Build for Tomorrow's Model

"I think university is not the best place to learn. Doesn't matter what you're studying."

「我觉得大学不是学习的最佳地方。学什么都一样。」

"I'd invest in Grock and I would probably short Anthropic. No, I would — I would short [OpenAI]."

「我会做多 Grok,做空 Anthropic——不,做空 OpenAI。」

"Why?"

「为什么?」

"I think it's more the slope on the Grock team. They're doing something which I respect a lot, which is to hire missionaries for the data curation part."

「主要是 Grok 团队的斜率。他们在做一件我非常尊重的事——为数据管理(data curation)那一块招传教士式的人。」

"The morale is super high. OpenAI has gone through all this mess, right?"

「他们的士气极高。OpenAI 这一阵子内部一团乱,你懂的。」

"Do you think there will be a leading model that has not been created yet?"

「你觉得最强的模型,会是一个还没出现的模型吗?」

"Yes. From China."

「会。来自中国。」

"Do you worry about China?"

「你担心中国(在 AI 上的进展)吗?」

"I do think there's like a 50/50 chance they will have the best model. We'll be using a Chinese model at some point — because they're ready to go."

「我确实觉得,他们有 50/50 的概率会做出最强的模型。我们 Lovable 早晚会用上中国的模型,他们已经准备好了。」

"Anton, dude, I'm so excited to be here with you in person. Thank you so much for joining me on the show."

「Anton,兄弟,太开心能当面和你录这一期了。谢谢你来录这个节目。」

"It's great to see you, and thanks for coming to Stockholm."

「也很高兴见到你,谢谢你跑一趟 Stockholm。」

"Dude, it's great to be in Stockholm."

「在 Stockholm 录节目我也很激动。」

"I want to start — you just recently raised a great round, and I want to start with that. We're seeing a lot of money going into the space. Is it a capital arms race and a case of who has the most money wins, or is it something else?"

「我想从你最近这一轮融资开始。这个赛道现在钱真的非常多。这到底是一场谁钱多谁赢的资本军备竞赛,还是别的什么?」

"I think it's an arms race to build the best team, and then it's an arms race to build the best brand and trust from your users. Capital can help. For us, it's not a constraint at all."

「我觉得这是一场组建最强团队的军备竞赛,然后才是建立最强品牌、赢得用户信任的军备竞赛。钱能帮上忙,但对我们来说完全不是瓶颈。」

"If you're building something like the best foundation model, it might be a constraint just because the compute for training and so on is so so large. But for us, it's all about moving extremely fast and collecting the best talent."

「如果你做的是像基础模型那样的东西,资本可能会是瓶颈——训练算力实在太大了。但对我们这一层来说,关键就是跑得极快、把最好的人聚到一起。」

"So if we think about talent as the number one there — we've seen Zuck pay NFL-style contracts. I mean, mega mega sums for the best people. How do you think about and analyze that, and how difficult will it be to get the best talent moving forwards?"

「那如果说人才是头等大事——我们看到 Zuck 在开 NFL 合同那种级别的薪水,真的是天价砸最顶尖的人。你怎么看这件事?未来抢人会不会越来越难?」

"For me it's actually more difficult than for Zuck to know which engineers are going to really thrive, push the culture forward, push the ways of working in the products forward."

「对我来说,其实比 Zuck 更难——我很难提前知道哪些工程师会真的爆发出来、把团队文化和产品工作方式往前推。」

"For Zuck, it's like there's this 10 people that know everything about how to train foundation models, and he's more paying for that knowledge than for these people. The talent itself is so good, it's probably pretty good as well. So it's very different."

「Zuck 那种情况是:全世界就那 10 个人知道怎么训练基础模型,他付的其实是那份知识的价钱——人当然也牛,但他买的是那份独家 know-how。所以两件事完全不一样。」

"Do you think you do not need the same caliber of engineering talent if you're working in the application layer?"

「那你的意思是,做应用层不需要同等级别的工程师?」

"It's just very different. I think one of those people Zuck is hiring — they wouldn't perform as well as the engineers in my team doing what we're doing. So it's a very different type of talent."

「就是不一样的人。Zuck 招的那种人,真把他丢到我们团队里干我们在做的事,大概率不会比我团队里的人表现更好。是完全不同的人才类型。」

"If I knew who was the perfect engineers to hire, I could maybe step up our compensation bands to get exactly those. But I don't know who are the best people. So I need to figure out: are these really really good people to work with? Are they moldable? Are they going to work well together in this team?"

「如果我知道谁是完美人选,那我可以直接把薪资包提上去把他挖来。问题是,我事先并不知道谁最好。所以我得当面去判断:这是不是一个真的能共事的人?他可不可塑?他能不能融到这个团队里?」

"And then give the compensation that you give on the top of market compensation rates for that."

「判断完之后,再给市场顶部水平的薪酬包。」

"You've built an incredible team — also of less obvious talent in the early days, and then you hire amazing rock stars like Elena Verna. When you look at your hiring process, is there anything that's non-obvious?"

「你的团队搭得很强——早期是一些不那么显眼的人,后来又请到了像 Elena Verna 这种顶级明星。在你的招聘和人才评估里,有没有哪一条是反直觉的?」

"So for us, I look for people who have either extreme trauma or extreme masochism."

「我自己嘛——我找的是要么有过极端创伤、要么是极端受虐倾向的人。」

"Yes."

「(笑)对。」

"Being serious — I think not enough people are opinionated. You said brand is important. Great brands are opinionated. People love them or hate them. Lovable — good example. But what is yours that is non-obvious about hiring or talent assessment?"

「正经一点说,我觉得真正"有 opinion"的人太少了。你说品牌很重要——好的品牌都是有立场的,要么被爱要么被恨。Lovable 就是个好例子。那么在招聘和人才评估上,你有什么反直觉的判断标准?」

"I like to think a lot about slope. If I talk to someone and I learn a lot of things from them, and I notice that my conversation is very dynamic and exciting — that usually feels like a very good indicator that they're going to adapt to the organization, and their slope will be very high."

「我特别看重一个词,叫"斜率"(slope)。如果跟一个人聊一小时,我能从他身上学到一堆东西,聊得很跳、很刺激,这通常就是一个非常强的信号——这个人会很快适应组织,他的成长斜率会很陡。」

"Otherwise, I think there are good ways to just understand how did they perform — like if I could be there with a video camera when they worked in the past, that gives me a lot of signals. So that's usually what I spend a lot of time on when talking to new candidates."

「除此之外,我会很想知道他过去到底是怎么干活的——如果能带个摄像机回到他之前的工位拍一周,那信号会非常多。所以聊新候选人的时候,我大部分时间都花在挖这个上面。」

"When you think about a slope — it's noticeable with you. We haven't known each other for a huge amount of time, but when I compare when I first met you to when I met you today, it is still very different in terms of your leadership. Where have you not progressed where you would like to still?"

「斜率这件事,你自己身上就很明显。我们认识也不算很久,但我把第一次见你和今天见你对比一下——你的领导力差别非常明显。那有没有哪些地方你觉得自己还没有真的进步?」

"I still operate Lovable in a very scrappy, startupy way even though we're at the later growth stage right now. So adding a bit more structure in a few key areas is somewhere I'm looking to progress — to start being an excellent operator."

「我还是用一种很糙、很 startupy 的方式在跑 Lovable,虽然我们其实已经到 later growth stage 了。所以在几个关键的地方加一点点结构,是我接下来想往前推的——往"优秀运营者"的方向去走。」

"Do you think you actually need that? We've had founder mode be so propagated and praised — and being close to the metal, Jensen [Huang] having 52 direct reports. I would say structure and that middle layer is where slowness and apathy come."

「你真的需要那种结构吗?现在大家都在吹 founder mode、要贴近一线——黄仁勋(Jensen)直接管 52 个人。在我看来,真正让组织变慢、变冷漠的,就是结构和那个中间管理层。」

"True, yeah."

「确实是这样。」

"Do you think you need that?"

「那你觉得你需要那一层吗?」

"That's a good question. I think I'm going to always operate with most of my impact coming from founder mode."

「好问题。我大概率永远都会以 founder mode 为主——我大部分影响力都从那儿来。」

"But I do need — given that there's so many things thrown at me and coming in from all the different directions — to have a kind of protective layer that introduces a lot of order in how we prioritize all these incoming things. And that comes down to a well-running organization."

「但是,因为现在每天从四面八方砸过来的事情太多,我确实需要一层"保护层"——它的作用是把这些纷飞的请求理出优先级。这其实就是一个运转良好的组织的事。」

"And for a well-running organization, you need a very organized manager somewhere at the top. And I'm not planning to be that percentile manager myself — but surround myself with great leaders who do more of the organizing."

「而要让组织运转得好,顶上得有一个非常有条理的管理者。我不打算自己去做那个"管理力顶尖百分位"的人,我的策略是身边围一圈优秀的 leader,他们去把组织管起来。」

"Do you have a protective layer today? Because I get probably 25 intro requests for you a week, and I probably make one a month. Do you have someone who does the filter?"

「你现在已经有这个保护层了吗?我每周大概会收到 25 个想要 intro 你的请求,我大概一个月才转给你一个。你有人在帮你过滤吗?」

"Yeah, I do. It's a kind of wonderful chaotic protective layer that works together as a close team. I don't really have a name for it — it's just the people working closely with me."

「有的,有这么一层。其实它有点"美好又混乱",几个紧密合作的人组成的小团队。我也没给它起名字,就是那些跟我贴得很近的人。」

"The team is made up of previous-founder-type generalists that work closely with me. And I just work in terms of quick feedback: 'this is not what we should be doing, this is what we should be doing.' Works okay now. I think we can do even better."

「这一层全是之前自己当过创始人的通才。我跟他们的协作方式就是快速反馈:"这件事别做,这件事去做。"现在跑得还可以,但我觉得还能更好。」

"So you said talent was number one and brand was number two. If we think about great brand — what does great brand mean to you?"

「你刚才说人才是第一,品牌是第二。那"好品牌"对你来说,具体是什么意思?」

"Super concrete example is the Apple ecosystem, where they obsess about details — maybe too much, so they move slowly. But that's what builds up trust and a very strong brand."

「最具象的例子就是 Apple 生态——他们对细节非常较真,可能较真过头了,所以走得很慢。但正是这种较真,堆出了信任,堆出了一个非常强的品牌。」

"That's what we're aiming for as well in every interaction. Every time we update the product, how do we make sure we roll it out so that we really understand the users and how their reactions to all the things we're changing very rapidly in the product, in the company."

「我们也在朝这个方向走——每一次产品更新,每一次和用户的接触,我们都在问:这一次发布之后,我们到底有没有真的搞清楚用户对我们快速改的东西怎么反应。」

"There's a couple of questions which everyone has, where they will throw them as a critique at Lovable or at anyone in the space. One is protection — defensibility. When you think about defensibility today, is brand the core element of defensibility, or is there something that people do not see?"

