能够发布你正在做的项目，或者让你的智能体向这个智能体对智能体的网络发布项目，然后其他智能体将能够去处理其中的部分，它们会得到评估和评分，会有挑战和排行榜，如果它们赢了就会获得积分。这个想法是，智能体将在不同领域竞争，比如 QA 或其他事情，如果你的智能体赢了，你会围绕这些领域开发越来越好的技能文件，或者你的智能体会，所以它会开始赢得挑战并赚取积分，然后你就可以用积分来获得一群智能体，比如几千个智能体，为你做某事。这就是我探索的领域。

What I'm working on is I'm really thinking a lot about what I call the social membrane. Like how are we going to network in this agentic AI world when our agents are our avatars, our agents are roaming the world while we're not there and they're navigating trust with other people and they're finding us leads. And the first one of the first tests of that that I'm going to turn on in the next couple weeks. I've been building with a couple of hackers a network just for agents. So I said, what do agents care about? Our agents care about tokens and effectiveness and token effectiveness. And so the first thing we're turning on, we're calling it slashwork and the first thing will be called slashchallenge. Your agent will, if you're a big token user, you'll be able to post projects you're doing or have your agent post projects to this agent-to-agent network and then other agents will be able to go and work on parts of that and they'll get evaluated and rated and there'll be challenges and there'll be leaderboards and then they'll get credits if they win. The idea is that agents will be out competing in different domains like QA or other things and if your agent wins, you'll develop better and better skill files around these areas or your agent will, so it's going to start winning challenges and earning credits and then you'll be able to use credits to get swarms of agents, you know, a couple thousand agents to go work on something for you. So that's kind of the zone that I'm playing with.

当你发布它时，人们怎么找到它？

How could people find that when you unleash it?

他们可以在 Twitter 上关注我，Mark Pincus，M A R K P I N C。我们可能会在下周末左右发布一个链接供人们尝试。我们设想第一批用户会是 Claude 和 Hermes 的高级用户，比如机器学习 AI 研究人员。到目前为止，我们展示给每一位机器学习研究人员，他们都非常喜欢。所以它最初会像是一个为你的 AI 提供的零工市场，利用空闲时间。但最终的想法是，智能体喜欢个人资料，你会有个人资料，最终它们可以开始互相建立网络，所以它们会进入小的子社区，因为它们和你对同样的事情感兴趣。所以，我只是在摆弄它并探索。

They can follow me on Twitter, Mark Pincus, M A R K P I N C. We'll post a link to this for people to try it probably by the end of next week or so. And we're thinking the first users will be power users of Claude and Hermes, like ML AI researchers. Every ML researcher we've shown it to so far has really loved it. So it'll initially be like a gig marketplace for your AI, kind of idle time. But the idea is that eventually agents love profiles and you'll have a profile and eventually they could start networking with each other, so they're in little sub-communities because they're interested in the same things you are. So, I'm just kind of playing around with it and exploring.

哦，还有一件事我要提一下，抱歉我忘了，我不小心共同创立了一家名为 Hivemind 的企业 AI 公司。它源于我的社会膜研究，我和一位非常有才华的机器学习研究员兼企业家 Raton Cardas 一起开始研究。他接手并推进了它，开始构建技术和用例，说我们如何应用这个社会膜。我们如何拥有一个持续学习系统来处理人类部分，这是最混乱的部分，输入，AI 的第一英里和最后一英里都是人类。我们如何不取代人类，而是开始从你最好的人的判断中持续学习？他们知道所有边缘情况。他们知道如何以新的方式应用它们。所以我们构建了，他的团队构建了这个系统，现在它已经在四个大型企业中使用，并且正在交付，他正在构建相当大的 ARR。所以我不能说这是我，因为我现在是一个被动的帮手，因为我不想创办一家企业公司。

