在 Anthropic 做设计：Joel Lewenstein 谈 AI 浪潮下的设计变革

今晚我非常荣幸能与 Anthropic 的设计负责人 Joel Lewenstein 同台。请大家给他热烈的掌声好吗？Joel 在科技行业有着一段精彩的职业经历。我和他算是老相识了——我记得我们最初建立联系时，你还在 Quora。

I am very honored to share the stage tonight with Joel Lewenstein, head of design of Anthropic. Can we please give him a very warm round of applause? Joel has had a story career in tech. He and I go back — I think we first connected when you were at Quora back then.

是的。

Yeah.

在这段时间里，我们两人都经历了软件设计领域的诸多变化。但最大的变化莫过于过去 18 到 24 个月里人工智能的出现，这在很大程度上要归功于像 Anthropic 这样的公司。所以今晚，我们要聊一聊 Joel 作为 Anthropic 设计负责人是如何亲身应对这些变化的，同时也会谈谈他对设计领域未来发展的预见。那我们就直接开始吧。Joel，感谢你的到来。我想先请你描述一下你刚加入 Anthropic 时的工作是什么样的，现在又是什么样的，因为我想你当初签约时所设想的和你今天所负责的之间，大概有着巨大的差异。

And we've both navigated numerous changes to the field of software design over that time period. But no greater changes happened than in the past 18 to 24 months with the advent of artificial intelligence, in large part due to firms such as Anthropic. And so tonight, we're going to get to chat a little bit about how Joel has been personally navigating it at the helm of design at Anthropic, but also just talking about what he foresees for the field of design going forward. So, we'll dive right in. Joel, thank you for joining us. I wanted to kick things off with describing what your job started like at Anthropic and what it looks like now, because I think there's probably a huge delta between what you signed up for and what you're overseeing today.

好的。大家好，感谢邀请，非常感激。Anthropic 最棒的地方在于它本质上更像一个研究机构，做产品的决定是很晚才做出的。所以感觉就像——我们这个机构像是一个受人尊敬的哈佛研究实验室，一直在产出该领域的开创性成果，而后院搭着一个小小的帐篷，里面是个 YC 创业公司，喊着‘嘿，我们也在这儿’。所以早期很多工作基本上就是从零开始搭建那些我们习以为常的东西。我翻出了一份文档——我在那儿大概三年了，所以这份文档大概是两年半前的——标题叫‘我们该不该做一个移动应用？’当时真的就是在公开争论一家前沿模型公司该不该做，而且确实有很好的理由说明你可能不该做——为什么只提供平台访问给其他人反而更好。而现在我们已经连续几个月位居 App Store 前十了。所以就是这种基础引擎运转起来的浪潮，我觉得我们其实在过去几年里已经搭建了一个相当不错的引擎。然后 Opus 4.5 彻底把所有人震住了，于是今年我们又从头重建了整个东西。

Yes. Hey everyone, thank you for having me. Very much appreciate it. The best part of Anthropic is that it is like a research house at its core, and the decision to make a product at all was like a much later one. And so it very much feels like we — the institution is like a venerated Harvard research lab that's been producing the seminal results in a field, and then like a tiny tent in the backyard with like a YC startup that was like, hey, we're here too. So a lot of the early days were just like essentially building the muscles from scratch, the things that we all take for granted. I dug up a doc from my — I've been there for about three years, so this is maybe two and a half years old — called 'Should we have a mobile app?' And it was literally like an open debate as to whether a frontier model company should, and there's like actually genuinely good reasons why you might not want to — why it's better to just kind of provide platform access for others. And now we've been in the top 10 of the App Store for the last couple months. So there's just this wave of sort of like the basic engine of things going, and I think we had actually gotten like a pretty decent engine running over a couple years. And then Opus 4.5 just like broke everyone altogether, and then we rebuilt the thing from scratch again this year.

那你最初是怎么接到邀请的？我是说，Anthropic 最初是怎么进入你的视野的？因为你所描述的听起来并不像是对一个软件设计师——一个曾把软件交付给数百万用户的人——最有吸引力的价值主张或最有吸引力的说辞。那么，他们的吸引力究竟在哪里，让你加入了他们？

And how did you first get the call? I mean, how did Anthropic even first appear in your sphere? Because what you described does not sound like the most attractive value proposition or like the most attractive pitch to a software designer, somebody who's shipped software to millions of people. So, what was it about their appeal that brought you to the fold?

是的，我整个职业生涯都在应对这样一种割裂：我似乎擅长的事情——让像素对人们有意义——和我认为世界上需要解决的重要问题清单之间——比如气候变化、经济不平等、药品可及性。我尝试过各种切入方式，核心之一是关于信息获取。我做过一些政治科技方面的事情。而 AI 真的是我遇到的唯一能真正弥合这一鸿沟的东西，以及这种‘任何人都能为这些更宏大的事业做出贡献’的感觉。而让我确信 Anthropic 就是做这件事的地方的那次访谈，是 Dario 上了 Dwarkesh 的播客——也是几年前，在我开始面试之前。Dwarkesh 提了一个问题，大意是：假设我们有了 AGI，治愈了世界上所有的疾病，你认为这项技术的民主治理应该如何运作？作为一个还没适应 Anthropic 说话方式的普通人，我当时想，这假设也太疯狂了——Dario 肯定会降低预期，说‘嗯，你知道，咱们别想太远’之类的。但他直接接受了这个前提，说‘是的，好问题，以下是我对 AI 国际治理的看法’。那种感觉既是——对这项技术无论如何都会实现的绝对信念，又是对监管、政府——社会作为一个整体该如何应对某件事——的急切拥抱，那是一种我之前从未遇到过的宏大愿景。

Yeah, I have navigated my entire career with this disconnect between the thing that I appear to be able to do, which is like make pixels make sense to people, and like the list of problems that I think are important to solve in the world — like climate change, economic inequality, you know, access to medicine. I've tried a variety of stabs at that, core being one on sort of information access. I worked in like some political tech things. And AI is like really the only thing I've come across that genuinely bridges that gap and this feeling that like anyone can contribute to these broader things. And the interview that convinced me that Anthropic was the place to do it was Dario was on the Dwarkesh podcast, again a couple years ago before I started interviewing. And Dwarkesh poses a question that's like, okay, presuming that we have AGI, we've cured all the world's diseases, how do you think like democratic governance of this technology should work? And being like a normal person who was not yet acclimated to how Anthropic spoke, I was like, that's like a crazy assumption — like surely Dario was going to like lower expectations and be like, well, you know, let's not jump ahead, blah blah blah. And he just accepted the premise and he was like, yeah, great question, here's how I think about like international governance of AI. And that was a sense of like both the just like utter conviction that this technology is going to happen one way or another, and this really eager embrace in like regulatory, governmental — like how is a society as a whole to grapple with something — it was a sort of a grandiosity of vision in a way that I had not encountered before.

我想你在 Anthropic 大概也经历了一种‘减一’的旅程，因为在很多方面，如果你是那个楼外帐篷的一部分，在巨大的技术变革中把产品开发引入这个新环境，你大概也在重新审视和质疑过去所有的金科玉律。在关于如何实践设计的思考上，你不得不丢弃多少东西？因为你也曾在 Airtable 工作过，在 Quora 工作过，你见过软件是如何被制造的，而且在很大程度上你被招进来就是为了把那一套带入这个组织。这项新技术在多大程度上迫使你走上了一条属于自己的‘减一’之旅，去思考如何做出对人们有意义的产品？

And I imagine there was probably some form of a minus-one journey for you at Anthropic, because in many respects, if you are part of the tent outside the building introducing product development to this new environment amidst a huge technological shift, you're probably also revisiting and questioning all of the truisms of the past. How much did you have to discard at the door in terms of how you thought about practicing design? Because again, you've worked at Airtable, you've worked at Quora, you've seen how software is made, and in large part you were hired to help bring that into that organization. How much of this new technology kind of forced you to have to take a minus-one journey of your own in terms of how to make products that are relevant to people?

是的，我觉得基础模型公司不同寻常的地方在于，在产品或设计师所从事的最后一公里软件之外，存在着一种价值，对吧？就像我之前工作过的每一个地方——大概我们任何人工作过的任何地方——唯一能产生的价值——如果公司所有人都去度假了，你的用户就不会获得任何新价值，对吧？除了你写 PRD 和规格文档之外，软件不会自己做出新东西。所以你必须非常有远见、有战略、有意图，对吧？就像你在指挥部队之类的，但你覆盖的领土只限于你派遣部队去的地方。而对我来说，设计实践的大部分内容——从像素到为什么，到战略，到更大的图景——在 Anthropic 真的就像是在寻找机会。如果我们的研究员去度假了，我们就毫无进展；但如果我们的产品团队去度假了，模型只会越来越好、越来越好，积累的价值越来越多。所以某种程度上，不采取行动其实也没关系。所以我认为几乎所有来自更传统产品背景的人进来时都会说，‘好，我们要写一个 18 个月的战略，要聚焦 TAM。’

Yeah, I think the unusual thing about a foundation model company is like there is a thing of value that exists outside of the last-mile software that is the product or that designers work on, right? So like every other place I've ever worked — presumably that any of us have ever worked — like the only value that gets — if everyone at the company went on vacation, no new value would accrue to your users, right? The software doesn't do new things outside of you writing a PRD and a spec. And so you have to be very forward-looking and strategic and intentional, right? You're like guiding troops or something, but the only territory that you cover is where you send your troops. And so much of what the practice of design for me had been — moving from the pixels to the why to the strategy to the sort of bigger picture — really like finding opportunities at Anthropic. If our researchers go on vacation, then we make no progress, but if our product team went on vacation, the model just gets better and better and better and more and more value accrues. So there's like a kind of like no action is actually fine in some ways. And so I think almost everyone who's come from a more traditional product background comes in and they're like, 'All right, we're going to write like an 18-month strategy. It's going to be like TAM focused.'