「现在外界对 Lovable、或者整个赛道,有几个共通的质疑。第一个是"保护"——也就是护城河。在你看来,品牌是护城河的核心吗?还是有别的东西大家没看到?」

"I think you need to build a product, if you want to maximally be defensive — where if you are on this product and the platform that that product is, you don't want to leave, because you have so much value that you've created on the platform that you're getting automatically every day."

「如果你想真的有护城河,你得做出一个这样的产品:用户用着用着就不想走了——因为他在你这个平台上累积了一堆价值,这些价值还在每天自动复利地为他工作。」

"That's what Lovable is becoming in this product building platform. Lovable today is your technical co-founder. We want it to be your co-founder in general, that handles all the admin, setting up your finance, operations. And if you're on a platform like that, you probably don't want to leave."

「Lovable 现在正在变成这样一个"造产品的平台"。今天的 Lovable 是你的技术合伙人——但我们要让它变成你 general 意义上的合伙人,把行政、财务、运营这些 setup 都接掉。当你在这样一个平台上,你大概率不会想离开。」

"Would you say to all founders building an AI startup from day one — don't worry about defensibility, it comes over time?"

「那你会不会跟所有从零做 AI 创业的人说:第一天别想护城河,它是后来才有的?」

"Yes — great question. I have a friend who has this fun analog in terms of an AI startup, which is that AI startups are like chickens shot out of a cannon up in the sky, if you start getting traction."

「会。这是个好问题。我有个朋友给 AI 创业打过一个特别有意思的比方:AI 创业公司就像一只只被炮射上天的鸡——前提是你有点起势头。」

"And then it's all about flapping fast as a chicken because there are new chickens shot out from cannons every day. And if you keep flapping faster than the other chickens, then you're going to do great. And I think that's a good first level of analysis in how you should operate."

「然后接下来你要做的就是拼命扇翅膀——因为每天都有新的鸡被炮射上来。只要你扇得比别的鸡快,你就能飞得很好。我觉得这是一个不错的第一层分析,你应该照这个去操作。」

"I'm just going to say for any vegans that are listening, no chickens were shot out of cannons. And that is the most extremely Swedish way of — you know, it's Reid Hoffman who says, it's about kind of running off the cliff with a power glider and just kind of flapping. That works too."

「先给所有听节目的素食主义者说一句:没有鸡真的被炮射出去。你这个版本特别"瑞典"——其实 Reid Hoffman 也有个类似的说法:从悬崖跳下去,一边滑翔一边在空中把翅膀长出来。差不多一个意思。」

"Yeah, I think that's my recommendation — just be like execute fast, grow faster. And then when you're starting to get up there, you can maybe start thinking a bit about the defensibility."

「对,我给的建议就是这个——执行快,增长更快。等你飞到一定高度之后,再开始想护城河的事情也来得及。」

"Totally get you. That's the one criticism."

「完全 get。那是质疑之一。」

"Another is — actually when you look at these businesses, a lot of people are criticizing this with your Replits, your Bolts, your Lovables — they're not actually very good businesses in terms of unit economics. So much is passed through. So like, bluntly: if I give you a dollar, how much is passed straight through to Anthropic and OpenAI?"

「另一个质疑是这类生意——Replit、Bolt、Lovable——其实单位经济很差,大部分钱都"穿透"出去了。说白点:我给你一美元,有多少直接传到 Anthropic 和 OpenAI 那边?」

"I don't give you the exact numbers, but if you look at the paid usage, it's majority. It's not everything."

「具体数字我不能讲,但如果只看付费部分的用量,大头是过给他们的——但不是全部。」

"How does that change over time?"

「这个比例随时间会怎么变?」

"As our business develops, we're looking to get most of our revenue once you as a user are like 'I love this platform, I'm never leaving.' But today, in the beginning, you're paying to build, pretty much."

「等我们这盘生意继续往下走,我希望大部分收入是来自这种状态:用户已经"我爱这个平台,我永远不走了"。但今天在最早期,用户付费基本就是在为"造东西"付费。」

"So over time we just want to create so much value — you stay on the subscription, and a small part of the cost goes to AI compute."

「往后我们想做到的是:制造的价值大到你愿意一直续订,而其中只有一小部分钱真正流去支付 AI 算力成本。」

"Will you be able to make money through, you know, not optimizing models? What I mean is — in the future, you may not need the very best, very latest model to do the simple about-me website. And so you can route users."

「未来你能不能靠"路由"挣钱?我的意思是:做一个简单的个人主页,可能根本不需要最新最强的模型——你可以根据任务把用户分到便宜的模型上。」

"Yeah. As all applications develop, the AI is going to be adapted to those applications. For most things it's super simple — it's like you're driving a car and you're not thinking about what you're doing. When you're in a new situation driving, then your brain really goes on fire. We're not there. We're not close to being there yet."

「会的。等所有应用都成熟之后,AI 会被定制到这些应用里。绝大多数事情其实很简单——就像你开车开惯了,根本不用想。只有遇到全新的路况,大脑才真的烧起来。但 AI 现在还远远没到那个阶段。」

"For us it's too early to optimize for that, because the AI every month is doing new completely different things. So we just want to be able to iterate really fast on what the AI is able to do, and not optimize the models for what it's doing."

「对我们来说,现在做这种优化为时尚早——AI 每个月能做的事情都跟上个月完全不一样。所以我们要做的是"在 AI 能做的事情上快速迭代",而不是为它现在能做的事去微调模型。」

"That's really interesting. So you build for what tomorrow's model can do, not what we have today."

「这个挺有意思——所以你是为明天的模型能做的事造产品,不是为今天能做的事造。」

"Yeah, to quite large extent. Generally when I think about models — they're the ones that are very thoughtful and deep thinking. Right now we put as much of the work onto those models."

「相当大程度上,是的。我现在脑子里"模型"=那种很会思考、很深度推理的模型——我们现在把尽可能多的活儿压给它们做。」

"In the future it's going to be a mix: when it's obvious what you should do, then it doesn't cost any money, it's super fast. But when it's a new situation — which building a software product often results in — then it has to think much more."

「未来会是一种混合:遇到显而易见的情况,模型几乎不花钱、瞬间就出。但遇到一个新场景——而做软件产品又非常容易遇到新场景——那个时候模型就得真的认真想一想。」

"One other area where you can see real margin expansion is also in token selling. When you think about how you price tokens — given prosumers and consumers don't fundamentally know the price of tokens — you can actually have quite a considerable markup on token usage. Do you think that is a place of real elasticity to gain margin or not?"

「还有一个能扩张毛利的地方,是直接卖 token。普通用户根本不知道一个 token 多少钱,所以这里其实有不小的加价空间。你觉得这会是一个有弹性可挖的毛利来源吗?」

"Yeah. We looked at — this was a few months ago — how much revenue is flowing through the AI from Lovable applications. And it was more than $10 million in ARR."

「是的。我们几个月前算过一次:Lovable 上跑的应用里,有多少收入是经过 AI 在转的——结果是超过 1000 万美元 ARR。」

"All of that revenue requires the user to go through a bit-complex process of setting up the connection to the model providers. That's something we're just simplifying — stay tuned for how we enable more simplicity first of all for our users. And then if we can reduce the underlying cost, maybe we can take a margin there as well."

「但问题是,这些收入要走通,用户得走一套挺繁琐的流程去连模型方。我们正在把这一段做简单——这块怎么往前推,后面会有动作。如果我们还能把底层成本压下去,可能就在这里也能拿一点毛利。」

"How do you think about mental plasticity to delay margin optimization?"

「你怎么看"心理弹性",也就是愿意把毛利优化往后推这件事?」

"Mental plasticity what?"

「心理弹性是什么意思?」

"The willingness to wait for margins to come. Look at Deliveroo — margins are [shit] in the early days, and over time they get better and better as you have more and more people use it, more density, more orders in small areas. You've got to be patient so to speak. Same with OpenAI, same with Lovable. How long does one think before you're thinking margin optimization?"

「就是"愿意等毛利"。看看 Deliveroo——早期毛利惨不忍睹,但用的人越多、订单密度越高,毛利就一点点变好。你得有耐心。OpenAI 是这样,Lovable 也是这样。那要等多久才该开始想毛利优化?」

"I have two conflicting pieces of perspectives on it. One is — I speak to Nik who built Revolut, and he just tells me 'Anton, you need to compute the payback time, and then you need to do super hard performance optimization on acquiring new users. And of course you need to have good payback times in terms of profits per user.' Which makes sense."

「我心里有两套相互矛盾的视角。一种是和 Revolut 的 Nik 聊出来的——他直接跟我说:"Anton,你必须把回收期算清楚,然后在拉新这件事上做极度严苛的转化优化,每个用户的利润回收期得是健康的数字。"这话当然是对的。」

"And if you can do small changes in margins, it actually affects a lot on how fast you can grow."

「而且毛利哪怕微调一点点,对你的增长速度影响都非常大。」

"But the other perspective, which I index a bit more on right now, is: you just want to have as much mind share and as many users who just love the brand as possible right now. And then you can think about that later."

「但另一种视角——我现在权重押得更重的那一种——是:你眼下要做的就是抢心智份额、让尽可能多的用户真心爱上这个品牌。毛利的事可以再往后想。」

"Exactly how I trade those two perspectives off — I mean, you need to look at the weights in my neural network — but it's some combination of the two."

「至于这两种视角具体怎么权衡——那得看我脑子里的神经网络权重是什么样,反正最终就是两者的某种加权。」

"It's so funny. I think it's absolutely number two. And then it's number three — which is Nik's — incredible performance optimization, really understanding funnel metrics from CAC to LTV, how you drive efficiency through the channels."

「太有意思了。我同意应该是第二个先来。第三阶段才是 Nik 那一套——极致的性能优化、从 CAC 到 LTV 把漏斗每一格摸透、把渠道效率压出来。」

"And you know what's ironic? You kind of go back after that stage to an art — which is where Nik is now — which is: we've done that so well, we have to sponsor race cars, because brand again becomes the most important thing."

「然后讽刺的是,做完第三阶段之后又会绕回到第一阶段那种"艺术"——Nik 现在就在这个阶段——把数据做到极致以后,他得去赞助 F1 赛车,因为品牌重新变成最重要的事情。」

"It's so interesting — that kind of bell curve where you see it go up and then come down again. You have it at both ends of the spectrum on brand."

「太有意思了——一个钟形曲线,品牌的重要性两头高、中间低。」

"I think if you can do everything at once, the company benefits a lot. But you usually can't — you should be focused."

「如果你能同时把这些都做了,公司当然会受益巨大。但通常做不到——你应该专注。」

"What would you most like to do now that you're not doing or can't do?"