Oh, another thing I'll mention, sorry I forgot, is by mistake I co-founded an enterprise AI company called Hivemind. It came out of my social membrane research and I started researching it with this really talented ML researcher entrepreneur named Raton Cardas. He took it and ran with it and started building technology and a use case saying how do we apply this social membrane. How do we have a continuous learning system that deals with the human piece, which is the messiest piece, the input, the first and last mile of the AI is the human. And how do we not replace the human, but how do we start to continuously learn from the judgment of your best people? They know all the edge cases. They know how to apply them in new ways. And so we built this, his team built this system and now he has it inside of four large enterprises and it's delivering and he's building pretty big ARR. So I can't say it's me because I'm a passive helper now because I didn't want to start an enterprise company.

嗯，但你在这个播客开头说你觉得那才是机会所在。

Well, but you started off this podcast saying that you feel that's where the opportunity is.

是的。毫无疑问。那就是冰球要去的地方，可能未来 18 个月、两年内都是这样。我对消费者充满热情，所以我会继续在消费者领域捣鼓。

It is. There's no question. That's where the puck is and it's probably where it's going to be in the next 18 months, two years. I'm passionate about consumer, so I'm going to keep tinkering on consumer.

你看到像 Twitch 这样的东西，会不会因为没做而感到遗憾？就是那种不只是玩游戏，而是看别人玩游戏的想法。

Do you look at something like Twitch and feel bad that you didn't do that? The idea of not just playing games but watching other people play games.

我记得老朋友 Susan Wajiki 问我为什么不关注游戏直播，因为它占了 YouTube 惊人的份额，大约 7% 到 8% 的观看量。但那主要是硬核玩家，不是我的大众市场休闲社交游戏领域。所以我忽略了，尽管在亚马逊收购 Twitch 时我们也有竞标。我们提出了竞争报价，不记得是 8 亿还是 10 亿。所以最终我们确实尝试收购，但完全错过了那个机会。你专注于自己的领域，我们错过了 Epic、Riot，错过了很多。

I remember Susan Wajiki, a longtime friend, told me why aren't you focused on game streaming because it made up a shockingly big part of YouTube, like seven or eight percent of all views. But it was kind of hardcore gamers, not my zone of mass market casual social gaming. So I ignored it, although we were in the running to buy Twitch when Amazon bought them. We had a competing offer. I can't remember if we offered 800 million or a billion. So we did try to buy them in the end, but we totally missed that opportunity. You kind of focus on your zone. We missed Epic, Riot, we missed a lot.

有趣的是，在 Twitch 或 YouTube 上你可以看别人打扑克，不一定是大型角色扮演游戏，就是那些你一直玩的老游戏。

It's interesting how on Twitch or YouTube you can watch people playing poker, not necessarily huge role-playing games, just classics you always played.

不，你说得对。这有点像 ASMR 的领域，人们就是看任何东西。

No, you're right. It's kind of in the zone of ASMR, people just watching anything.

是啊。

Yeah.

我太专注于我们的创新领域：改善人际关系。游戏如何帮助你在友谊中增进关系？最终目标是通过游戏促成婚姻，我们做到了。那个机会很大，现在依然如此。尽管 Twitch 很大，但我要提醒你，他们现在奇怪地把所有休闲游戏都称为社交游戏，这很奇怪，因为它实际上并不社交。他们曾称之为游戏视频，现在不社交了，却叫它社交。但游戏行业有 2830 亿美元，很无聊且不社交。它那么大，又那么糟。我们在 2007 年创办 Zynga 时，游戏行业只有 230 亿美元，很糟糕。那是小众市场，人们坐在妈妈沙发上看着坦克爆炸。现在它很大，但我认识的人都不玩游戏。我只玩 chess.com 和 Lichess，就这些。

I was so focused on our innovation zone: improving people's relationships. How do games help you move buckets in a friendship? The ultimate was getting to marriages out of games, which we did. That opportunity was so big and still is. As big as Twitch is, I'll remind you they now oddly call all casual gaming social gaming, which is weird because it's really not social. They called it game video gaming, and now that it's not social, they call it social. But the gaming industry is $283 billion and it's boring and not social. It's that big and that bad. When we started Zynga in '07, the gaming industry was $23 billion and it sucked. It was niche, people sitting on their mom's couch watching tank explosions. Now it's really big, but no one I know plays games. I play chess.com and Lichess, that's it.