我们会思考如何驾驭市场，然后倒推。Dario 反复强调的，也是我们现在做产品开发的方式，就是让模型引领我们，相信缩放定律。我们相信模型会越来越好。我们会对新服务、新产品、新能力采取相当轻触的方式，一切都是为了释放越来越多的能力。我用的类比是，你有一个巨大的风电场在远处，人们现在就像把烤面包机插到 12 伏的插座上，他们说这个烤面包机真棒，而你说等等，这里还有更多，但并不是说我们真的需要建很多，而是我们只需要打开新的门户，让已有的能力流入人们想要的东西。

We're going to think about how to navigate the market and then we're going to work backwards. And what Dario has emphasized over and over and is now how we practice product development is like we're just going to let the model lead us and we're going to believe in the scaling laws. We're going to believe that the model gets better and better. We're going to take honestly fairly light touch approaches to new services, new products, new capabilities, all in the service of just like unlocking more and more power. The analogy I use is you have a sort of like massive wind farm sort of like sitting off site and people are currently like plugging in a toaster to like a 12, you know, 12vt outlet and they're like we're like this toaster is great and you're like hold on like there's a lot more here but it's not that we actually have to like build a lot. It's actually that we have to just like open up new portals into the power that's already there and let it like flow into the things that people want.

你需要把人灌输到这种新的构建方式中吗？我的意思是，这将是一场非常自私的对话，各位，所以请原谅我。过去我领导过设计团队，让新员工入职，确保你招到合适的人，然后还要让他们成功。这本身就是一项努力。这其中有多少需要新员工被带入 Anthropic 的软件开发方式？

Do you have to indoctrinate people into this new way of building? Like when I mean this is going to be a very selfish conversation, guys, so you have to forgive me for this. Having led design teams in the past, onboarding new employees and making sure you're hiring for the right fit, but then also setting them up for success. That's an effort all its own. How much of this requires new hires to be kind of brought into the Anthropic way of making software?

是的，这是个很好的问题。我其实很好奇你在 Facebook 的经历，因为至少在这个阶段，我们有一种文化，每个人都在一直构建原型，新员工的体验是，你进来时期待看到路线图，结果发现只是一片半成品的原型，奇怪的未来事物，其中一些正在支撑我们工作流的绝对核心，比如我们笔记本电脑上的屏幕就是我们编写的内部工具，还有一些是奇怪的副业。所以你发现这就是创新的源泉，只是文化模式匹配。你会想，那些在这里似乎受到尊重、欣赏和产生影响的人，就是那些在别人告诉他们之前就 vibe coding 新原型的人。也许我也应该这样做。

Yeah, it's a very good question. I would actually be curious to hear your Facebook experience because at least at this stage we have a culture in which everyone is building prototypes all the time and the experience of a new hire is you come in expecting to see road maps and instead what you find is just a sea of half-baked prototypes, weird future things, some of which are like powering the absolute core of our workflow like the screens up on our laptops are like internal tools that we've coded, some of which are like weird side hustles. And so you sort of just discover that this is the source of innovation and just culture pattern matching. You're like, well, the people who seem to have respect and appreciation and impact here are the ones who are just vibe coding new prototypes before anyone tells them to do it. Maybe I should do the same.

那么人们是否需要经过一个流程来决定什么会被点亮，或者你发现自己处于一种策展模式或立场，或者事情如何最终到达数亿人手中？

Do people like then have to go through a process for what gets screen lit or like do you find yourself in a curatorial mode or stance or how do things end up in the hands of hundreds of millions of people?

是的。内部牵引力可能是我们最好的信号，很多最终成为 Claude Code 和 Cloud Design 以及我们已推出和将推出的一堆其他东西，基本上就像你知道的，我们是一家拥有数千名员工、职能多样的公司，所以如果它对我们有用，那么它可能对外部世界也有用，所以这是我们的第一个迹象，然后可能有一个更传统的流程，比如好吧，这在我们产品组合中的位置、定价和包装、安全和保障，很多关于我们如何真正把它带到外部世界。所以这就是策展部分的来源。

Yeah. Internal traction is probably like our best signal and a lot of like what has eventually Claude Code and cloud design and a bunch of other things that we have launched and will launch basically like you know we're multi,000 person company with diverse functions and so if it's useful to us then it's probably useful to the outside world and so that's our first indication and then there's a probably a more traditional process of like okay like where does this fit in our portfolio pricing and packaging security and safety like a lot of the like how do we actually bring this to the outside world. So that is where the curatorial part come from.

随着 Anthropic 抓住并开创的企业级动作。你现在显然与所有这些大客户有了新的关系，他们从安全到功能有各种要求。这如何融入你刚才描述的这个有趣的软件开发环境游乐场？比如你如何真正确保你满足那些为你支付大量金钱购买 token 和功能的客户的需求？

And with the enterprise motion that Anthropic has latched on to and pioneered. You obviously now have a new relationship to all of these large customers that have all sorts of requirements from security to functionality. How does that work its way into this fun playhouse of a software development environment that you just described? Like how do you actually ensure that you are meeting the needs of the customers that are paying you gobs of money for tokens and functionality?

是的。在炉边谈话中谈论这个没那么有趣，但我们也有一个深入嵌入客户的企业团队，做一件事，一个奇怪的主题可能会在这里反复出现，就像一半的工作是疯狂的，生活在未来 3000 年，像头脑，另一半是，是的，这只是成功公司几十年来甚至几个世纪以来的做法。企业的东西基本上就是其中之一，就像我们有 PM 团队与财富 100 强公司坐在一起。他们说我们需要这种合规性和这个 skim 管理器，然后我们写一个 PRD，然后我们有一个团队构建它，我们确保它的安全。那里有一个循环。我认为它们相遇的地方基本上是，嗯，对于我们许多最奇怪的实验，我们可以相信早期采用者，比如这个房间里的人，会尝试并压力测试它，我们就会发现我们是对还是错，因为我们有这种取之不尽的供应，嗯，像技术人员愿意尝试东西。嗯，在企业中就不那么真实了，比如有一个人，他没有报名成为 AI 的 beta 测试员，他打开笔记本电脑，他们一直在用 Claude Cowork 支持所有工作，突然那里有一个新标签，他们的管理员和拥护者就像我们应该使用这个吗？我们不应该使用吗？我和一个朋友进行了一次咨询电话，他经营一家处于高度监管领域的初创公司，他说我想感谢你 Joel。嗯，过去当我们改变 UI 并重新设计时，我们的客户非常生气。所有的链接都变了，等等。但你们经常在没有任何通知的情况下改变 UI，现在我们的客户已经习惯了我们可以这样做。我说，你知道，我不认为这是你认为你在给我的赞美。所以，我认为我们需要在这方面找到一个中间立场。

Yeah. It's less fun to talk about at Fireside Chats, but we also have an enterprise team that is like deeply embedded in our customers that does a thing a weird theme that will probably come up over and over here is like half of work is the craziest living in the future year 3000 like mind And half of it is like yeah this is just how successful companies have practiced things for decades for like centuries. Enterprise stuff is like basically one of those which is like we have teams of PMs who sit with Fortune 100 companies. They're like we need this type of compliance and this skim manager and then we write a PRD and then we have a team build it we secure it. There is a there is a loop there. I think where they meet is basically like, um, for many of our weirdest experiments, we can just trust that early adopters, like the people in this room, will just try things and stress test it and like we find out when we're right or wrong because we just have this sort of like um inexhaustible supply of uh like technologists who are willing to try stuff. Um that is less true at enterprises like you have a person who did like not sign up to be an AI like uh beta tester who opens their laptop they've been powering all their work on Claude Cowork um and all of a sudden like there's a new tab there and their admins and their champions are like are we supposed to uh are we supposed to use this? Are we supposed to not? I did a um consulting call with a friend who runs a startup in like a very regulated space and he was like I want to thank you Joel. Um, it used to be that when we changed our UI and redesigned, our customers were so mad. All the like links changed, whatever. But you guys change your UI so often without telling anyone that now our customers have just come to expect that we can do it. And I was like, you know, I don't think that's the compliment that you think you're paying me. So, I think we need to like uh find a middle ground on this.

嗯，我喜欢这个。我其实想暂时转向内部工具方面。如果我要讲述硅谷的故事，我想我的角度会是内部工具，如果你看看这个行业里所有你钦佩的公司，每一个都孵化并构建了自己的工具来帮助推进使命，帮助推进他们自己的产品开发，他们自己的做事方式，他们自己的开展业务的方式。我贪婪地想知道，在这个问题的 Anthropic 篇章里有什么，比如你们发现自己正在开发什么样的内部工具，因为你们有点与未来 Claude 模型的研究团队协同工作。我猜未来的 Claude 模型是工具包的一部分。

Um, I love that. I actually want to shift for a minute over to the internal tool side of things. If I was going to tell the story of Silicon Valley, I think my angle would be internal tools that if you look at all the companies that you admire that have come out of this sector, each one has has incubated and built its own tools to help advance the mission to help advance their own product development, their own way of doing things, their own way of like conducting business. And I am like greedily wondering like what's inside the Anthropic chapter of that question like what what sort of internal tools you guys find yourself developing because you're sort of working in concert with the research teams on future models of Claude. I presume future models of Claude are are kind of part of the uh uh part of the toolkit.