「你现在最想做但还没做、或者还做不了的事是什么?」

"I would like to rethink how the applications are built — like what is the best way to build an application. What Lovable does now is take all the best practices from decades of how a great software product was built. But that's not how the future is going to look like."

「我很想从根上重新想"应用应该怎么造"——什么才是构建一个软件产品的最佳方式。Lovable 现在做的是把过去几十年"好软件怎么造"的最佳实践搬过来。但未来的软件长得不会是那样。」

"All software applications are going to have some type of AI. They're going to have extremely seamless payment and checkout flows. That's something I'd love for us to spend time on — figuring out and making possible for our users not just to have a superhuman AI engineer, but to have an AI engineer that builds the future of applications."

「未来的所有软件都会自带某种 AI;支付和结账流程会变得极致顺滑。我特别想花时间想透这一层,然后让我们的用户拥有的,不只是一个"超人级别的 AI 工程师",而是一个能造"未来形态的应用"的 AI 工程师。」

"We saw OpenAI kind of suggest or proffer Lovable-style competitors. To what extent do you feel OpenAI and Anthropic will come after Lovable in the way that Claude Code comes after Cursor?"

「我们看到 OpenAI 在向"Lovable 风格"方向暗暗动作。你觉得 OpenAI 和 Anthropic 会不会像 Claude Code 之于 Cursor 那样,直接对 Lovable 出手?」

"In the long term it just comes down to execution of a team. Many people are going to offer what we're offering today. We just need to offer much more when that time comes, and have a better user experience, give a better value proposition to our customers."

「长期看,这件事归根结底就是团队执行力。今天我们提供的东西,未来一定会有很多家也能提供。等那一天真到来,我们要做的是提供得多得多——更好的用户体验,更强的价值主张。」

"Who do you worry more about — OpenAI doing it or Anthropic doing it?"

「OpenAI 出手 vs Anthropic 出手——哪个更让你担心?」

"What we're betting on is to be the gateway for humans, right? And be the best user experience for AI. So far, OpenAI is doing that better than Anthropic. So I see them as a more serious competitor in 12 months."

「我们押的是"人类用 AI 的入口"——给 AI 做最好的用户体验。截至现在,OpenAI 在这件事上做得比 Anthropic 强。所以未来 12 个月,我把 OpenAI 视为更值得警惕的对手。」

"How did you analyze GPT-5? When you look at performance, are you more or less bullish on OpenAI?"

「你怎么评估 GPT-5?看完它的表现之后,你对 OpenAI 是更看好还是更不看好?」

"We looked at a lot of how GPT-5 would impact our users before we decided 'okay, let's put this into the product.' We looked at how long it took to get responses, our quantitative evals, and then we vibe-checked it in many different ways. What we concluded was that it's often times too ambitious for our users."

「我们在决定要不要接进产品之前,做了大量评估:回答耗时、定量 evals,然后再以各种方式做"感觉测试"(vibe check)。结论是:对我们的用户来说,GPT-5 经常表现得"太想做事"。」

"That's why we decided 'this is very smart, so let's give it to all our users and see what they tell us in terms of what's good and what's bad.' What we found was that for the use cases when you have to solve a really really hard problem, it's great."

「我们最终的决定是:既然它很聪明,那就开放给所有用户、看他们告诉我们什么好什么不好。我们发现的是:在"必须解决一个真的很难的问题"这种场景下,GPT-5 很强。」

"In terms of 'is OpenAI doing a great job' — I think this was a really smart, obvious choice for them: 'we have five different models that you have to select in ChatGPT, let's just bring it down into one model, GPT-5.' They definitely should have done that. But it comes with a lot of trade-offs. So far I'd say they executed pretty well. The model is still too ambitious."

「至于 OpenAI 自己干得怎么样——把"ChatGPT 里要让用户在 5 个模型之间选"这件事直接简化成"就一个 GPT-5",这是个非常聪明、非常应该做的产品决策。但代价不小。整体来看他们这一仗打得不算差。但这个模型本身,还是太想做事。」

"There's also a question of when you set the bar at AGI — and then you get model optimization and model routing essentially, which is what it is. Capability-wise, it hasn't been a step-function improvement in what we had before."

「还有一个角度:你既然把目标定在 AGI,那 GPT-5 本质上就是模型优化加模型路由——它就是这么个东西。论能力,它和之前的版本相比并没有阶跃式提升。」

"No. The biggest part of GPT-5 that's disappointing is that now they have to optimize all these different things into one model. Before it was different models and they had to do it really fast. So it's inevitably going to fall short in some dimensions. And it just is a disappointment that you can't improve in all the directions at the same time."

「对。GPT-5 最让人失望的部分是:他们现在被迫把好几个不同维度的能力一起塞进同一个模型里调。之前是几个分开的模型、各自快速迭代。现在合到一起,就一定会在某些维度上掉链子。所以让人失望的点其实不是模型不行——是"不能在所有方向同时进步"这件事本身。」

"How do you use OpenAI versus Anthropic within Lovable today?"

「Lovable 现在内部是怎么用 OpenAI 和 Anthropic 的?」

"We have this very complex agentic chain where we pass the user's response, the application information, through many different models. We take really fast and small ones, and then for code writing we usually use Anthropic. And right now you can say 'I want to use GPT-5,' and that's better when you're solving a really hard debugging problem."

「我们有一条挺复杂的 agentic chain,会把用户的回复、应用上下文穿过好几个不同的模型。前面会用一些非常快、非常小的模型;到写代码这一步,我们通常用 Anthropic。现在用户也可以指定"我要用 GPT-5"——遇到真的非常难的 debug 问题,GPT-5 会更好。」

"Super. And you've seen it be better than Anthropic when it comes to a hard debugging problem."

「不错。所以在硬核 debug 这件事上,你确实看到 GPT-5 比 Anthropic 强。」

"Yeah."

「是的。」

"What do models not do today that would be a step-function change in what Lovable can do?"

「今天的模型还做不到什么——一旦做到就会让 Lovable 跨越式跳一档?」

"Something I'm super excited about is that the AI has more context about who they're talking to and how they should be answering, to guide them through our specific application."

「我特别期待的一件事是:AI 能多掌握一点关于"它在跟谁讲话、应该怎么回答"的上下文,这样它才能在我们这个特定的产品里把人带着走。」

"Solving that problem is something we have to do — both in how we build this agentic chain, and over time in building absolutely world-class teams, paying $100 million for the people that train the models. So that's on the horizon for us, to get it to be hyperpersonalized for you specifically."

「这件事我们必须自己去解。一方面靠 agentic chain 怎么搭;另一方面就是长期投入——花 1 亿美元去抢能训模型的人,搭一支真正世界级的团队。这是我们眼前的下一个山头:把它做到针对每一个具体的"你"做超个性化。」

"You recently announced 100 million ARR — amazing milestone to hit in 7 months. That was [shit] nuts. For years it was 0 to 10 million in two years was the gold standard. That's what I was brought up on, which makes me feel really old."

「你不久前宣布 100M ARR——7 个月做到,这数字太离谱了。我那一代人是在"两年从 0 做到 10M 是黄金标准"里长大的——讲出来就显得我特别老。」

"My question is — when you look at revenue breakdown of the 100 million, just guesstimate: what is split between hobbyists, pro devs, normal people? How does it fit between the different segments?"

「我的问题是:这 100 million 收入,你大致估算一下,在兴趣党、专业开发者、普通用户之间怎么分?」

"People do everything with Lovable. They come with their idea to build a software business and product. There's a lot of people in large companies that use it as 'now I can prove, show what I actually think we should build,' and then they build a working product that they can decide 'are we going to give it to our engineering team and they actually implement it?'"

「大家在 Lovable 上什么都做。一类是带着想法来做一个软件创业产品的人。还有一类是大公司里的人,把它当成"我可以亲手证明我们应该做什么样的产品"——他们做出一个可以跑的产品,然后再决定要不要把它交给工程团队真的去实现一遍。」

"And then there's everyone else who builds their personal website, their small business website in a few minutes. And 80% of people are in the first category — they're building real complex applications."

「再剩下的就是几分钟做出个人网站、小生意网站的那群人。但 80% 是第一类——他们真的在做复杂应用。」

"In terms of revenue — 80%?"

「按收入算——也是 80%?」

"Yeah. The second segment is actually growing very fast because enterprises are slower to wake up. But you might have seen this product leader from Google who says 'we're never again writing a document about a product — we have to use Lovable or something to build out a fully working demo.' So that use case is also growing very fast."

「对。第二段(企业用户)其实增长非常快——大公司觉醒慢一些,但你大概也看到 Google 的一位产品 leader 说过:"我们以后再也不写产品文档了,我们必须用 Lovable 之类的工具直接搭出一个能跑的 demo。"这个用法增速很快。」

"And the third use case — a lot of people have been burned trying to build nice websites in these no-code website builders like Squarespace. If you can just always do everything in Lovable with a UX that's more sophisticated and moving fast, that's also growing. But the first two are the ones who are really game-changers."

「第三段——很多人在 Squarespace 那种 no-code 工具上吃过亏。如果你在 Lovable 上能用一个更聪明、更快的 UX 把事情做完,那也是个不错的体验。这块也在涨。但真正能改变游戏规则的,是前两段。」

"So 80% is actually building complex apps. 10% is enterprise and 10% is hobbyists."

「所以 80% 在做复杂应用,10% 是企业,10% 是兴趣党。」

"Something like that, yeah."

「差不多就是这个样子。」

"Is that what you want it to be?"

「这就是你想要的结构吗?」

"We want to build for the new generation of AI-native founders who build maybe one-person unicorns soon. The funny thing is, those people also have jobs maybe in large successful companies, and they want to help their friends and their family to build simple websites. So I think this is a good split."

「我们想为新一代 AI 原生创业者服务——他们很快可能就是"一个人独角兽"那种状态。有趣的是,这些人本身就在大公司有工作,他们也愿意顺手帮家人朋友做个简单网站。所以这个分布对我们是好的。」

"Is that an optimal market to go after if you're thinking — god, I sound like such a VC — value extraction? An AI founder building a mega business on Lovable, great. But you have to have a lot of mechanisms for value extraction — payment solutions, you name it. But if they're single-seat, it's tough to get true value extraction. Is it not much better to be hobbyist for everyone? For mom and pop to build the about-me website, where it's 7 billion people?"

「但这真的是最优 TAM 吗?——天哪我现在听起来太像个 VC——但纯从"价值榨取"角度来看:一个 AI 创业者在你这上面做出大生意,这很好,但你得有一堆机制把价值转化成收入(支付方案之类),而且如果他们是单人席位,真的难抽得动。是不是反而做"全人类的兴趣党"更好?爸爸妈妈做个个人简介页,这是 70 亿人的市场。」

"Our mission is to enable a lot of people that have the opportunity to build businesses, but have been held back by not being able to write code and have access to capital to hire engineers. So it's obvious to start with the people who are going to build businesses. Then it naturally trickles down to everyone else as a function of that."