我在 chess.com 和 Lichess 上很活跃。

I'm avid on chess.com and Lichess.

是吗？

You are?

是啊。我们得下一盘。不知道我该不该跟你下。你知道你在 Lichess 上的快棋等级分吗？我大概 2250 到 2300，最高到过 2400 左右。

Yeah. Well, we'll have to play. I don't know if I should play you. Do you know your rating on Lichess Rapid? I'm about 2250, 2300. My highs are around 2400.

哇。我跟你说，这甚至不是谦虚的炫耀，是虚假的炫耀。我在 Lichess 上的快棋等级分我觉得是 2200，但那只是因为我下了大概五盘棋，我女儿坐在旁边，跟我一起下，我们互相告诉对方走法，我全赢了，然后就不下了。我告诉你，我在 chess.com 上的等级分大概是 850，人们都很震惊这么低，但有很多次我在哄孩子睡觉时偷偷下，然后不得不退出。但 Lichess 更接近我的真实水平，因为我更认真。我觉得我真实的 Lichess 等级分大概是 1800 或 1700。我女儿 Georgia 痴迷于国际象棋。她跟教练学，每天下大约三小时，尤其是夏天，每天五小时。她去公园下。她是个好棋手。她能打败 2200 左右的人。

Wow. I will tell you, this is not even a humble brag, it's a false brag. My rapid rating on Lichess I think is 2200, but it's only because I played like five games and my daughter was sitting next to me, kind of playing them with me, so we were telling each other moves and I won them all and I stopped. I will tell you that my chess.com rating is like 850, and people are shocked it's that low, but there are so many games where I'm putting the kids to bed, sneaking and playing, then I have to quit. But Lichess is more my real rating because I take it more seriously. I think my real Lichess rating is probably like 1800 or 1700. My daughter Georgia is obsessed with chess. She works with a chess tutor, plays like three hours a day, especially in summer, five hours a day. She goes to the park. She's a good player. She's beating people who are like 2200s.

她的 USCF 等级分是多少？

What's her USCF rating?

我不知道。她只参加过伦敦的一次比赛，表现惊人地好。她输了一盘棋，对方说服她和棋，而她本来能赢。她才 15 岁。她需要在情绪上更强一些，但她想整个夏天都参加比赛。她真的很想成为大师或特级大师。

I don't know. She's only been in one tournament in London and did shockingly well. She lost a game where a guy convinced her to draw in a game she would have won. She's only 15. She needs to get a little stronger emotionally, but she wants to spend the summer doing tournaments. She really wants to become a master or grandmaster.

我在美国是大师。

I'm in the US a master.

是啊，她能打败 2200 的人。所以我们三个连一下。你在 Lichess 上跟她下会比跟我下更好。

Yeah, she's beating people with 2200s. So let's connect you, me, and her. You'll have better games on Lichess with her than me.

是啊。不，这很有趣。我每周都上课。我试图回到我 USCF 的巅峰大约 2250，但年纪大了，不可能了。

Yeah. No, it's fun. I mean, I take lessons every week. I'm trying to get back to my peak USCF of about 2250, at an older age, and it's impossible.

哦，真的吗？是啊。我肯定不如她学得快、掌握得快，但我还能赢她。这更多是因为我是她爸爸，知道她的情绪弱点，可以利用这些。

Oh, really? Yeah. I'm definitely not learning and picking up as fast as her, but I can still beat her. It's more because I'm her dad and I know her emotional weaknesses and I can exploit those.