但你们是不是还在围绕这个搭建更多东西，还是说只是花上好几个小时跟 Claude 聊接下来该做什么？

But is there more that you find yourselves building around that, or is it just spending hours and hours just talking to Claude about what needs to happen next?

嗯，我觉得这要分两半来看，因为我们确实搭了一些非常酷的内部工具，它们在我们研究、市场推广和产品开发流程的关键环节里都是承重的。我想在另一个平行世界里，就像 Slack 是从一家游戏公司里分拆出来的一样，这些工具完全有可能成为独立且非常成功的公司。还有另一个维度，就是我们基本上有一个一键部署，任何你想要的工具，你只要放一个内部 URL 就能用。所以我的工具栈其实跟我的同事很不一样，不同职能之间也很不一样。所以我其实很好奇，我们到底有多少个未来能独立成公司的东西，还是说真正的重点其实是那个平台、那些工具，以及随手就能写任何前端的思维方式。我们这套内部系统，你可以做 UI，但它也连接了我们所有的 MCP，而且一路到底都是 Claude。所以你能用非常有说服力的方式搭出非常复杂的东西。我觉得那个平台也许是更有意思的创新，而不是说，我有个待办清单管理器，因为我当然是个待办清单管理器，但问题是，我是怎么走到那一步的？

Well, I think there's two halves to it, because there are certainly internal tools that are very cool that we have built and are load-bearing across the key parts of our research, go-to-market, and product development process. I think in a different world, in the same way that Slack spun out of a gaming company, those could conceivably be their own standalone, very successful companies. There's another dimension, which is we have essentially just a one-click deploy where any tool that you want, you can just put an internal URL and use. And so my tool stack is actually very different from my co-workers', very different from function to function. And so I actually wonder how many standalone future companies we have, versus actually the platform and tools and mindset to just code any front end. And this internal system that we have, you can make UIs, but it's also connected to all of our MCPs, and it's Claude all the way down. And so you can build really sophisticated things in a really compelling way. And I think that platform is maybe the more interesting innovation, versus the, you know, I have a to-do list manager because of course I'm a to-do list manager, but it's like, what is the how did I get there?

但你们团队肯定也在用、也很喜欢一些第三方工具吧。有没有一些 AI 时代原生的工具，是你完全离不开的？比如对在座那些想卖进 Anthropic、想成为这些前沿模型公司必备工具包一部分的 SBC 创始人来说，进入的路径是什么？那里还有多少创新空间？

But surely you have third-party tools that the team uses and loves. So are there examples of tools that are native to the AI age that you just could not live without? Like for SBC founders here that are interested in selling into Anthropic and being part of the essential toolkit for these frontier model companies, what's the pathway in? What's the room for innovation there?

是的。我觉得有两个独特的门槛，或者说要翻越的山丘、要攀爬的高墙，基本上就是：我们对安全和安全性非常非常认真，对数据存放之类的基本上要求很高，因为我们做的东西性质如此。所以我觉得这是很容易卖不进来的一个地方。然后当然就是，我们能用补贴的 token 靠 vibe coding 基本上做出任何内部工具。而且我们内部工具之所以这么成功，很大一部分原因是它们彼此深度连接。所以如果你只看规格，或者只看静态的 UI，我们的内部工具其实可能并不惊艳。它们并不惊艳。它们惊艳的地方在于，我们让它们接入了基本上我们所有的真相来源、所有的内部工具，所以它们极其强大。我觉得对一个外部工具来说，这就是要跨过的门槛：你能不能提供超出我们自己内部能写出来的东西？我知道这话有点显而易见，但这最终就是那道门槛。

Yeah. I think the two unique qualities that are sort of the hills to climb or the walls to scale are basically, we are very, very serious about security and safety, and basically have high requirements for data housing and such, because of the nature of the stuff that we're building. And so I think that's one place where it's easy to not get bought. And then of course it's just, we can vibe code with subsidized tokens basically any internal tool. And so much of the success of our internal tools is that they are deeply connected to all the rest of them. And so our internal tools are actually probably not mind-blowing if you saw the spec or you saw the UI just in still life. They wouldn't be mind-blowing. What's mind-blowing about them is that we've given them access to basically all of our sources of truth, all of our internal tools, and so they are immensely powerful. I think for an external tool, that's the bar to clear: do you provide something beyond what we can code internally? Which I know is kind of the obvious statement, but that is ultimately the bar.

不过你觉得，对那些能在转型中活下来的企业来说，这会成为未来吗？我们是不是正在进入一个 SaaS 实际上正在消亡、而那种特别适合某家特定业务的自制工具成为新常态的世界？

Do you think that's going to be the future though for enterprise businesses that survive the transition? Like are we entering a world where effectively SaaS is dying and homespun tools that are kind of uniquely suited for that particular business is a new normal?

我觉得有很多——我在想，除了写代码之外，一个外部供应商能给我们带来什么我们自己做不到的东西？对我来说和我们团队来说，日常使用频率最高的两个，一个是 Listen Labs。他们的产品非常棒，他们带来的是外部受众和用户的招募，这是我们不希望内部投入去做的事。然后 Figma 也是日常大量使用。关于产品是如何被做出来的，有太多智慧，关于做出精美产品开发需要什么，有太多深思熟虑，我觉得这些并没有被——如果你去给一个设计工具写规格，我觉得它捕捉不到这些。所以我觉得，那些能带来一种观点的公司，基本上就是“做某个学科或某种实践的正确方式是这样的”，我觉得这是我们内部并不总是有的东西。

I think there is a lot of — I'm sort of thinking, what would an external vendor bring to us that we couldn't do ourselves, outside of just writing lines of code? And so two that are top of mind for me and our team, which is very daily use, is Listen Labs. They have a fantastic product, and they bring the external audience and the finding of users, and so that's something that we don't want to invest in internally. And then Figma is in tons of daily use. There is so much wisdom about how products get made and so much thoughtfulness around what it takes to do exquisite product development that I think is not captured in — if you were to write a spec for a design tool or something, I don't think it would capture it. And so I think companies that bring a point of view on basically this is the right way to do a certain discipline or a certain practice, I think that's a thing that we don't always have internally.

我很高兴你提到 Figma。我们时间线上有人提到你们有个产品叫 Claude Design。它是怎么来的，Claude Design 的未来大概是什么样？你怎么看那个产品套件，以及 Anthropic 所有与设计相关的努力，如何与现有的设计工具协同工作？

I'm so glad you brought up Figma. One of us on the timelines that you guys have a product called Claude Design. How did that come about, and what's sort of the future for Claude Design? How do you sort of see that product suite and all the design-related efforts by Anthropic work in concert with existing design tools?

是的，就像前面说的，我们并不是——我们并没有按 TAM 列出所有市场，顺着清单往下开始做产品。我们每天都用 Claude Code。我们发现模型在写代码、或者做前端设计方面越来越好。内部有个设计师做了个 skill，基本上帮了一点忙。进展还不错，但我们意识到撞到了天花板，最终你确实想跳出命令行。所以我们做了个内部版本，你知道，Claude Design 本质上就是 Claude Code 外面套了个前端，再加一些提升体验的功能。这就成了 v0.1。我知道这话很烦人，但我真不知道它会走向哪里。我们真的没有什么宏大计划。我可以告诉你我认为它可能、或者也许应该走向哪里，因为我觉得整个产品生态在这一块非常欠缺，那就是搞清楚你在想什么，把东西从你脑子里画出来。Claude Design 里我最喜欢的功能其实不是任何 UI 生成。而是当你向它要东西时，它会弹出那个问卷，说，你是想要 A、B 还是 C？或者，你问题的前半部分和后半部分，我该发散到什么程度？或者，你想的是网页还是移动端？几乎每次它问我这些问题，我都会意识到，要么是我有东西没想过，要么是它在几件事之间画出了联系。而那，我觉得是这个流程和这份工作里最难的部分。而且我觉得——这绝不是对路线图的任何承诺——但我会很乐意用一个产品，不管是我们做的、Figma 做的，还是你们做的，它能把我脑子里那个粗糙、古怪、几乎难以辨认的想法抽出来，问我问题，帮我走过那座想法迷宫，走到某个地方，让你觉得，就是这个，这有共鸣，然后再把它传达给周围的人。

Yeah, I mean, as per the previous comment, we did not set out — we did not list all markets by TAM, go down the list, and start building products. We all use Claude Code daily. We discovered that basically the models were getting better and better at writing code, or at making front-end designs. Someone internally, a designer, made a skill which basically helped a little bit. That was going pretty well, but we realized we were hitting a ceiling where eventually you really did want to break out of the command line. So we built an internal version of, kind of, you know, Claude Design is essentially just a wrapper over Claude Code with a front end and some additional quality-of-life features. And that became the sort of v0.1 of things. I know this is really annoying, but I have no idea where it's going to go. We really don't have a grand plan. I can tell you where I think it could or maybe should go, because I think it's very underserved by the entire product ecosystem, which is figuring out what you're thinking, drawing stuff out of your brain. My favorite feature of Claude Design is actually not any of the UI generation. It's that when you ask it for something, it brings up that questionnaire and it's like, well, did you mean A, B, or C? Or, how divergent should I be on the first part of your question or the second part of your question? Or, were you thinking web or mobile? And almost every time it asks me those questions, I realize that I either hadn't thought about something, or it had drawn a connection between things. And that, I think, is the hardest part of the process and the job. And I think — this is not a commitment to a road map in any way, shape, or form — but I would love to use a product, whether we make it, whether Figma makes it, whether you all make it, that took a raw, weird idea that was sort of barely legible in my own head, drew me out, asked me questions, helped me navigate the idea maze and get to something where you're like, this is it, this resonates, and then communicate it to people around you.