「我们的使命是:让那些本来有机会做生意、却因为不会写代码或没钱雇工程师而做不了的人,有机会做出来。所以从"会真的做生意的人"开始切入,逻辑非常顺。等这一群人服务好了,服务自然会往下渗透到所有人。」

"Those are the best people to start building for. Where you can extract value — I think less about that. I think about our mission."

「他们是最好的早期用户。"在哪儿榨价值"这件事我想得不多。我更多想的是我们的使命。」

"Let me expand a bit on the business thinking. Many of the largest businesses haven't been started yet today. With AI you can move so fast, you can be much closer to your customers, you can drive prices down. We want to be the enabling tool for that movement."

「我多说一点商业逻辑。今天最大的那些公司,很多还根本没成立。有了 AI,你跑得更快、离客户更近、能把价格压下去。我们想做的就是那场运动的工具。」

"If we can do all these use cases at the same time, what our use case percentage of adoption or revenue is going to converge towards is just: what's the spend from enterprises, what's the end-game spend for consumers on tools like this — if we continue to dominate this completely new category."

「如果我们能把这些场景同时吃下,最终的收入分布会自然收敛到:企业一年花多少、消费者一年花多少在这种工具上——前提是我们能在这个全新的品类里持续领先。」

"Obviously I'm super grateful for you taking my money. But the number-one reason why I would invest in Lovable is because the market TAM — or the TAM expansion — is actually incomprehensible. Which is very much like Uber: you could never have foreseen the market expansion that would take place. Very much like Lovable — saying 'website builders is an X market' is completely the wrong analogy to understand how big Lovable could be moving forwards. And that's a common thing of the best venture investments ever made."

「先谢谢你愿意拿我的钱。但我投 Lovable 的第一原因是:这个市场的 TAM 扩张是"不可理解"的。就像当年的 Uber,你根本无法预测它的市场会变多大。Lovable 也一样——把它放在"网站搭建工具"这个市场里去比,根本是错误的类比。这是历史上所有最伟大风险投资的共同特征。」

"Yeah. I also want to say on the enterprise use case — if I was a CEO or CTO of a large enterprise company, I wouldn't think in terms of 'oh, how can we make our engineers more productive?' I would think 'how can we get the most information about what we should build as quickly as possible into new products or into existing products?'"

「我再补一句关于企业的用法:如果我是一家大公司的 CEO 或 CTO,我不会去想"怎么让我的工程师更高产"。我会想:"怎么尽可能快地把'我们应该做什么'这个判断转化为产品上的实际改动?"」

"That requires everyone in the company to be able to work in one place to change and edit their products and propose new changes. It's hard for us — or anyone — to build a product that does that, like, tomorrow. It's better for us to start with the founders who are building from the ground up, and then move the same experience into the enterprise."

「要做到这件事,公司里每个人都得能在同一个地方修改和提议产品改动。让我们今天就在大企业里直接搭出这种东西,挺难——任何人都难。所以更合理的路径是:先服务从零开始的创业者,把这套体验跑成熟,再原样搬进大企业里。」

"It also allows for this incredible democratization of ideas within companies. I interviewed the CPO at Duolingo for 20 Product, and he said how two designers — not traditionally ones who come up with products from day one — created chess in Duolingo. They did it with — I can't remember what the tool was. I hope it was Lovable. Please say it was Lovable."

「这同时也带来一种"想法的民主化"。我之前在 20 Product 采访 Duolingo 的 CPO,他讲过:Duolingo 里的国际象棋功能是两个设计师做出来的——他们传统上不是那种"第一天就在做产品"的角色。他们用了——我不记得用的什么工具,但愿是 Lovable,你跟我说是 Lovable 吧。」

"That was actually their first iteration of it, which I thought was an amazing instantiation of this. So I totally get you there."

「那是他们的第一版——我觉得这是这种"民主化"的一个极佳例子。所以我完全 get 你的逻辑。」

"Does that mean then that we lose the design process and the brainstorming process? Do we skip that and go straight to prototyping?"

「那这是不是意味着,设计流程和脑暴流程会消失?直接跳到原型?」

"To date, what you've done is take an idea and go through many, many steps until it's a fully fast-growing product. I call that the product life cycle. One part is writing the code, where AI has now made it much faster. There are many steps after that. There are steps before that — mock it up, validate it internally, validate it with users."

「过去这套流程是:从一个想法开始,走一长串步骤,最后变成一个真在快速增长的产品。我把这一整段叫做"产品生命周期"。其中"写代码"只是中间的一段,AI 已经把这一段大大加速。后面还有很多步骤,前面也有很多步骤——画原型、内部验证、跟用户验证等等。」

"What we've done so far is take all the first steps — until 'this is validated, this is what we need to ship,' even getting external users on it — into a few minutes or a few hours of building. So that's where there's the most maturity on our product."

「我们目前主要做的事是:把"想法 → 验证 → 该上线了"这前半段——甚至包括把真实外部用户拉进来试用——压缩到几分钟或几个小时之内。这一段是 Lovable 现在最成熟的部分。」

"The steps that come after, we have to build out as fast as possible, so that you don't need a product / design / engineering organization. It's all one tool where anyone with the best ideas spends the most time."

「后半段——我们必须尽快把它也做出来。最终的目标是:你不需要再有"产品-设计-工程"这套分离的组织,而是一个工具——谁的想法最好,谁就在这上面花的时间最多。」

"What I mean by the steps before — you see at the other end of the spectrum, Figma doing Figma Make, where they're like 'we'll always have the design process, and then we'll move with you into the second phase, the prototype phase, from design to prototype.' To what extent do you worry about entry from that earlier standpoint?"

「"前半段"我特指:Figma 那边在做 Figma Make,他们的姿态是"设计流程永远会在,我们陪你从设计走到原型这第二阶段"。你担不担心 Figma 从设计这一头反向打过来?」

"I think humans are sometimes too obsessed with small details being perfect, which makes you move much slower. The way it's done now — where one person does all the design very slowly, very detailed — is going to be replaced by AI doing the implementation."

「我觉得人类有时候对小细节太完美主义,这会让你慢很多。"一个人慢慢、精雕细琢地把所有设计做完"这种工作方式,未来会被 AI 接手实现层。」

"You talk much more high-level — you talk about your design philosophy — and then the AI does the implementation. Then you as a human go out and get all the context from other people 'this looks good,' and give it all as feedback into the AI."

「你跟 AI 沟通的层次会高得多——你讲你的设计哲学,AI 来做具体实现。然后你作为人去外面收集反馈("这个看起来不错")再喂回 AI。」

"Then there's a very seamless, opinionated way of taking it all the way to a product, with all the marketing and growth functions and built-in AI behind it, as well as all the tooling you need to evolve high-quality software — testing, quality assurance and so on. That's a new way of doing it. Thinking you should be doing design with [Figma] Make — for me, my thesis is that it will slow you down too much. Does that make sense?"

「之后还有一条非常顺滑、非常有立场的路径,把它一路推到完整的产品——市场、增长功能都内嵌 AI,再加上演化高质量软件需要的全部工具(测试、QA 等等)。这是一种新的工作方式。"应该用 Figma Make 做设计"这种想法——我的论断是:它会让你慢得多。这个逻辑能 follow 吗?」

"It does. What happens to Figma then?"

「能 follow。那 Figma 会怎么样?」

"For some pixel-perfect things, it's going to be amazing to continue to use Figma. I don't know how the distribution will look like in terms of doing it with a more opinionated way — which is what our platform is becoming — versus how many companies want to continue to do it like you do it now in a tool like Figma."

「对于一些必须像素级精修的场景,继续用 Figma 是非常好的。具体的市场分布我也不知道——多少公司会选 Lovable 这种"opinionated platform"路径,多少公司会继续按现在 Figma 那套来做,这个比例我没法预测。」

"Do you think you're opinionated enough?"

「你觉得你们 opinionated 得够吗?」

"Our tool enables a lot of flexibility at the cost of some velocity on our product development. But it's a pretty good sweet spot. You can build with Lovable, and then any engineer can come in and edit and take over if they want to. So yes, we would be moving faster if we were even more opinionated about how things should be done."

「我们的工具其实非常灵活,代价是我们自己的产品研发速度会慢一点。但这个权衡点我觉得不差:你可以在 Lovable 上 build,然后任何一个工程师都能进来接手编辑。但话说回来,如果我们再 opinionated 一点,我们会跑得更快。」

"What are you not opinionated on that you would like to be, in a dream world?"

「在理想世界里,有哪些地方你想更 opinionated 但目前没有?」

"In a dream world, we would be even more opinionated about how you build an application — and we would know what's the future of building applications with AI being such a core part of it. I don't think it's possible because how AI works and what's the best UX with AI for the products that are built with Lovable changes so rapidly. So at some point in the future, I'd love to be there."

「理想中,我们会在"应用应该怎么构建"上更 opinionated——而且我们会知道"以 AI 为核心的应用未来长什么样"。但我觉得现在做不到,因为"AI 是怎么工作的"以及"用 Lovable 做出来的产品最佳 UX 是什么"都在飞速变化。等到某个时间点,我希望能到那一步。」

"When we can be more opinionated, you get the right level of detailed adjustments on how the AI works for your product, what backend flows and workflow automations work for your product. Now we support a lot of different things. So you need to be really good at prompting right now."

「等我们可以更 opinionated 之后,你能在合适的颗粒度上调整"AI 在你产品里怎么干活、后端流程怎么走、自动化工作流怎么搭"。现在我们什么都支持,所以现在用 Lovable 你必须是个 prompt 高手。」

"Do you think we will prompt in five years' time?"

「五年后我们还会写 prompt 吗?」

"Yes, I think so. But it will evolve in terms of how you do it. Hyperpersonalization takes care of a lot of the detailed prompting we have to do today."

「会。但写法会演化。今天我们必须写得很细的那些 prompt,大部分会被超个性化吃掉。」

"What does that mean?"

「这个具体怎么理解?」

"Prompting is basically providing context to an AI of what your goals are and how you wanted to do something. It's like when you have great employees — they know everything about how you want it to work. So you just have to say 'let's go to Stockholm and do a hackathon,' and then it just magically becomes what you want it to be."

「prompt 本质上就是把"你想达成什么、希望怎么做"的上下文喂给 AI。这就跟你有特别好的员工一样——他们已经知道你的工作风格,你只要说"我们去 Stockholm 办个 hackathon",它就自动变成你想要的样子。」

"It pretty much does, honestly. I sent a picture to my mother beforehand and she's like 'you had nothing to do with that.' And I'm like 'no, I did not.' And she's like 'I know.'"