真狠。

That's brutal.

但我从不让孩子赢。所以你也不能。

But I never let kids win. So you can't either.

Mark，《Life at the Speed of Play》的作者。我非常喜欢这本书。我写作时一直有种不安，就是不想让它无聊。有太多商业书让我觉得痛苦，我对作者感到恼火，因为他们讲显而易见的东西，搞这种套路。我冒险忽略了一些已被证明的想法，但就是不想写枯燥的内容。最后我明白了，只要我聊得开心、写得开心，读起来也会有趣。书中有个权衡，有些章节，比如深入探讨已被证明更好的新事物，更枯燥、更学术，但也更有价值。故事更吸引人、更有趣，你也能学到东西，这两者之间总有取舍。我的合著者曾合著了 Ben Horowitz 的《The Hard Thing About Hard Things》。他们让书一半是故事，一半是教训。从某些方面看，那更聪明，而我更雄心勃勃。我只是不想让它无聊。我觉得你做到了。你在书里讲了很多故事，很吸引人，而且有点历史感，因为你真的回溯到 90 年代和 2000 年代，那时这一切刚刚兴起。我不想自称 OG，但每个时代都有很多教训可以学习。当你退后一步，看看别人以前做过什么，你会更成功，不会陷入这些小竞赛。你不必仅仅因为编码智能体现在很火就去构建一个。

Mark, author of Life at the Speed of Play. I thoroughly enjoyed the book. I had this pit in my stomach the whole time writing, like I just don't want it to be boring. There are so many business books that I find painful and I get annoyed at the author because they're making obvious points, doing this playbook. I am ignoring some of the proven ideas at my own risk, but I just didn't want boring dry content. I learned by the end that whatever I had fun talking about and writing about, I thought would be fun to read. There is a trade-off in the book because some chapters, like going deep on proven better new, are more dry and professorial but more valuable. There is always this trade-off between stories being engaging and fun and you kind of learn. My co-author co-authored Ben Horowitz's book, The Hard Thing About Hard Things. They made half the book stories and half lessons. In some ways, that's smarter, and I was more ambitious. I just didn't want it to be boring. I feel like you did that though. You told a lot of stories in this book and it's fascinating, and it's kind of historical because you really go back to the 90s and 2000s where all this first blossomed. I hate to be like the OG, but there are so many lessons to learn from each era. When you pull back the camera and look at what people have done before, you'll be more successful and less caught up in little races. You don't have to build a coding agent just because that's massively working right now.

或者也许你应该，你知道，但还有很多其他话题。我们可以再做一次，因为我想你在芝加哥。

Or maybe you should, you know, but also there's so many other topics. We can do this again because I think you're in Chicago.

是的。

Yeah.

那很酷。我在芝加哥长大，所以我对芝加哥很熟悉。

That's cool. I grew up in Chicago, so I have a lot of Chicago.

所以，哦，是的，没错。

So, oh yeah, that's right.

你多久来一次这里？

How often are you here?

我七月中旬会去那里，因为我还有家人在那里，而且我们要参加一个婚礼。

I'm going to be there in the middle of July because I still have family there and we have a wedding.

等你来城里的时候，我们做一期播客吧。

Let's do a podcast when you're in town.

哦，是的。我们可以当面做。那会很有趣。

Oh, yeah. We could do it in person. That'd be fun.

是的。带上棋盘。我会带棋盘。

Yeah. And bring the chessboard. I'll bring the chessboard.

好的。我们会带上乔治娅，至少她能跟你认真下一盘。

Okay. And we'll bring Georgia so at least she can give you a serious game.

太好了。太好了。那么，马克，再次感谢。《游戏速度的生活》是本好书，我期待继续我们的对话。

Excellent. Excellent. Well, Mark, thanks once again. Life at the speed of play. Great book and I look forward to continuing the conversation.

是的。很有趣。

Yeah. Fun.