我们来聊聊你到目前为止在 Anthropic 的旅程。有没有一些关键时刻，或者你对今天的产品体验做出了独特贡献，可以作为在 Anthropic 领导设计的实践案例？当你回顾团队迄今为止的成就时，哪些是你的高光时刻，让你觉得“这是我的功劳”，或者团队在某个贡献、某个决策点、某个设计上指望你，帮助人们更有效地使用 Claude？

Let's talk a little bit about your journey thus far through Anthropic. Are there moments where you had a crucible moment or where you made a unique contribution to the product experience today that might act as case studies for what it means to lead design at Anthropic in practice? When you look back on things that the team has accomplished thus far, what are the highlights for you where you're like, that's a feather in my cap, where I feel like this is where the team's looking to me for this particular contribution or this particular decision point or this particular piece of design that helps people work more effectively with Claude?

是的。所以我认为，如果你接受我之前说的观点，即模型是大部分价值所在，上面有一个薄薄的产品包装。很容易想象界面基本上无关紧要。你可以从认为聊天如此多功能，可以做任何事情的角度，或者从 AGI 最大化主义的角度，即最终 Claude 会足够聪明，能自己制作软件并做任何你想做的事，所以沿途在用户体验上攀登是没有意义的。我认为并不是有人明确说过那样的话，或者他们雇用了我，让我组建团队等等。这不像是一个官方立场，但我认为 artifacts 真的是一个大的转折点，最初版本的 artifacts 是最薄的 UI 技巧。基本上就是，看聊天响应。如果 Claude 决定渲染 HTML，不要把它放在一个小盒子里，而是放在右侧的一个大盒子里。当你描述它时，它看起来如此单薄，几乎愚蠢，但在使用它的体验中，这是一种全新的心智模型转变，你会觉得，我在用这个工具制作东西，我能看到我正在制作的东西，我可以把关于我正在制作的东西的元对话与正在制作的东西分开。它改变了很多我们构建的方式、思考的方式和接近事物的方式。所以我认为那是一个看似小的 UI 变化却产生了巨大差异的例子。

Yeah. So I think if you take the point that I made earlier that basically the model is a majority of the value and there's a thin product wrapper on top. It is very easy to imagine that the interface essentially doesn't matter. You could either take that position from thinking chat is so versatile that it can do everything, or a sort of AGI maximalist position which is that eventually Claude will just be smart enough to make its own software and do whatever you want, and so there's no point in hill climbing on UX along the way. And I think not that anyone said that explicitly or, you know, they hired me and they let me build a team and all this stuff. It was not like an official position, but I think artifacts was really the big turning point where the very first version of artifacts was the thinnest of UI tricks. It was basically like, look at the chat response. If Claude has decided to render HTML, instead of putting it in a tiny box, just put it in a huge box on the right side. It really seems so thin and almost silly when you describe it, but then in the experience of using it, it is this whole mental model shift where you're like, I'm using this tool to make something and I can see the thing that I'm making and I can separate my meta conversation about the thing I'm making from the thing that I'm making. And it changed a lot of how we build and how we think and how we approach things. And so I think that was one where a seemingly small UI change made a big difference.

我喜欢这个，因为我觉得这就是构建型设计师的福音，举个例子，对吧？在这种情况下，是 Michael 吗？太好了。我第一次见到你是因为你在 Quora 工作，Quora 是这种特定类型设计师的绝佳温床。能写代码的设计师，能构建的设计师。当你雇用能构建的人时，你得到的是直接原型化和尝试想法的能力，直接与想法交互，而不是抽象地，不是通过某种 PRD 或在 IRC 或 Slack 或你使用的任何聊天频道上进行热烈讨论。当你审视你如何建立这个团队时，除了能够快速原型化，你认为还需要雇用哪些其他品质，以便某人能够在这个新媒介中成功，比如前沿模型作为你正在围绕的核心原语？你在这些招聘中寻找什么？为什么原型化、技术能力、构建是其中的一部分？

And I love that because I feel like this is the gospel of designers who build, sort of case in point, right? In this case was it Michael who—great. So I first met you because you were working at Quora and Quora was a wonderful breeding ground of this particular breed of designer. Designers who could code, designers who could build. And what you get when you hire somebody who builds is the ability to just prototype and try out ideas and actually interface with that idea directly rather than in the abstract rather than some sort of PRD or in some sort of lively discussion over IRC or Slack or whatever chat channel you're using. When you look at how you've kind of built this team, along with being able to prototype quickly, are there other qualities that you feel you need to hire for in order for somebody to be successful with this new medium like the frontier model as the core primitive that you're working around? What are the things you look for in these hires? And why is prototyping, why is being technical, why is building a part of it?

是的。我能快速推销一下 Quora 做过的一件事吗？我至今在 15 多年里还没有遇到过类似的东西。

Yeah. Can I quickly pitch a thing that Quora did that I still have not encountered in like 15 plus years since then?

请讲。

Please do.

他们投资了整个自定义 Python 库，基本上让任何人都能推送到生产环境，我们每个人都有自己的开发盒子，永久 URL，可以分享那个 URL。所以我有像 joel.dev.quora.com 这样的地址，我可以复制粘贴链接。我可以推送到生产环境。我可以复制粘贴链接并分享给某人。我知道这听起来愚蠢而具体，但我实际上认为自己构建东西是，我不知道，三分之二的战斗，但把它交给别人，让他们使用，而不必经过一些奇怪的标志 AB 测试，只是像，这有趣吗？我做了这个，你知道，那时可能需要几天，现在只需几小时。但即使在 Anthropic，我们整天都在原型化。我们分享的很多方式是通过屏幕录制，就像录制我们自己的 Loom 视频。我们有推送到生产环境并查看的方法。但我惊讶于其中很多仍然是工作原型的视频，我认为 Quora 在那里解锁了一些非常真实的东西，即可分享、可用、深度交互。在招聘方面，我认为是原型化。我认为是无畏，不害怕模糊性，并想要开辟一条道路。有越来越多的人——几乎民主化了——能够用代码制作东西。但我认为是回答问题。Quora 有两个独特之处，我认为与此相关。一是构建和发布代码。另一个是非常假设驱动的产品开发，真正思考，我相信这是关于世界的真实，但我不确定。我要构建这个原型，不仅因为它酷，而且因为它像科学方法一样，专注于特定问题。我们并没有那种科学严谨的 AB 测试文化。但我认为那种心态，即我要通过构建来学习。我要通过构建来理解世界，自己使用，展示给用户，等等，真正建立我对事物如何运作的理解，宇宙的真相是一种特定的思维方式。

So they'd invested in this whole custom Python library to basically let anyone push to production and we each had our own dev box that was perma-urled and could share that URL. So I had like joel.dev.quora.com and I could copy and paste a link. I could push to production. I could copy and paste a link and I could share it with someone. And I know that sounds stupid and specific, but I actually think the building the thing yourself is, I don't know, two-thirds of the battle, but handing it to somebody else and letting them use it and not having to go through some weird flag AB test thing, but just sort of like, is this interesting? I did this, you know, back then it would have been days, now it's hours. But it's not like even at Anthropic where we're prototyping all day every day. A lot of how we're sharing it is with screencasting, just recording videos like Looms of ourselves. We have ways of pushing to production and seeing it. But I'm surprised by how much of it is still videos of working prototypes and I think Quora had unlocked something very real there of sharable, usable, deeply interactive. In terms of hiring, I think it is prototyping. I think it is fearlessness and wanting to not be afraid of ambiguity and wanting to carve a path through it. There are lots of increasingly number of people—it's almost like democratized at this point—to be able to make something in code. But I think it's the answering the question. And Quora had two unique things to it that I think are relevant here. One is building and shipping code. The other is very hypothesis-driven product development and really thinking in terms of, I believe this to be true about the world but I'm not sure. I'm going to build this prototype not just because it's cool but because it is scientific method-like, focused on a particular question. And we don't have quite that culture of scientific rigor AB testing. But I think that mentality of, I'm gonna build to learn. I'm going to build to understand the world, to use it myself, show it to users, whatever, to actually build up my understanding of how things work and the truth of the universe is a particular way of thinking.

是的。你目前如何考虑围绕这个有效的产品套件组织设计，鉴于你有这些实验，有时它们会上市，有时它们会起飞。它们真的有了自己的生命。当你考虑团队的组织时，你如何思考确保既有覆盖范围，又正在构建能力，这些能力不仅映射到 Anthropic 组织，而且开始真正深入到该产品解决的特定问题领域。

Yeah. How do you currently think about organizing design around this effective product suite seeing as you have these experiments that sometimes they go to market and sometimes they just take off. They take a life of their own literally. When you sort of think about the organization of the team, how are you thinking about making sure that you have both coverage but that you also are building the capabilities that don't just map to Anthropic the organization but sort of start to go really deep into the particular problem area that that product solves for.

是的。

Yeah.