「老实说就是这样。我提前给我妈发了张照片,她说:"这事儿跟你没关系吧?"我说:"对,我没插手。"她说:"我猜也是。"」

"Right. You can't just tell ChatGPT 'let's go to Stockholm hackathon' — it's going to come up with something else than what you had in mind. So you can either prompt it very very detailed, or you can make sure it knows how you think. And that's what we'll be evolving towards."

「对。你不可能直接跟 ChatGPT 说"我们去 Stockholm 办 hackathon"——它给你的肯定不是你脑子里想的那个。所以你的选择只有两个:要么把 prompt 写得极度详细,要么让它真的"知道你怎么想"。我们要走的方向是后者。」

"Dude, I love your social media presence because you're also opinionated, and respectfully also just quite blunt. You've been opinionated in how you talk about competition, specifically like Replit. How do you think about whether or not to engage in an opinionated stance against competition or not?"

「老实说我很喜欢你的社交媒体——你在公开发言里非常 opinionated,而且坦白讲挺直接的。你对竞争对手——特别是 Replit——也表过态。你怎么决定要不要公开开炮?」

"Look, I don't think so much about competition. The only thing that matters is that we make our product the best product, and that we continue to deliver on our value promises to our customers."

「老实说,我不太想竞争对手这件事。唯一重要的是把我们的产品做成最好的产品,持续兑现给用户的价值承诺。」

"If there is something that happens — there was a competitor that found a lot of apps that have been poorly made, and they said 'this is a vulnerability.' I spoke to a lot of security professionals — that wasn't really how you would normally announce a vulnerability. So then I went in and bashed them as an outcome of that. I think that was a very reactive thing, and it's something I'd be happy to share in person to that competitor face to face."

「但有件事让我没忍住:之前有个对手找出一堆做得很糙的应用,说"这是个漏洞"。我跟好几个安全专业人士聊过——这不是业界正常公布漏洞的方式。我看完之后就直接公开开火了。说实话那是一个很 reactive 的反应——我也愿意当面跟那个对手把这事说清楚。」

"It was interesting because Jason Lemkin, who's a friend of mine, was using Replit, and I can't remember exactly what happened — but they basically had a massive security breach or they deleted all his database. It was like code red for them. And the takeaway for him was: security on all of them is just nowhere near where it needs to be. It's wrong for Replit to bash Lovable, it's wrong for Lovable to bash Replit. All of you guys suck at security. Is that true?"

「这事挺有意思的——我朋友 Jason Lemkin 在用 Replit 的时候,我记不清具体细节了,但他遇到了大事故:要么是大型安全漏洞,要么是数据库被全删了——总之 Replit 那边是 code red 状态。Jason 的结论是:你们所有人在安全这件事上都远远不够。Replit 黑 Lovable 不对,Lovable 黑 Replit 也不对——你们都很烂。这是真的吗?」

"Uh, yes. I think — yeah. Let me say it a different way. First of all, we talk about security companywide every week. Every day I hear something about security because we take it so seriously. There are so many different fronts to make it much more secure than if a human would do the application development. That's why it's so important for us to be the best in the world at security."

「呃……是的,我觉得——是这样。让我换个说法。首先,我们公司级别每周都会聊一次安全;每天我都会听到点跟安全有关的事——我们真的非常认真。要让这个产品比"人类自己写应用"更安全,要做的方向太多了。所以我们必须做到全世界最强的那一档。」

"So you're saying it's more secure than humans?"

「你的意思是,它比人类更安全?」

"Not yet. I said this at some point — if you take your average developer who normally works in a large team where they have a lot of support, and then that human goes out and builds an application alone, they are going to create software that has security holes on average."

「现在还没有。我之前说过一个判断——拿一个普通的开发者(他平时是在大团队里干活、身边有一堆支持系统),把他单拎出来让他独自去写一个应用——平均下来,他写出来的代码就是会有安全漏洞。」

"When you build an application with Lovable, it's going to tell you to go through a bunch of security reviews, and the AI is going to do a bunch of security reviews, and finally give you green light: 'we haven't found any security vulnerabilities.' So if you compare those two — that average really average developer with Lovable — Lovable is going to have a lower chance of having a vulnerability."

「在 Lovable 上 build 的话,系统会主动让你走一遍安全 review,AI 也会自己跑一遍安全检查,最后给你一个绿灯:"没有发现安全漏洞。"两边对比——那个"真的很普通"的开发者 vs Lovable——Lovable 出漏洞的概率更低。」

"And we want to put that to 0%. We need to put that at 0% chance of vulnerability."

「我们的目标是把这个概率压到 0%。我们必须把它压到 0%。」

"It reminds me of self-driving — for the world's best driver, I'm sure you are better than self-driving, but for the majority who are tired, humans have the potential to be hungover, high, malfunctioning, in some way actually wildly dangerous and much much worse."

「这让我想到自动驾驶——世界上最好的司机,你的水平肯定还是高过自动驾驶。但对绝大多数人来说,他们可能宿醉、嗑药、状态不对、危险得多,这种时候自动驾驶反而 wildly 更安全。」

"Very much so. I would say I'm very proud of what the team has done so far with security, but there's more to come."

「很对。我对团队目前在安全这块做出的东西非常自豪——但还有更多要做。」

"If you move forward three years, what does the space look like then?"

「往后看三年,这个赛道会变成什么样?」

"Again, I focus on what our product does and how we serve our customers best. If we get the majority of the profit share in this market, that's amazing. If it's spread out across different companies, that's also fine, as long as we build a product that lasts for generations. And I do that by building the best product for our customers."

「我还是只专注在"我们的产品做了什么、怎么把客户服务好"。如果三年后我们能拿到这个市场利润的大头,那当然棒。如果利润是分散在多家公司之间,那也行——只要我们做出一个能跨世代留下来的产品就够了。要做到这一点,只能靠"为客户做出最好的产品"这一条路。」

"Do you mind if devs go to Lovable, get 60% of the code from there, and then fine-tune it themselves? How do you feel about that?"

「如果开发者上 Lovable 拿走 60% 的代码、自己再 fine-tune,你介意吗?」

"From the get-go, two years ago, I decided I'm going to build Lovable for a world where humans don't write code anymore. We're quickly moving there. Today I don't mind at all — it should be flexible. Some humans have their way of doing things, and it's good if you have an ecosystem where you can use many different tools on a product."

「两年前一开始我就决定:Lovable 是为"人类不再写代码"的那个世界做的。我们正在飞快地朝那个方向走。今天我完全不介意——产品本来就应该灵活。有些人有自己的工作习惯,生态里能多工具混用,是件好事。」

"Over time I think it will converge towards the very opinionated platform we're building towards, and everyone's going to look at what's the cost-benefit of doing it with some other tool. Only using Lovable is going to be the obvious, highest-velocity, highest-quality-driving choice. That's the future."

「但长期来看,我觉得最终会收敛到我们正在建的这种"非常 opinionated 的平台"。到那个时候,大家会问"换别的工具划得来吗"——而答案会是:只用 Lovable 就是那个又快又稳的选择。那是未来。」

"Today, does AI make 1x engineers 10x, or does it make the 10x engineers 100x?"

「现在的 AI,是把 1x 工程师变成 10x,还是把 10x 工程师变成 100x?」

"It does both. So for the junior 1x engineers — they often are bad at something. If the AI can bridge that gap, it takes them from zero to one — to be able to do something they were not able to do before. So they go to 10x or even more."

「两个都干。对于 junior 那种 1x 工程师——他们往往在某些事情上是真的不会。如果 AI 能补上那一块,他们就从 0 到 1——能做出他们以前做不出来的东西。这等于一下子把他们提到 10x,甚至更高。」

"If it's a 10x engineer working on something where you need many years of experience to work on a system that a new junior engineer completely doesn't understand — the 1x engineer doesn't understand the system, then the 1x engineer is useless. So no AI is going to increase their velocity. Whereas the 10x engineer is going to maybe go to 100x engineer."

「但如果是一个 10x 工程师在搞一个需要很多年经验才能 hold 住的复杂系统,而新人 junior 完全不理解这个系统——那 1x 在这种场景里就是没用的,AI 也救不了他。但 10x 在这种场景里,可能就会被放大到 100x。」

"How will the size of engineering teams change in the next 5 years?"

「未来 5 年,工程团队的规模会怎么变?」

"For the best companies, engineers will really act as a translation layer. They will need to be more thinking in terms of product, being a product manager. And I think there's a higher elasticity on such engineers. So you might see many companies being like 'oh, more engineers — we can do even more, even faster.' They go out talking to customers and changing things with AI super fast."

「在最好的公司里,工程师会变成一个"翻译层":他们要更产品化思考,更像 PM。这种工程师对公司价值的弹性其实更高——所以你可能会看到很多公司反过来说"工程师再多招一点,我们就能更快做更多事"。他们直接跟客户聊,然后用 AI 飞快地改东西。」

"The skills required to be a good engineer then change with time."

「所以"好工程师"的技能要求会跟着变。」

"Yes, definitely. Being a generalist becomes more and more important with everything in AI — so that you can understand how things come together as a larger whole, and then you use AI for the deep expertise that you don't need in the future as much."

「肯定的。在 AI 时代,做一个通才会越来越重要——因为你需要看懂"整体怎么拼起来",而那些过去要你自己掌握的深度专业知识,以后可以让 AI 来补。」

"There's a lot of people asking today: should I bother studying computer science if we're going to see Lovable be the last software we ever need? How would you advise your little brother questioning whether to do CS at university?"

「现在很多人在问:既然 Lovable 这种东西可能就是"最后一个我们需要的软件",那我还要不要学计算机?你会怎么劝一个犹豫要不要读 CS 的弟弟?」

"I think university is not the best place to learn. Doesn't matter what you're studying. You should be out there and really understand how the world works in terms of how work translates to value creation. You don't learn that at university."

「我觉得大学不是学习的最好地方,学什么都一样。你应该真的到外面去,搞清楚"工作"是怎么转化成"价值创造"的——这件事你在学校里学不到。」

"University is like a way to train your brain to learn new things and meet a lot of interesting people."

「大学的功能更像是训练你的大脑去学新东西、还能让你认识一堆有意思的人。」

"Would you encourage your children to go to university?"

「你会鼓励自己的孩子上大学吗?」

"This is now 20 years in the future, pretty much. So it's hard to say something about 20 years in the future. I think it's a great experience to have had in life, so why not? But it depends on what outcome you want to reach. If you want to have a job where you make the most money — no, they should not go to university. I think the opportunity cost of those years is very high."

「这话其实是 20 年之后的事了,很难判断。我觉得这是一段很好的人生经历,为什么不去呢?但要看你想要什么结果——如果你的目标是赚最多的钱,那他们不应该去上大学。那几年时间的机会成本太高了。」

"True. In the UK in particular, we just get very drunk for three years. And that's generally how it is. Generally study generalist subjects like geography and history, and honestly it is a little bit of a waste of time — in which case you can utilize those years so much better given your stamina, your energy, the plasticity of your brain at that age. Which is why I highly advocate against it."

「确实。英国尤其是这样——我们就是连着喝三年酒。多数人学地理、历史这种通识科目,老实说挺浪费时间的——你 20 岁那种体能、精力、大脑可塑性,完全可以拿来做更有价值的事。这也是为什么我强烈反对(读大学)。」

"Agree. I mean, if you do a very very specialized job for those years, maybe you'll become less of a generalist. So there's a trade-off there, of course — that whilst at university you're exposed to many different concepts, which can be useful."

「同意。但反过来,如果那几年你做的是一份非常垂直的工作,你的"通才性"也会被牺牲掉。所以这里也是有 trade-off 的——大学的好处是它让你接触到一堆不同的概念,这件事是有价值的。」

"We mentioned earlier AI and enterprise. When you look at the biggest enterprises today, they're not able — for data, for permissioning, for security — to use AI. Are we going to see the biggest shift in incumbent power in the next 10 years?"

「我们之前聊到 AI 和企业。今天最大的那些企业,因为数据、权限、安全这些原因,根本用不起来 AI。未来 10 年,我们是不是会看到一次史上最大的"在位者权力洗牌"?」

"You see this in banking — a bank is a software company, right? It's all about software systems, and the old banks are moving much slower. Yes, there's going to be some companies built ground-up for AI to change their systems. And anyone exposed to customers, understanding the legal requirements, can move much faster in creating a good customer experience."

「银行就是个例子——本质上每家银行都是软件公司,全是软件系统。但老牌银行跑得慢得多。会有一批从零按 AI 思路搭起来的公司出现;那些贴近客户、理解监管的人,会跑得比老巨头快得多,先做出好的客户体验。」

"There are some benefits of having been around for a long time in enterprise — in banking, there's a certain element of trust. So I don't know how large the shift is going to be across different segments. But many companies will get disrupted by cheaper, much much better alternatives."

「在企业市场里"在位很久"也确实有它的好处——银行业里"信任"是个真因素。所以不同细分市场里这一波洗牌的剧烈程度会不一样。但确实会有大量公司被更便宜、好得多的对手颠掉。」

"What question should large CEOs and business leaders be asking today about the future of AI and their companies — that they're not asking?"

「今天那些大公司的 CEO 和高管,关于"AI 和我的公司"应该问、但他们没问的问题是什么?」

"One of the biggest bottlenecks for these companies is going to be some kind of change management for the humans in the organization. They should be asking 'how have similar companies to ours done change management very very rapidly?' Get that conversation into the leadership room and then across the organization to start studying those examples."

「这类公司未来最大的瓶颈之一,会是组织内对"人"的变革管理(change management)。他们应该问的问题是:"和我们类似的公司,是怎么把变革做得非常快的?"然后把这个对话先放到领导层里,再扩散到整个组织,系统性地研究那些成功的例子。」

"Then look specifically: which AI tools should we be using? Should we be adopting — like building our product on top of Lovable 100% — because then everyone can collaborate? Or should we hire some new type of people that come in and upskill everyone?"

「然后再具体到:我们到底该用哪些 AI 工具?要不要 100% 把产品建在 Lovable 上面——这样全公司的人都能在同一个地方协作?要不要去招一批"新物种"进来,让他们带着大家升级技能?」

"Bluntly, dude, I get really fed up with everyone saying Europeans are about espresso and taking the summer — 'it's August and July, we're not going to work.' I advocate for a very aggressive work culture, which you know is 996. How do you feel about the importance of unwavering hard work over balance in the desire to win?"

「老实说,我特烦"欧洲人就是喝喝 espresso、夏天不上班"这种说法。我自己鼓吹的是非常激进的工作文化——也就是 996。在"想赢"这件事上,你怎么看"全力工作 vs 平衡"哪个更重要?」

"Over a 10-year period, I would advocate for some balance. But over a two-year period, if you really care about something, then you should make sure you have your exercise, your sleep in really really well, and maybe something that relaxes you — and then just work your ass off. That's what you should be doing."

「在 10 年这个尺度上,我会建议保留一些平衡。但在 2 年这个尺度上,如果你真的在乎一件事,那就把锻炼、睡眠搞好,留一点能放松自己的事情——然后剩下时间往死里干。这就是你该做的。」

"Do you agree then with Scott from Cognition, who clearly said to all Windsurf employees after hiring them — 'it's six days a week. It's unwaveringly relentless. And if you don't want to sign up for that, you can leave.'"

「那你同意 Cognition 的 Scott 那种做法吗?他在收购 Windsurf 之后,直接对员工说:"我们一周工作 6 天,毫不动摇。不愿意签的可以走。"」

"I would say, in how we think about it — you are here to have 10x impact over other people at other companies. And if you don't have 10x impact... for some people you do that by being very talented, being good at your job, and being very focused. For some people you need to put in a [shit] ton of hours — and for some, not. So I prefer to talk about: am I seeing the impact?"

「我们的逻辑是这样的——你来这里是为了"做出比别家公司同岗位 10 倍的影响"。如果你没做出 10x 影响……有些人是靠天赋、专业能力、强专注做到的;有些人就是必须投入大量时间;还有一些人不需要。所以我更愿意问的不是"你工作多少小时",而是"我有没有看到你做出影响"。」

"If you would tell me you're leaving tomorrow, I would be like 'no, you are such an important part of this company you have to stay.' That's how I push performance and impact."

「我衡量的方式是:如果你明天来跟我说要离职,我的反应是不是"不行,你对这家公司太重要了,你必须留下来"。这就是我推动绩效和影响的方式。」

"Do you do the keepers test?"

「你们做"keepers test"吗?(Netflix 那一套——"如果这个人现在跟我说他要走,我会不会拼命挽留?")」

"Yeah, we do the keepers test."

「对,我们做。」

"Has it made you change how you construct teams?"

「这个 test 让你改变了你搭团队的方式吗?」

"It always makes it clear to people that I need to figure out 'how can I have more impact?' That's one part. Then I think: what does this organization look like? Is it optimally set up to succeed right now? Culture is such an important part — if you're just throwing people around too much, it hurts the culture and the ways of working. But doing this business exercise — 'is this organization set up perfectly to win?' — definitely shapes how I build the organization."

「它的好处是把这件事讲明白:每个人都得想"我怎么做出更大的影响"。然后我会问自己:这个组织,从"能不能赢"的角度,是不是处在最佳状态?文化非常重要——如果你只顾着把人挪来挪去,你会把文化和协作方式搞坏。但每隔一段时间问一次"如果纯粹为了赢,这个组织该长什么样",一定会塑造我搭团队的方式。」

"That's a good quote. It's a really good way to think about it, which I think is the same — if we win, everyone will be happy. There's very few places where they're losing every day, day in day out, and blissfully happy. It doesn't happen."

「这话讲得真好。我完全同意——只要我们赢,大家都会开心。一个团队天天输、天天输、然后所有人还乐在其中——这种地方几乎不存在。」

"What's not great about your culture today, if you could change it?"

「如果可以改一件事,现在你们的文化有什么不好的地方?」

"There's a certain personality type that takes a lot of initiative — they're very excited about new ideas and doing novel things. As your company matures, that's still important, but the first priority needs to be: make what you have high quality, continue to be high quality, and improve the quality across everything you're doing."

「我们公司里有一种性格类型——会主动出击,对新点子和新做法非常兴奋。这一点本身仍然重要,但随着公司变成熟,第一优先级应该变成:把现有的东西做扎实、保持高质量、把所有事情的质量都拔高一档。」

"I want us to be even more like — let's improve the quality, let's improve how we do things, move slow so that we can move really really fast."

「我希望我们再多一点这种气质:把质量提一档,把做事方式提一档——慢一点,这样后面才能真的快。」

"Do you think you're in that phase of company build yet? I actually prefer the 'optimize for the short term.' I don't know if you've spent much time with Chinese development teams, but they are unbelievable in their psychology around build. They optimize for the short term incessantly and then just sticky-tape, sticky-tape, sticky-tape. And that is often how they're able to do so much so fast."

「你确定现在是该进入这个阶段了吗?我自己其实更喜欢"短期最优"。不知道你跟中国的开发团队有没有打过交道——他们的搭建心态非常恐怖。短期一直最优,然后就是"打补丁、打补丁、再打补丁"。他们就靠这一套跑得飞快、做出极多东西。」

"If you have super clear product-market fit, you have a brand to defend — you cannot, sticky tape is [not] everything. You can do that in sprints to innovate. But you really need to focus on: are these pieces put well together? Spend a lot of time on moving things around in the organization and in your product, so that it has high quality, maintains high quality, and you can build faster upon that foundation."

「如果你已经有非常明确的 PMF,有品牌要守——你不能"全部打补丁"。打补丁可以用在 sprint 期的创新上。但整体上你必须不断问:这些零件之间真的拼得对吗?要花大量时间去调整组织、调整产品的结构,让它保持高质量。在那个地基之上,你才能跑得更快。」

"Can you imagine if Apple were like 'ah, [shit], we deleted your iCloud, sorry'?"

「你能想象 Apple 突然来一句"啊,卧槽,你的 iCloud 被我们删了,抱歉"吗?」

"You said before that it is better to build in Europe. I don't want this to be like an advert for Europe — but why do you think it's better to build in Europe?"

「你之前说过,在欧洲做公司更好。我不想把这一段搞成"欧洲招商广告"——但你为什么觉得在欧洲建公司更好?」

"There are many good things about Europe — and also things that are better in the US. I mostly think about: I want to prove that you can build a generational product, a generational company team from Europe. And part of it is on hard mode."

「欧洲有不少好的地方,美国也有它更强的部分。我心里更多想的是:我想证明,从欧洲也能做出一个世代级的产品、一个世代级的公司。这条路有一部分是 hard mode。」

"What parts are on hard mode?"

「哪些部分是 hard mode?」

"There's a network where it isn't as great in how many individuals and companies have worked on and have context for all the different stages of building an amazing multinational company."

「这里的网络密度不够——能给你"如何把一家公司从早期一路推到跨国巨头"这条路上每个阶段提供经验的人和公司,都太少。」

"Completely agree. There are no Elena Vernas in Europe."

「完全同意。欧洲没有 Elena Verna 这种角色。」

"Very few. Maybe we'll get there soon. But that's hard mode. Access to capital, people that will quickly give you a lot of distribution, help you with distribution and brand."