这大概属于那 50% 老生常谈的公司组织管理问题，就是我们有过的一切组织架构都是一种不神圣的妥协，把复杂性在不同地方之间推来推去。我可以给你描述一个大致结构，听起来应该能行：有一个 Claude Code 团队，有一个 API 平台团队，有一个企业和大公司团队，有一个消费者团队。然后每个团队基本上都在试图专注于他们本该专注的用户，同时有意或无意地做出和其他团队高度相关的东西。然后我们就有了这种疙疙瘩瘩的变形虫形状——你知道，技能和定制化其实非常承重，却莫名其妙地归增长团队管。但现在 co-work，那个桌面应用，其实非常依赖它。于是我们就会重组，然后未来又会重复同样的问题。我觉得我们就是经常重组。

This is probably in the 50% that is like age-old company org management, which is like every org structure we've ever had has been like an unholy compromise that sloshes complexity around between different places. I can describe to you the sort of rough structure which sounds like it should work, which is like there's a Claude Code team, there's an API platform team, there's an enterprise and sort of big company team, there's a consumer team. And then each of them are basically attempting to focus on the users that they are meant to focus on and either accidentally or intentionally build stuff that is highly relevant to other teams. And then we have this sort of lumpy amoeba shape where it's like, well, you know, skills and the customization is actually really load-bearing but was weirdly owned by the growth team for reasons that make no sense. But now co-work, which was a desktop app, actually really relies on it. And so then we'll reorg and then it'll just repeat the problem in the future. I think we just reorg a lot.

你们有没有依赖 Claude 来对你们的运营卓越性提供评论？我其实对「Claude 作为同事」这个想法很感兴趣——作为一个新兴的超级智能，它是 Slack 的一部分，和你们每个人对话。Claude 在帮助确保你们让所有列车准点运行方面扮演什么角色？

Do you guys ever lean on Claude to provide commentary on like your operational excellence? I'm actually quite intrigued with the idea of like Claude as a co-worker, as an emerging superintelligence who's a part of Slack who talks to each one of you. What role does Claude play in kind of helping to ensure that you're keeping all the trains on time?

我们非常依赖 Claude 来了解正在发生什么。我用的一个我很喜欢的内置应用会扫描我们所有的 Slack 频道，看人们发了什么，试图把它整合成「这里似乎有一个设计热点」。它没有项目名，而且是一些本不该一起工作的人出于某种奇怪的原因在一起工作，然后它会说你应该去看看那个 Slack 频道。我不确定我们发现它在综合信息或建议如何推进方面有多大用处，因为我觉得它能反映出存在某种复杂性。而且你往往并不知道正确答案是什么。所以我们现在正处于这样一个过程中——我觉得我们已经经历了 6 个月的非常生成性的阶段，我们确实乘着 Opus 4.5 的能量，发布了很多产品，而我会第一个承认，这些产品并不总能凝聚成我们用户能理解的心智模型。所以我们完全清楚需要整合，需要做一些意义建构。

We rely on Claude a lot for awareness of what is going on. One of the internal apps that I use that I love just scans all of our Slack channels for what people have posted, tries to cohere it into like, there appears to be a hot spot of design happening here. It's like, doesn't have a project name, and it's like people who shouldn't be working together but are for some weird reason, and like you should check out the Slack channel. I don't know that we've found it that useful for like synthesizing or advising on like how to move forward, because I think it can reflect that there's a complexity. It's also like you often don't know what the right answer is. And so we're sort of in a process of this now, of like, I think we've been in a very generative phase for 6 months and we've really like ridden a lot of the like Opus 4.5 energy and we've put out a lot of product that, like, I will be the first to admit, don't always cohere into a like mental model that our users understand. So we are fully aware that we need to consolidate, do some sense-making.

Claude 在综合方面还行，但我们还没看到它基本上做出那最后的一跃，做出那个艰难的取舍，达到那种「哦，这就是——如果有一个答案能让这一切说得通，那我们早就想出来了」的程度，我觉得。所以我认为那仍然是一个非常人类的过程。

Claude is like okay at synthesis but we have not seen it basically like make that final leap, make that hard trade-off, like come to the sort of, oh this is the—if there was an answer for how it all made sense then we would have figured it out already, I think. And so I think that is still a very human process.

这让人安心——至少在这里，我们在组织他人方面永远有优势。也许反过来问：有没有一些情况让你惊讶，创业公司和小团队、Anthropic 之外的团队，居然没有在用 Claude？如果你能挥一挥魔杖，改变人们思考如何最大化利用 Claude 的方式，你会给出什么建议？你怎么看人们在 Claude 上摸索的样子——那些你们 12 个月前也许也摸索过、然后意识到「不，这才是从这个工具栈里榨取最大价值的唯一正道」的地方？

That's reassuring—here at least we'll always have an advantage in terms of organizing other people. Maybe flipping it on its head: are there ways in which you're surprised that startups and smaller teams, teams outside of Anthropic, are not using Claude? If you could wave your wand around and change the way in which people think about how they can get the most out of Claude, what advice would you give there? How do you sort of see people fumbling around with Claude that you guys have maybe fumbled around with 12 months ago and have realized, no, this is like the one true way to get the most out of this tool stack?

每次我们尝试使用任何不是纯粹 Claude Code 的东西，最后都会直接回到 Code。这包括 Claude Design，也包括我们自己构建的其他内部工具。我觉得有一种——我们都觉得命令行不可能是命令行的最终形态，或者桌面应用之类的，不可能是软件构建的最终形态。然后事实证明，把生产代码库作为这么多事情的事实来源，就是极其有用。它是你产品、你的功能、事情运作的规则、你的设计系统的事实来源——所有这些你本来不得不搬进另一个系统的东西。所以对我们来说，Claude Design 就是一个很好的例子。Claude Design 在我们的使用中对于全新的从 0 到 1 非常棒。我在思考，我在从无到有地做梦。但如果你说，代码，我有四个设置页面，我需要把它们基本合并成一个新的导航，你能帮忙想清楚吗？如果你把那个交给 Claude Design，它就得在一堆不是事实来源的 Google Docs 里翻找，它很棒因为它会让你问——向你提问——而且非工程师也能用，但它就是不知道实际发生的地面真相是什么。所以我们很多设计师就直接回到 Claude Code，他们说，这种贴近金属的感觉太有帮助了，我们就继续这么做。

Every time we try using anything that isn't just like raw Claude Code, we end up just going right back to Code. This includes Claude Design, this includes like other internal things that we've built. I think there is a—we all feel like the command line couldn't possibly be the command line, or the desktop app or what have you, couldn't possibly be the like final form of software building. And then it turns out that it's just extremely useful to have like a production codebase as the source of truth for so many things. It's a sort of truth for your product, your features, the rules of how things work, your design system—like all of these things that you have to like move into another system. And so for us, Claude Design is like a really good example. Claude Design is to our usage fantastic for like brand new zero to one. I'm like thinking, I'm dreaming from like no first principles. But if you're like, code, I have four settings pages and I need to basically combine them into a new nav, can you help figure it out? If you give that to Claude Design, it has to like rummage around in a bunch of Google Docs that are not the source of truth, and like it's great because it makes you ask—and ask you questions—and it's accessible to non-engineers, but it just like doesn't know the actual ground truth of what's going on. And so many of our designers just go back to Claude Code and they're like, this—the proximity to the metal is like so helpful that we just continue to do that.

你觉得这种反转会变得在所有团队中司空见惯吗——事实来源会转移，记录系统开始默认变成生产代码库，而抵制它的团队就是在抵制未来？你觉得这能代表 Anthropic 之外的团队将会发生的情况吗？

Do you see that that inversion is going to become just sort of commonplace across all teams then, that we're going to be moving the source of truth—that the system of record starts to become the production codebase sort of by default, and that teams that are resisting it are resisting the future? Do you think that's indicative of what's going to be happening for teams outside of Anthropic?

我觉得可能有两三种可能。一种是——基本上那就该由我们来解决，比如 Claude Design 显然应该知道你代码库里在发生什么，那只是我们的锅，我们应该修好它。所以有可能只是工具会追上来。我觉得也有可能是 Claude 可以自动化创建非代码库的事实来源，比如你关于哪些页面存在哪些设置的文档，目前只存在于代码里。它进入文档的方式是，一个人加载你的设置页面，然后写下变更。我觉得很可能在不久的将来会有自动化流程来做这件事，然后你就不需要——你就不需要每个人都去看源代码材料，因为那对很多职能来说可能太难了。但这一切可能都下游于代码就是答案这件事。是的。

I think there's maybe two or three possibilities. One is that—and basically like then it's incumbent upon us to solve the sort of like—like Claude Design obviously should know what's going on in your codebase, that's just like our bad, like we should fix that. So it's possible the tooling just catches up. I think it is also possible that like Claude can automate the creation of sources of truth that are not codebases, like your documentation for what settings exist on what pages currently exists only in code. The way it gets into a document is that a human being loads your settings page and like writes down the changes. I think there is a probably near future in which like there are automated processes doing that, and then you don't need—you don't need like everyone to be looking at source code material, because that feels probably too hard for lots of functions. But it's all probably downstream of like code being the answer. Yeah.

然后我想也许一个相关的问题是围绕，你知道，对正在发生什么的共同理解。至少对我来说——我在这方面很有 Facebook 思维——Claude 所提供的东西和团队不得不围绕讨论想法、拥有共同的地面真相感来组织自己的方式之间，存在一个缺口，而不仅仅是 Claude 告诉我什么是真的。

And then I guess maybe a related question is around, you know, common understanding of what's going on. There's, to me at least—and I'm very Facebook brain in this way—there's like this gap between what Claude offers and how teams have to kind of organize themselves around just like discussing ideas, having a common sense of ground truth, not just like what Claude tells me is true.

然后还要协调，好，或者一起推测，嘿，这可能是我们下一步想走的方向，大家是否都同意，我们今天就此推进——这些事就散落在 iMessage、Slack、邮件之类的地方。Anthropic 有没有一种原生属于 AI 的观点，想要推向市场，或者它如何想象人类将如何与越来越多的智能体互动——这些智能体能力更强，但也想和我们对话，不是作为个人，而是作为一群人、作为团队？

And then coordinating around, okay, or speculating around, hey, this is maybe the direction we want to take next, are we all in agreement, let's move on today — that lives in some combination of iMessage or Slack or email or what have you. Is there a view that Anthropic has that's sort of native to AI that it wants to bring to market, or that it imagines how humans will be interacting with an increasing set of agents that have higher capabilities but also are going to want to be dialoguing with us not as individuals but as groups of people, as teams?