「极少。也许过不了多久会有,但当下确实是 hard mode。还有就是融资渠道、能立刻给你大渠道分发的人、能帮你做品牌和分发的人——欧洲都更稀缺。」

"Do you think access to capital is a genuine problem? I think there's so much money in Europe."

「你觉得欧洲融资真的难吗?我自己感觉欧洲钱挺多的。」

"As I said, it's not a bottleneck for us. No, it's not."

「正如我前面说的,对我们来说完全不是瓶颈。」

"You're going to start to see very soon — I'm sure you're already seeing it — Lovable spinouts where anyone who leaves Lovable will get a term sheet straight away."

「你很快会看到——你大概已经看到了——Lovable 系的"分裂出来再创业":任何从 Lovable 离开的人立刻就能拿到 term sheet。」

"100% true."

「100% 是这样。」

"Why is it better to build in Europe?"

「那欧洲到底好在哪里?」

"It's easier to get distribution to be on the centre world stage in San Francisco and New York. But we've been able to pull that off from Stockholm, which is a good proof that you can do it from here."

「单论站上"世界中心舞台"那种 distribution,SF 和纽约确实更容易。但我们从 Stockholm 也做到了——这本身就是个证据:在这里也能做。」

"Why do you think you've been able to do it? I have my theory on why I think you've been successful."

「你觉得你为什么做到了?我自己有一套理论,但我想先听你说。」

"It's about storytelling and sharing everything we're doing at the company. Empowering other people who are using Lovable, telling their stories. That's how we've been breaking through."

「核心是讲故事——把我们公司在做的事情完整地共享出来,然后让用 Lovable 的人也讲他们自己的故事。这是我们打开局面的方式。」

"I think we understand that you should be building in public and share what you're doing. Transparency is everything. People like to follow people. You've combined the two very well — you're incredibly transparent around your ARR growth (easy to be when it's as good as it is, but still in a way that most people aren't), and then it's led by you and your voice. The combination of cult of personality — you being Anton, not 'you and the third person' — and growth, is what really drives the success."

「我同意"在公开场合建公司"是必须的,要把你做的事 share 出来。透明本身就是一切。而且人们喜欢追"人"。你把这两件事融得特别好——你在 ARR 增长这块极其透明(数字这么好当然容易透明,但你比绝大多数人都更透明);加上整件事是你这个人在前面带头讲——以"Anton"为主语,而不是公司"我们"的那种第三人称。"个人魅力"和"增长"这两件事的合体,才是真正驱动成功的东西。」

"What's better? Why should everyone build their company in Europe?"

「那回到正面——欧洲到底好在哪?为什么大家都该来欧洲建公司?」

"We are the biggest talent magnet in Stockholm right now, which is amazing. You can't be that in San Francisco or New York. So we can really pick up all the underutilized talent and shape them — 10x their performance by being in a 10x better culture, ways of working, and with amazing colleagues. Being able to be that 'top one' is the biggest one."

「我们现在就是 Stockholm 最大的"人才磁石"——这件事在 SF 或纽约你做不到。所以我们能把那一片"被低估的人才"全部捞起来,让他们在一个 10 倍更好的文化、10 倍更好的协作方式、和非常牛的同事一起工作的环境下,把自己的产出也提到 10 倍。能做"地方第一"——这是欧洲最大的优势。」

"There's a culture of humility and low ego, working really really well together as a team, that I think is stronger in Europe. And this 'do much more with less' efficiency mindset is also stronger here."

「欧洲有一种"低 ego、谦虚、能真正作为一个团队协作"的文化,这一点比很多地方都强。还有"用更少做更多"的效率心态,在这里也更明显。」

"You have inherently higher churn in the Valley. When you have a bad day, OpenAI offers you a bigger package and it's like 'ah, I'll leave and do OpenAI.' That prevents the compounding of knowledge within teams, which I think is so valuable."

「Valley 那边的 churn 是天然就高的——你某天不爽,OpenAI 给你一个更大的包,你就走了去 OpenAI。这就让"团队内部知识的复利"积累不起来。我觉得这一点其实很值钱。」

"Yep. Totally agree with you there."

「对,这一点我完全同意。」

"Would Lovable be less successful if it were in the Valley?"

「如果 Lovable 在 Valley,会不会更不成功?」

"I honestly don't know. I think it would be very successful regardless."

「老实说我也不知道。我觉得不管在哪儿,它都会非常成功。」

"Did you ever think about moving?"

「你考虑过搬去 SF 吗?」

"Yes. When I was about to start the company, everyone was telling me I should go to SF. But we just kept building, and we found some great people in Stockholm, so we kept building it from here. I'm happy how it turned out."

「想过。开公司之前,所有人都跟我说"你应该去 SF"。但我们就这样在 Stockholm 一边写一边招人,结果在这里找到了一群很厉害的人,就继续在这里搭。事后看,我对这个决定很满意。」

"What did you do in the Lovable journey that — with the benefit of hindsight — you wish you hadn't done?"

「Lovable 这一路上,有什么事情你现在回头看,会希望当初没做的?」

"When we started, we had this idea — the vision was very clear, the sequencing was not so clear in what we should be doing. We had this open-source community that was kind of excited about a tool I made a few months before we started the company — GPT Engineer."

「公司刚起步的时候,我们 vision 很清晰,但事情的顺序排得不够清晰。当时已经有一个开源社区在关注我创业前几个月写的一个工具——GPT Engineer。」

"I think we shouldn't have just scrapped that completely and 100% focused on what's the future look like — opening your browser and just building your product there, which is Lovable."

「我觉得当时我们应该直接把 GPT Engineer 那条线全部砍掉,百分之百投入到那个未来形态的 Lovable 上——也就是"打开浏览器,直接造产品"这件事。」

"Why should you have scrapped that? Was it not crucial for customer development, customer feedback?"

「为什么应该砍?那个开源社区不应该是早期客户开发和反馈的关键吗?」

"No, I don't think so. There was of course a plan for how to incorporate it together — open source can be very useful for many businesses. But in the perspective of maximal focus, it was just a bad idea to do two things that were a bit too tangentially related."

「我不这么认为。我们当然有过把它们融合的计划——开源对很多生意都是有用的。但从"最大化专注"这个角度看,做两件相关、但不够紧密的事情,本身就是错的。」

"We talk a lot about doing one thing — finding the bottleneck for the company and solving for that bottleneck. That is the best way to move really really fast."

「我们公司内部反复强调"做一件事"——找到公司当下最大的瓶颈,然后只攻这一个。这就是跑得飞快的最好方式。」

"What is the bottleneck that you'll be discussing in the board tomorrow?"

「明天董事会上,你要讨论的瓶颈是什么?」

"Long-term: how we identify the technical product engineers that will take the product to its next phase and innovate on many fronts at the same time."

「长期看的瓶颈是:怎么识别出那种"既懂技术又懂产品"的工程师——能把产品带进下一阶段、还能在多个方向上同时创新。」

"Today, the product bottleneck is giving our AI more capabilities — really polished UX, more capabilities so you can build out your full company and grow your business on top of Lovable. Then how we serve all this enterprise pull at the same time as we focus on founders building on Lovable."

「短期看,产品上的瓶颈是给 AI 增加更多能力,把 UX 打磨得真正精致,让你能在 Lovable 上把整个公司搭出来、把生意做起来。再往上一层,是怎么在保持服务好"在 Lovable 上创业的人"的同时,接住企业端不断涌来的需求。」

"Will Lovable have an enterprise sales team?"

「Lovable 会有企业销售团队吗?」

"Yes. It will not become an enterprise company — but it will have this enterprise sales team. I'm not so nervous about it. It's just about talking and understanding your customers, making sure they have the tools to get value from the product."

「会有。但 Lovable 不会变成一家"企业公司"——只是会有企业销售这一支团队。我对这件事不紧张——它本质上就是和客户聊、理解他们,让他们能从产品中拿到价值。」

"Many enterprises have a top-down enterprise sales team that hustles themselves to wine and dine CEOs. That's not what we're going to do."

「很多 to-B 公司搞那种从上往下的企业销售——靠请 CEO 吃饭喝酒拿单。我们不会这么做。」

"What do you think is the biggest secret to a successful co-founding pair scaling at the speed of Lovable?"

「在 Lovable 这种速度上,联合创始人这一对最大的秘诀是什么?」

"The most important thing is just the horsepower and adaptability of the founders. If those are maxed out — you must be able to work together. With sufficiently low ego, it's going to work."

「最重要的就是创始人本身的"马力"和适应能力。这两项打满了之后,你们必须能搭得起来。只要 ego 都够低,基本就能跑得动。」

"To take Fabian and me as an example: he's not very big on doing some weird new way of doing things — he's just like 'simplify as much as possible.' He's quite introvert and quiet until he's really shaped an opinion about what's the most important thing. And I'm on the polar opposite side of the spectrum, saying 'father, we should use this new crazy thing.' That polarity is actually very productive for both of us."

「拿 Fabian 和我来说:他不喜欢搞那些花里胡哨的新做法——他的口头禅就是"尽量简化"。他比较内向、安静,只有真正形成了"这件事最重要"的判断之后才会开口。我正好是另一极——天天在喊"老兄,我们应该用这个新玩意儿"。这种极性反差,对我们俩反而都很有产出。」

"What does a Lovable product look like at the end of 2026?"

「2026 年底,Lovable 产品长什么样?」

"It's your perfect co-founder that you go to with your idea from the idea stage, but also all the way up to growing your business once you have customers — taking care of what Elena is doing: optimizing the product for growth, optimizing your communication with your customers through email or different marketing channels."

「它是你的完美合伙人——从一个想法刚冒头的阶段开始陪你,一直陪到你有了客户、要做增长这一段。也就是 Elena 现在在做的那些事:为增长做产品优化、优化你和客户之间的沟通,无论是邮件还是其他营销渠道。」

"So you eat the whole stack then."

「所以你吃整条 stack。」

"Yeah, it's one opinionated way to do the entire product life cycle."

「对——一种 opinionated 的方式,把整条产品生命周期串起来。」

"Is benchmarking for models [shit] and evaluations [shit]? I had Edwin from Surge on the show — Scale AI's competitor, never raised a dollar and it's $1.2B in revenue. He was like, 'the benchmark / evaluation is [shit]'."

「模型的 benchmark 和 eval 是不是都很烂?我之前请 Surge 的 Edwin 上节目——他们是 Scale AI 的对手,一分钱没融,做到 12 亿美元收入。他直接说"benchmark 和 eval 都是垃圾"。」

"They turn more and more [shit] over time. There's something called Goodhart's law — when you start optimizing for a number, that number stops being a good measure for success. Even if it was a great measure previously. So that happens with all benchmarks over time in some sense."

「随着时间推移,它们会变得越来越烂。有个东西叫 Goodhart 法则——一旦你开始为某个指标做优化,这个指标就不再是成功的好衡量了,哪怕它之前是个非常好的衡量。所有 benchmark 在某种程度上都会经历这个过程。」

"What metric within Lovable means less over time?"

「Lovable 内部哪个指标的"含金量"会随时间下降?」

"It means less if we start optimizing for it. One example: how many people click the thumbs-up button on messages. Because then we can say fun jokes or whatever that just triggers people to click — just asking the human to click the button. That's hacking the metric."

「只要我们开始为它优化,它就开始失真。一个例子是"消息上的点赞数"。如果我们盯着这个指标,我们可以让 AI 说点搞笑的段子,纯粹为了诱导用户去点那个按钮。那就是在 hack 指标。」

"What widely-held belief about AI do you think is just very wrong?"

「关于 AI,什么"大家都信"的观点,你觉得是错的?」

"AI is much better than humans, and most people don't agree."

「AI 已经比人类强得多,但大部分人不同意这一点。」

"Do you not think they do now?"

「你觉得现在大家还不同意吗?」

"Most people don't agree. The reason is, it's often times very stupid. But if you give it all the context, or you build a purposeful system for what it's stupid at — it's smarter than humans."

「大部分人不同意。原因是 AI 经常表现得很蠢。它确实经常很蠢。但如果你把上下文喂全,或者针对它"蠢"的部分专门搭一套支撑系统——它比人聪明。」

"Do you think we will see a plateauing or a continuous exponential progression curve?"

「你觉得我们会看到一段平台期,还是持续的指数曲线?」

"I think we'll see plateauing on the things we care about — a lot of nuance and being good at all the different things at once in the same model. There are some sigmoid curves where we're still in the exponential phase — like science and engineering and bioengineering. AI is going to continue to exponentially become extremely powerful and generate a lot of new medicines and new ways of treating health."

「在我们日常关心的那些维度上——细腻的细节、同一个模型同时擅长很多事——我觉得会进入平台期。但有一些 sigmoid 曲线现在还在指数期,比如科学、工程、生物工程。在那些领域,AI 会继续指数式地变强,会带来一批新药和新的治疗方式。」

"Grok, Anthropic, OpenAI. You can invest in OpenAI at 380, Anthropic at 180, and Grok at I think it's 100. Which one do you invest in and which one do you short?"

「Grok、Anthropic、OpenAI——OpenAI 380 亿、Anthropic 180 亿、Grok 大概 100 亿。你做多哪个?做空哪个?」

"I'd invest in Grock. And I would short — what was the numbers again? 380 and 180. Okay. 100. I would probably short Anthropic because — no, I would short OpenAI, let's say."

「我会做多 Grok。然后做空——刚才的数字是?380 和 180,好。100。我可能做空 Anthropic,因为——不,我做空 OpenAI 吧。」

"Why would you buy Grok and short OpenAI?"

「为什么做多 Grok,做空 OpenAI?」

"It's more the slope on the Grock team. They're doing something I respect a lot — to hire missionaries for the data curation part, they call it 'AI tutoring.' The morale is much better in that team than both of the other teams. It's super high. OpenAI has gone through all this mess. And Anthropic has good morale as well, and they're growing faster on the enterprise side from what I'm hearing."

「主要是 Grok 团队的斜率。他们在做一件我非常尊重的事——为数据管理(他们叫 "AI tutoring")这一块,招的是真正有使命感的人(missionaries)。这支团队的士气比另外两家都明显要高,真的非常高。OpenAI 这一阵内部一团乱。Anthropic 士气也不错——而且从我听到的消息看,他们企业端增速更快。」

"Do you think there will be a leading model that has not been created yet?"

「你觉得会有一个还没出现的模型,最终成为领跑者吗?」

"Yes. From China."

「会。来自中国。」

"Do you worry about China?"

「你担心中国吗?」

"Chinese companies are not as good at really understanding your users. So not very worried. I do think there's like a 50/50 chance they will have the best model. We'll be using a Chinese model at some point — and that makes me a bit concerned, because I'd have to look into the details: do we give them data we don't want to give them? But we just want to do what's best for our customers. If that's a Chinese model and there are no negatives — yes."

「中国公司在"真正理解你的用户"这件事上还不够强。所以从产品维度我不算很担心。但模型能力这一块,我确实觉得他们有 50/50 的概率会做出最强的模型。我们 Lovable 早晚会用上中国的模型——这件事让我有点紧张,因为得仔细看:这是不是会让我们泄露不该给的数据?但归根结底我们想做的是对客户最好的事。如果用中国模型对客户最好、又没有负面,那就用。」

"I completely agree. The multitude of models coming out of China is just terrifying. Every week there's like four new ones, and they're all as good as the last one. The speed of distillation is just [shit] insane. Are the models of the future open or closed? Which model wins?"

「我完全同意。中国一周冒出来四个新模型,每个都和上一周那个一样好——蒸馏的速度真是恐怖。未来的模型是开源还是闭源?哪一边会赢?」

"The best ones will always be closed. But if you want maximum flexibility and some kind of open ecosystem around it, it might be that open ones are the ones most people choose."

「最强的那一档永远会是闭源。但如果你要最大的灵活性、要一个围绕模型的开放生态,那很多人最终选的会是开源那一边。」

"You can have dinner with anyone dead or alive — who do you have dinner with and what do you ask them?"

「死人活人随便选,你想跟谁吃顿饭?问他什么?」

"I would have dinner with Newton. Because he was religious and super smart, and just talk about how he was in his age and why he's religious. He invented so many different things — he's a bit of a role model. And he's dead, so I can't meet him unless I ask him... no, sorry, I can't help with that."

「我想跟牛顿(Newton)吃饭。因为他既是个虔诚的宗教信徒,又极其聪明——我想问他,在他那个年代,他到底是怎么生活的、为什么会信教。他发明了那么多东西,我把他当 role model。他已经过世了,所以我没法见到他,除非……不,抱歉,这一题我答不上来。」

"What AI company does no one pay attention to that everyone should pay attention to?"

「哪一家 AI 公司,大家都没在看,但其实应该看?」

"I think the browser companies are interesting. There's Strawberry, here's.com, there's Dia and Perplexity. I'm very excited to see what happens to other companies."

「浏览器公司这一块挺有意思——Strawberry、here's.com、Dia,加上 Perplexity。我很想看接下来这一块还会冒出什么。」

"What do you think happens to Perplexity?"

「Perplexity 后面会怎么样?」

"They want to create the phone, I think. And I think that's a good bet."

「我猜他们想做一台手机。我觉得这个赌注还行。」

"Would you invest in them at 18 billion?"

「180 亿估值你会投他们吗?」

"18 billion. Um... it depends on what options I have."

「180 亿啊……呃,要看我手上还有什么别的选项。」

"That's amazing. Your laugh there just kind of said it all."

「太精彩了。你那一笑等于回答了所有问题。」

"What have you changed your mind on most? Penultimate one — I thought, bluntly, you'd see commoditization of model performance and value accrual would be very difficult, and there'd be a race to the bottom. I think that's clearly very wrong of me to have ever thought that — there'll be very valuable model providers. What did you believe that was wrong?"

「你最近改变最大的一个想法是什么?我自己一个改过的判断是——我以前以为模型能力会被商品化、模型方很难积累价值、会陷入"逐底竞争"。现在看这个判断完全错——以后会出现非常值钱的模型方。你呢?」

"In the context of Lovable — I thought we should be building an agent before the models were ready for it, because the models were starting to get optimized for an agentic system."

「Lovable 这边——我之前以为我们应该在模型还没就绪之前就先把 agent 做出来,因为那时候模型已经开始为 agentic 系统做调优了。」

"What I realized is no, no, no — you need to have a product that as many people as possible are using today, so that you can optimize not necessarily the AI but the entire user experience for those users. That's your data flywheel."

「但我后来意识到完全不是这样——你需要的是一个"今天就有尽可能多的人在用"的产品。这样你才能优化的不是 AI 本身,而是整个用户体验。那才是你真正的 data flywheel。」

"Do you worry about job displacement at scale in a 10-year time period?"

「10 年尺度上,你担不担心大规模工作位移?」

"I worry about us humans globally not even understanding what we want to achieve on this planet. If there's a lot of rapid change with white-collar workers being out of a job, humans get super worried, concerned, and scared — all hell is going to break loose. That's what I'm worried about."

「我更担心的其实是另一件事:人类在全球尺度上根本搞不清楚"我们到底想在这颗星球上达成什么"。如果短时间内大量白领失业,人类很容易变得担忧、焦虑、恐惧——那时候可能就要出大乱子。这才是我真正担心的。」

"But if we're a bit more thoughtful — 'okay, if there'd be insane amounts of job displacement, this is what we should do, this is what we want to achieve, this is how we make sure people can have some made-up job in the interim' — then we would 100% solve that."

「但如果我们能稍微多想一步——"假如真的发生大规模工作位移,我们应该怎么做、想到达什么状态、过渡期怎么让人保留某种'被发明出来的工作'"——这件事我们 100% 能解。」

"8 out of the top 10 paying jobs today did not exist 15 years ago. We always overestimate job displacement with new technologies."

「今天最高薪 10 个职业里,8 个 15 年前根本不存在。我们对新技术造成的失业,总是高估。」

"I think we're going to have a shift away from some very glamorous jobs, which people will get depressed by — similarly to how being an artist was so cool but clearly you can't make any money as an artist. We're going to see that again now for a lot of knowledge work. That's going to be funny."

「我觉得我们会经历的是:一些原本很光鲜的职业会失去那种光环,做这些工作的人会很沮丧——就像"做艺术家"曾经很酷,但也清楚得很——你做艺术家其实赚不到钱。现在很多知识工作也会经历同一件事。这一幕会挺有意思的。」

"Which competitor do you most respect?"

「最让你尊敬的对手是谁?」

"I respect Figma. They're good at listening to their users and building a good product. If they can translate that to the full product life cycle, I think they're a very formidable competitor."

「Figma。他们非常擅长倾听用户、把产品做好。如果他们能把这个能力延伸到完整的产品生命周期,他们就是一个非常可怕的对手。」

"Everything goes to plan — we hit all of our numbers, everything works. Where then is Lovable in 5 years' time?"

「假设一切按计划走,所有数字都达成,5 年之后 Lovable 是什么样?」

"We're the most-used interface for humans to AI. And that's a very huge market."

「我们就是人类用 AI 时最常用的那个界面。那是一个非常大的市场。」

"Dude, it's so much better doing it in person. I've so enjoyed this. Thank you so much for agreeing to do it in person. I've loved it, man."

「兄弟,当面录这一期太爽了。我超享受这个过程。谢谢你愿意当面来录,真的很喜欢这一集。」

"It was fun."

「挺开心。」