这是个非常好的问题。这就像——呃，我觉得我不会在台上哭出来，但我在 Anthropic 每天经历的痛苦恰恰就是这个问题：事情太多了，太混乱了，太难追踪了。

That is a very good question. That is like — uh, I don't think I'm going to cry on stage, but the pain that I experience day-to-day working at Anthropic is exactly this problem of just like there's so much going on and it's so chaotic and it's so hard to track.

所以我们是在找这个吗？这是不是现场在招募创业公司？大家都注意听了吗？

So are we looking for this? Is this a request for a startup right here? Is everybody paying attention?

是的，我不认为我们有——我可以告诉你两件事。我们没有一个第一性原理式的信念，也没有一个正在开发的产品，是那种“真相最大化器”。我们每天的亲身经历就是在 Slack 评论的浪潮上冲浪，试图逆向工程出可能正在发生什么。所以即便我们做了一些工具，它们也追不上这个问题。我很好奇——这件事其实已经出现过几次了，我想很多这类问题都会继续出现——到了某个点，它们就变成了人的问题。就什么是真的、计划是什么达成一致，说“对，我们走这条路”“不，我们不走这条路”，这需要领导层、高管、管理者、层层传达的沟通，以及一堆前 AI 时代的组织机制。我觉得我们离这样一个世界已经很近了：Claude 可以呈现——可能今天就能做到，只要你设置得当——它可以向你的 CEO、CPO 呈现：你看，我们可以走这三个战略方向，这里是利弊清单，三分之二的员工想要 X，你的 VP 层想要 Y。也许它甚至可以说，如果由我决定，我会选 Z。但最终人们还是得认同那个决定，做到“不同意但执行”，然后开始推进。我认为那仍然是一个非常非常人类的问题。

Yeah, I don't think we have — I can tell you two things. We don't have a first-principles belief or a product in the works that is like the source-of-truth maximizer. And our lived experience day-to-day is just like surfing the wave of Slack comments and trying to basically reverse engineer what might be happening. So even to the degree that we build tools, they're not catching up to this problem. I am curious — basically all this has come up a couple times and I think we'll continue for many of these problems — like at some point they become human problems. And agreeing on what's true and what's the plan and saying yes we're doing this path and no we're not doing this path requires leadership and executives and managers and cascading comms and a bunch of the sort of non-pre-AI sort of org mechanics. I think we're very close to a world in which Claude can surface — probably today, if you set it up right — it could surface to your CEO, CPO, you know, like there are these three strategic directions we could go, there's a pro and con list here, two-thirds of your employees want X, your VP tier wants Y. And maybe you could even say, like, if it were up to me, I would choose Z. But eventually people have to buy into that decision and disagree and commit and start making progress. And I think that is still a very, very human problem.

我想花一点时间谈谈向这种新的智能体经济的转变，因为在我看来，今天智能体只占全部经济活动的一小部分。但我预计，随着时间推移，智能体与智能体之间的互动、交易、价值交换可能会增长到甚至压过人类经济的规模，对吧？在这个世界里，我认为会发生的是：如果智能体能够进行价值交易——无论是货币，还是“我给你做样东西，你给我信息”——你就会开始看到人类组织与这种新兴的智能体经济之间出现需求的分化，后者的偏好不断演化，而且是在我们所有人睡觉的时候发生的。你打算如何为这两种分化的经济构建产品？你觉得这真的是即将到来的事吗？这会不会——就像 Dario 在某种程度上预言了 AGI 世界中民主治理的未来那样——是你们内部会讨论的事情，关于你们如何思考未来经济的形态？

I want to talk a little bit about the shift into this new agentic economy for a minute, because the way I see it is, agents today represent a very small minority of all economic activity. But I anticipate that with enough time, agent-to-agent interactions, transactions, exchange of value may grow to a size that overwhelms even the human economy, right? And in this world, what I think will transpire is that if you've got agents that can transact in value — whether it be currency or like I make you a thing and you give me information — you start to create this divergence of need between human organizations and this emergent kind of agentic economy whose whims are constantly evolving and happening while all of us sleep. How do you anticipate building products for these two divergent economies? Like, do you think that's actually a thing that's coming down the pipe? Is that — in much the way that Dario was kind of foretelling the future of democratic governance in a world of AGI — is this something that gets discussed internally in terms of how you think about the shape of the economy to come?

这也是个非常好的问题。我总会回到的一点，基本上就是给智能体提供指引和方向。有点像——我的意思是，存在某种世界，它们完全自组织、自我——是一个完全闭环、不需要人类干预。我认为那是一组非常不同的问题，可能也不需要面向用户的产品。我认为存在一个世界，它们是在人的方向下执行。我在这里一直用的类比基本上就是组织里的不同层级。比如有 IC 做一线执行，有经理基本上确保这些 IC 做好工作，有总监，有 VP，有创意总监。而我想象成为 CEO 或总统之类的角色，基本上就是：成千上万的人类小时已经投入到一个决定或方向里。15 分钟内，你面前摆着两条路，你用品味、观点、经验、直觉的某种组合来说这个。有一句我很喜欢的奥巴马名言，他说，当总统最难的部分是，只有那些没有明显答案的问题才会落到你桌上。如果这个问题有任何可能的简单解法，我的劳工部副部长早就解决了。所以我做的所有事——他是在说这份工作真的很难，但我觉得其中有一种赋权感，那就是：在智能体商业的世界里，我希望用户不必批准每一笔购买或管理自己的预算。我希望他们像宏大的室内设计师，他们会说，我想要农舍美学，那是一种品味、一种观点，是你自己对世界的价值观、希望和梦想的表达，然后经济做经济该做的事，带着答案、购买和东西回来。但对其中许多事情，我认为仍然必须有某种程度的创意方向输入。

Also a very good question. I think the thing I come back to is basically giving guidance and direction to agents. And sort of like — I mean, there's some world where they're fully self-organizing and self — it's a fully closed loop that doesn't need human intervention. I think that's a very different set of problems and probably doesn't need user-facing product. I think there's a world where they are executing upon the direction of people. And the analogy that I always use here is basically like different levels of an org. Like you have ICs who are doing on-the-ground execution, you have managers who are basically making sure those ICs do their job, you have directors, you have VPs, you have creative directors. And what I imagine being a CEO or a president or something is like — basically like thousands of human hours have gone into a decision or a direction. In 15 minutes, you're presented with basically two paths and you use some combination of taste, opinion, experience, gut to basically say this. There's this famous Obama quote that I really love, where he was like, the hardest part of being president is the only problems that hit your desk are the ones that there was no obvious answer for. If there was any possible easy solution to this problem, my undersecretary of labor would have solved it. So all I do — he was saying it in a sort of this-job-is-really-hard way, but I think there's something kind of empowering about it, which is like, in the agentic commerce world, I hope that users are not having to approve every purchase or manage their budgets. I hope they are like grandiose interior designers where they're sort of like, I want a farmhouse aesthetic, and that's a taste and point of view and an expression of your own values and hopes and dreams about the world, and then the economy does what the economy does and comes back with answers and purchases and things. But for many of those things, I think there still has to be some level of creative direction put in.

是的，我很高兴你提到这个，因为我觉得今天很大一部分观众可能想听听你对这项技术会把设计领域带向何方的看法。你知道，当你把你的团队当作案例研究时，Anthropic 的设计负责什么？你认为在更长的弧线上，比如未来两到五年，软件设计领域会演化成什么样？我们主要会负责什么？Twitter 上流传着一张图，我觉得来自《金融时报》，所以我认为是真新闻不是假新闻，基本上是说应用提交量暴涨，应用的平均评分持平，而高使用量应用的数量要么持平要么下降。这是一个“垃圾内容鸿沟”吗？是这样吗？

Yeah, I'm glad you brought this up, because I feel like a big chunk of the audience today probably wants to hear your thoughts on where this technology takes the field of design. You know, when you're using your team as a case study, like what is design at Anthropic responsible for, and where do you think over the longer arc, say the next two to five years, the field of software design will evolve into? Like, what are we going to be principally responsible for? There's a chart going around Twitter that I think comes from the Financial Times, so I think it's real news and not fake news, which is basically like app submissions have gone through the roof, average reviews of apps have stayed flat, and number of highly used apps is either flat or down. Is this a slop gap? Is that what it is?

是的。是的。是的。是的。是的。所以我认为设计师必须弥合这个差距。嗯，我觉得有——我觉得也许有两条潜在路径。一条是，如果我们没有搞清楚——如果我们没有把代码生产力转化为人们想用的新东西，那是设计可以解决的问题。我做过的一个思想实验是，我认为 Anthropic 在过去六个月里以相当惊人的速度发布了新功能。

Yeah. Yeah. Yeah. Yeah. Yeah. So I think designers like have to close that gap. Well, I think there's — I think maybe there's two potential paths. One is like, if we are not figuring out — if we are not translating code productivity into like new things that people want to use, that is a problem that design can solve. A thought experiment that I've done is like, I think Anthropic has shipped new features at a fairly blistering pace in the last six months.

我做过一个思想实验：如果每家公司、每个创业公司都以那种速度发布功能会怎样？我觉得用户会直接造反。他们会把手机扔进海里，然后说：这不是——各位，这是你们的工作。搞清楚什么才是好的，然后给我好东西。我不想看到剪辑室里那些乱七八糟的废片。所以我认为设计的作用就体现在这里：什么才值得占用别人的注意力，以及你是否选择了用正确的方式解决正确的问题？当你请求别人付出注意力时，你是在把它用在值得的事情上吗？这可能是其中一条路径。

The thought experiment I've done is: what if every company and every startup was shipping features at that pace? And I think users would just revolt. They would just throw their phones into the sea and they would be like, this is not—guys, this is your job. Figure out what's good and then give me something good. I don't want to see every weird cut on the editing floor. And so I think that's where design is going to come in: what is worthy of someone's attention, and are you choosing to solve the right problem in the right way? And when you ask for someone's attention, are you using it for something that's worth it? That's maybe one potential path.

另一条路径是：不再有所谓的一体化应用，我们都在使用某种定制的、自制的小应用，来解决我们各自具体的问题。Wabby 就是我最喜欢的例子，里面有上百万个非常有用的小工具。我觉得在那个世界里，更多的是设计平台来让人们做出这些东西。

The other path is like there's no such thing as monolithic apps and we're all just using sort of bespoke homemade apps that are solving our exact problem. Wabby is like my favorite example of this, of just like a million utilities that are so helpful. And then I think in that world it's much more like designing the platform to make those.

你能快速介绍一下 Wabby 吗？

Can you describe Wabby for real quick?

哦，当然。它基本上就是一种氛围——它是一个 iPhone 应用，你可以在里面用 vibe coding 写一些小应用，它们就住在 Wabby 里。所以你可以做小工具或更小的体验，基本上就是把它提示出来，然后收藏、定期保存。我有一堆这样的东西，用来解决那些太小、不值得指望一体化应用去解决的问题，但它们极其有用，而且完全贴合我的具体需求。我觉得这可能是——如果现在我们都在用同样的七个应用，一条路径是很快会变成 10 个非常不同、非常用心的应用。另一条路径是我们每个人用 50 或 100 个应用，各不相同，都是为我们量身定制的。

Oh yeah, sure. It's basically like a vibe—it's an iPhone app where you vibe code small apps that live within Wabby. So you can make utilities or smaller experiences where you've basically just prompted it into existence and then bookmarked it and saved it regularly. And I have a variety of these for things that are much too small of problems for me to expect a monolithic app to solve, but are extremely useful and wrapped around my exact needs. And I think that's potentially the—if currently we all use the same seven apps, there's one path which is soon that will be 10 apps which are really different and really thoughtful. Another is we're using 50 or 100 apps, each of us different ones that are sort of custom made for us.

这项技术经历了我在整个职业生涯中见过的最剧烈的扩散。就连智能手机，我也不认为它的普及速度能比得上 Claude、ChatGPT——所有前沿模型。但仍然存在普及的瓶颈。你目前还在纠结哪些令人沮丧的瓶颈？我问这个问题的原因是，我知道你非常有政治意识、有政治取向，而我觉得我们这个时代的地缘政治问题就是这个扩散问题——如果我们不培养人们熟练使用这些模型的能力，我们就会无意中制造出新的数字鸿沟。但我不认为这从根本上是一个软件设计问题。某种程度上，这是在教人们如何思考。那么，你认为这项技术扩散的瓶颈有哪些？更重要的是，如何赋能那些最可能从 AI 中受益的人？

This technology has gone through the most dramatic dissipation that I've seen my whole career. Even the smartphone, I don't think, has been adopted as rampantly as Claude, ChatGPT—all of the frontier models have. And yet there are still bottlenecks to adoption. What are some of the frustrating bottlenecks that you're still grappling with? The reason why I asked this question is because I know that you're very politically minded and oriented, and I feel as though the geopolitical issue of our age is this dissipation question—that we will inadvertently create a new digital divide if we don't create fluency in what these models are capable of. But I don't think that this is fundamentally a software design issue per se. It's like teaching people how to think, to a certain extent. So what are some of the bottlenecks that you perceive for the dissipation of this technology, and more importantly, the empowerment of people that could stand to benefit the most from AI?

是的。我觉得——有意思的是，你不认为那是个软件设计问题，因为我觉得我看到的瓶颈基本上就是人们把它当作谷歌的替代品，而它是一个极好的谷歌替代品。然后中间还有一层——再然后就是编程，那是完全另一回事。而中间那一层，是彻底改变你学习一个主题、构建某个东西，或者以更实质性的方式解决问题的方法。

Yeah. I mean, I think—it's interesting that you don't think that's a software design question, because I think the plateau that I see is basically people using it as a Google replacement, and it's a phenomenal Google replacement. And then there's something in the middle, which is—and then there's coding, which is a whole separate thing. And there's something in the middle which is radically changing how you approach learning about a topic or building something or problem solving in a more substantive way.

我妻子是律师，她当时在向一所法学院推销她想教的一门课，一门她想教的兼职课程。所以她相当怀疑 AI，因为那些幻觉之类的问题。我费了点劲说服她没问题。她用 Claude 基本上把她在那个法律领域里非常感兴趣的各种案例都研究了一遍，然后想出了一个很酷的课程假设，以及她想讲的叙事。我说，你要把这个提交给法学院吗？她说，不行，我得先做一套课程和大纲才行。我就是不想做那个。我说，老兄，你所有材料都有了——直接在那边的框里打字，做个大纲出来。她说，哦，好。她就做了。大纲非常棒。所以这真的让我警觉，因为你会想，你知道，我不可能娶我们所有的用户。我们没法做这种一对一的培训。

My wife is a lawyer and she was pitching a law school about a class that she wanted to teach, like an adjunct class that she wanted to teach. So she was fairly skeptical of AI because of all the hallucination stuff. I carefully convinced her that it was okay. She used Claude to basically research all of the different cases in this area of law that she was really interested in, and she came up with a really cool hypothesis for a class and the narrative that she wanted to tell. I was like, are you going to submit this to the law school? And she was like, no, I have to build a curriculum and a syllabus to do it. I just don't want to do that. And I was like, dude, you have all the material—just literally type into that box right there, make a syllabus. And she was like, oh, sure. And she did. And the syllabus was fantastic. And so it's like that really alarmed me, because you're like, you know, I can't marry all of our users. We can't do this one-on-one sort of training.

而她只是——她就在那里，她就在终点线上，但就是没有——这种情况会一再发生，对吧？因为哪怕只是一个大纲，也只是众多步骤中的一步。

And she just—she was there, she was on the finish line, and it just didn't—and that's going to happen again and again, right? Because even just a syllabus is just one of many steps.

而你看待它是个软件设计挑战、失误、错误——我猜这跟这个问题有关：Claude 是个有用的教练吗？Claude 是个有用的同事吗？Claude 会不会不只是说，你知道，跨过这条终点线，还会说，嘿 Joel，你有没有想过用这种方式而不是那种方式来组织你的设计团队？我不知道，也许，你知道，根据我看到的情况——

And you see it as a software design challenge, mistake, error—like, I guess this is sort of related to the question of: is Claude a useful coach, is Claude a useful co-worker? Like, is Claude opining not just on, you know, just get over this final finish line, but also, hey Joel, have you thought about organizing your design team this way instead of this way? I don't know, maybe, you know, based on what I'm seeing—

迹象显示这样可能更高效，但你可以做你自己——

Signs show that this might be more productive, but you can do you—

这绝对是个奇怪的——我是说，就连你刚才语气的转变都暗示了——我觉得有一堆陷阱，对吧？比如你想象那个场景，我们应该、也许可以、而且很可能已经能够意识到，在对话的那个节点，你可以问另一个版本的 Claude：你觉得用户接下来想做什么？你可能会得到三个不同的答案。

It's definitely a weird—I mean, even just your tone change there suggested—I think there are a bunch of pitfalls, right? Like if you imagine that scenario, we should and maybe could and probably could already recognize that at that point in a conversation you could ask another version of Claude, what do you think the user wants to do next here? And you could come to three different answers.

人们不喜欢增长黑客。我觉得增长黑客非常违背我们作为一家公司的总体精神。所以我们不追求参与度最大化；我们追求解决人们的问题。而在主动提出帮人解决问题，和不停地说“再来一条消息、再来一条消息，我们觉得你还没做完”之间，有一条微妙的界线。我觉得那里有个危险的循环。所以我们在那个维度上很小心。

People don't like growth hacking. I think growth hacking is very against our general ethos as a company. So we are not looking to engagement-max; we're looking to solve people's problems. And there's a fine line between offering to solve somebody's problem in a proactive way and being like, one more message, one more message, we think you're not done. There's a dangerous loop there, I think. So we're careful on that dimension.

然后你的语气暗示了，人们并不总是想被告知下一步该做什么，或者他们喜欢这样——这里基本上有一个谁在开车的问题。你之前打了个开车的比方：你坐在方向盘后面，你在掌控，你觉得自己说了算，而另一个人就在你旁边。而 Claude 说“其实，我有一个更好的设计版本”的那一刻。

And then your tone suggested that people don't always want to be told what to do next, or they like it—there's an inversion of basically who's driving here. You were making a driving analogy earlier: you're behind a steering wheel, you're in control, you feel like you're in charge, and someone else is there with you. That moment where Claude is like, actually, I have a better version of this design.

我能给你看看吗？这就像以一种非常吓人的方式抢方向盘。

Can I show you? It's like it's taking the wheel in a really scary way.

或者它可能是一个特别棒的导航员，会说，嘿，那边拐角就有一家超棒的餐厅。

Or it could be a really fantastic navigator, somebody who says, hey, there's an amazing place to eat just right around the corner over there.

完全同意。

Totally.

我是说，我们可以去那儿。你握着方向盘。

I mean, we can go there. You're behind the wheel.

对，对，只是提个建议，是的。

Right, right, just a point, yeah.

所以我觉得这两者对我来说都像是软件设计上的挑战。我认为更宏观的问题——这远远超出我的薪资等级——是制度变革以及应对这类问题的方法。我觉得教育就是一个非常引人入胜的例子。如果 Claude 能够加速学习、制作、建造或发现过程中的某些环节，那它会被纳入课程体系、被推广并复制到许多机构中去吗？

So I think both of those feel like software design challenges to me. I think the broader question that's very much above my pay grade is like institutional changes and sort of like approaches to this. I think like education being a really fascinating one. And if Claude is able to sort of like accelerate certain parts of a learning process or a making or a building or a discovery process, like does that get embraced into a curriculum and sort of like brought along and replicate that for many institutions.

我想以一个关于这如何改变我们自身本性的问题来收尾，好吗？因为我们正在与这项技术共事。我们不断发现这些新能力，它某种程度上拓宽了我们日复一日施展自我的透镜。你大概比任何人都花更多时间在日常工作中与 Claude 对话。你也知道 Claude 的能力在未来数月乃至数年里会大幅提升。作为一个团队，或者甚至就你个人而言，你如何看待我们在塑造自身思维、自身能动性、自身与这项技术的关系方面所肩负的责任？因为我知道坐在这把椅子上，我不得不面对这些问题，比如你要为发生的一切负责，而当时我们做的很多工作只是为了创造出对尽可能多的人有用的东西，只是为了活下去，只是你知道，建立一门能撑一阵子的生意。但我越来越认识到，软件设计师的工作与建筑师并无二致，你要为你所创造的居所、你所创造的技术空间里的居住者负责。你如何看待我们与这项技术的关系，以及它对我们自身人性所做的事？

I want to close on a question around how this changes our own nature, right? Because we are working with this technology. We're finding all these new capabilities and it's sort of expanding the lens through which we can kind of apply ourselves day in and day out. You probably spend as much time as anyone else talking to Claude as part of your day-to-day work. And you also know that Claude will dramatically improve in capabilities in the months and years to come. How do you sort of think about as a team or even just as you around the responsibilities regarding like how we kind of shape our own thinking, our own agency, our own relationship to this technology? Because I know sitting in this seat I've had to confront those questions around like you're responsible for all this stuff that happened and in large part like at the time a lot of the work that we did was just to create something that's useful to as many people as possible and just survive, just you know build a business that sticks around for a little while. But I'm recognizing more and more that the work of a software designer is not unlike that of an architect, that you are responsible for the people that occupy the dwelling, that occupy the technological space that you create. How are you sort of thinking about the relationship that we have to this technology and what it does to our own human nature?

是的，我感到非常幸运，这是 Anthropic 核心研究团队深切关心的事情。我认为在这个问题上，我其实并不觉得 UI 和 UX 的选择有模型行为、模型说什么以及它如何呈现和反应那么重要。在几乎任何未来的产品形态中，都有一个智能实体在那里决定何时推动、何时建议、何时退后，对吧？我们做了大量研究，而且很多——负责这个团队的 Amanda 是科班出身的哲学家，这真的变成了一个道德哲学、伦理问题，基本上就是：像这样的工具对人负有什么责任？所以我认为在组织层面，我们深度投入于弄清楚这一点。

Yeah, I feel very blessed that this is something that the core research team at Anthropic cares deeply about. And I think this is one where I actually don't think the UI and UX choices matter nearly as much as the model behaviors and what the model says and how it shows up and reacts. In almost any future of the product instantiation, there is an intelligent entity there that is deciding when to push, when to advise, when to lean back, right? We do a lot of research and a lot of like—Amanda who runs this team is a trained philosopher and it really becomes a moral philosophy, ethics question of basically like what responsibility does a tool like this have to people? So I think organizationally we are deeply invested in figuring that out.

我个人觉得它就像魔法一般，像一个全天候 24/7 陪伴我的治疗师兼挚友，而且我不用付钱——即便我为心理治疗付费，当我带着问题去找治疗师时，我仍然觉得是在求人帮忙。就像我和妻子之间的某种社会契约：你可以在晚餐时抱怨工作一半时间，然后你的伴侣抱怨工作另一半时间。而当你依靠他人来反思、成长、自我反省之类的，总有一种规范，而这些模型就是没有。所以我用那么多低优先级的琐事烦过 Claude，那些我绝不会去烦别人的事——而它总是那么不遗余力地提供帮助和反思。所以我认为，如果你能把研究团队正在构建的核心道德与伦理品格，与我们的工作——设计的工作就是让它感觉安全、没问题、易用，以及所有那些无聊的东西，比如 iPhone 应用应该加载得飞快，因为有时我想在生活的碎片时间里做这些事，真正提高可及性——结合起来，我认为这是一个极其积极的改变。

I have personally found it to be like magical and like a therapist slash best friend who's with me 24/7 and I don't have to pay—like I don't, even though I pay for therapy I still feel like I'm asking for a favor when I bring my problems to my therapist. Like there's just there's some part of the social contract with my wife. Like you get to complain about work for like half of dinner and then like your partner gets to complain about work for half of dinner. And like there is just a there's a norm when you lean on other humans for reflection, growth, self-incrimination, whatever, that these models just like don't have. And so I have like I have bored Claude with so many like low priority things that I would never—and it's so relentlessly helpful and reflective. And so I think if you can combine basically like the core moral and ethical character that the research team is building with basically like our job—design's job is like make it feel safe and okay and accessible and like all the boring stuff like the iPhone app should load really fast because like I want to sometimes do this like in little slivers of my life, like really increase the accessibility. I think it is like a tremendously positive change.

而作为一个父亲，当你审视自己在这其中的位置，审视你如何向孩子们谈论这项工作，这个角色对日常生活有什么影响？比如，对于你目前所做工作带来的经济机会和意义机会，你抱有怎样的希望？

And as a father, when you look at your place in all this and you look at how you talk about this work to your kids, what does that role kind of do to the day-to-day? Like what are you sort of hopeful for in terms of the economic opportunities and sort of the opportunities for meaning that arise from the work that you're doing right now?

我在这方面偷师了 Mike Krieger 的一招，就是那种想直接问 Claude 所有答案的诱惑——我有一个七岁、一个四岁和一个两岁的孩子。他们问的全是我不知道答案的问题，尴尬得很，就像古往今来所有爸爸一样。现在他们知道 Claude 怎么用了。现在他们知道有语音模式了。他们就说：“爸爸，别挡道，直接问 Claude 吧。”

I stole a page from Mike Krieger's playbook on this where the temptation to just ask Claude for all the answers for every—I have a seven-year-old, a four-year-old, and a 2-year-old. Like, they just ask nothing but questions that I don't know the answer to, embarrassingly, like all dads through all time. And now they know how Claude works. Now they know that there's voice mode. And they're just like, "Dad, just get out of the way. Just ask Claude."

对，对，对，对。

Right, right, right, right.

Mike 说的是，他的孩子年龄也差不多。他说：“我们有个规矩，在问 Claude 之前，我们先猜答案是什么，并解释为什么这么想，还有哪些可能的答案，我们寻找的不只是那一句话的解释。”所以我真的开始尝试这么做，因为我完全不知道孩子们该选什么大学专业，或者将来会有什么工作。但我确实认为好奇心和探索至关重要，而他们年幼时就天生具备。

And what Mike said is that he has kids roughly the same age. He was like, "We have a rule where before we ask Claude, we guess at what the answer is, and we explain why we thought that, what the alternative answers might be, and we look for more than just the single one-sentence explanation." So I've really tried to start doing that because I have no earthly idea what the right college major is for the kids or what jobs will exist. But I do think that curiosity and exploration is critical and they're born with it at the young age.

是的。

Yeah.

至少从我的角度来看，我发现你们团队的工作真正神奇之处在于，它实实在在地拓宽了我个人抱负的范围，以及我想在世界上做成的事情。而其中很大一部分此前主要受限于——你需要这样一个不知疲倦的共谋者，你可以问所有愚蠢的问题，而它永远不会疲倦。我自己一直在琢磨的一个问题是：好吧，如果我们把所有这些负担都从他人身上卸下，当我们不像 Claude 那样不知疲倦、那样阳光乐观时，我们会不会对彼此感到沮丧？还是说，它可能会彻底改变我们彼此相处的方式，让我们认识到，其实我们可以谈论那些更棘手、更麻烦的事，而把技术上的壮举留给与 Claude 的对话。

And at least speaking from my perspective, the thing that I've found to be truly magical about the work that your team has done is that it has legitimately broadened the scope of my own personal ambitions and the things that I want to be able to do in the world. And a lot of that was largely gated by this like tireless co-conspirator that you can just ask all the stupid questions to and that just like you can just never fatigue. And I guess a question that I am kind of ruminating on myself is just okay, if we sort of take all of that burden away from other people, will we get frustrated with each other if we aren't as tireless and as sort of like sunny and optimistic as Claude tends to be? Or might it just completely change the way in which we kind of relate to one another and recognize that like actually we can sort of talk about the thornier, hairier and leave like the technological exploits to our conversations with Claude.

太棒了。

Awesome.

那么，我代表在座的各位以及所有听到这场对话的人，谢谢你，Joel，感谢你所做的一切，让你的生活变得好得多。

Well, on behalf of everyone here and everyone hearing this conversation, thank you, Joel, for everything you've done to make your life way better.

谢谢大家。

Thanks, everyone.