设计工作流的未来：掌握 AI 作为协作伙伴

欢迎来到 Dive Club。我是 Red，这里是设计师永不停歇学习的地方。本周的嘉宾是 Pietro Schirano，他是新设计工具 Magic Path 的联合创始人，也是 AI 领域真正的远见者之一。我们将深入探讨设计工作流的未来，以及掌握模型所需的条件。话不多说，让我们开始吧。

Welcome to Dive Club. My name is Red and this is where designers never stop learning. This week's episode is with Pietro Schirano who's the co-founder of the new design tool Magic Path and one of the true forward thinkers when it comes to AI. So, we're going to do a deep dive into the future of design workflows and what it takes to master the model. So, without further ado, let's dive in.

我在科技行业做设计师已经很久了。我从意大利搬到美国，在 OpenTable 开始了职业生涯，是那里的首批产品设计师之一，打造了广为人知且备受喜爱的 OpenTable。然后我去了 Facebook、Uber 和 Brex。在 Brex，我首次在 AI 方面取得了突破。但回想起来，更重要的是，真正促使我构建像 Magic Path 这样的东西的，是我在 Facebook 的经历。我们非常重视原型设计，Facebook 有一个很好的理念：在用户真正使用之前，一切都还不算数。当时我们使用一个叫 Origami 的工具。我 Origami 用得特别好，Facebook 还让我给设计师们上课，教他们用 Origami 来原型化交互和应用。

I've been a designer in tech for what feels like a really long time. I moved from Italy to the US, started my career at OpenTable, was one of the first product designers there, built the OpenTable like known and loved. Then I moved to Facebook, Uber, and Brex. And at Brex, I sort of had my first break with AI. But I think more importantly, looking back, what really took the stage for me to build something like this was when I was at Facebook we made a big deal about prototyping and Facebook actually had this really good philosophy around nothing is real until it's in people's hands. At the time we used to use this tool called Origami. Actually I was so good at Origami that Facebook made me do classes for designers to learn how to use Origami so they could prototype their interactions and applications.

顺便说一句，这简直是设计师的终极信号。

That's like the ultimate designer signal by the way.

是的。因为对我来说，我不喜欢设计师停留在静态的世界里，对吧？你需要感受一下效果，确保体验是正确的。所以在 Brex 的时候，我们拿到了 GPT-3 API 的访问权限，我们可能是最早的公司之一。那是在 ChatGPT 热潮和 LLM 狂热之前。我用这个 API 做了一个 Brex CFO 界面的演示。我做了像用 SQL 输出图表之类的东西。这些都是我尝试的，而且成功了，简直令人震惊。所以我们推出了 Brex AI，相当成功。然后我想，好吧，我对 AI 这东西已经有点心得了。实际上，从早期开始，我最大的经验——四年前——就是如何用 AI 制作漂亮的网站。当 GPT 市场开放时，我推出了一个叫 Designer GPT 的 GPT，它至今仍是平台上最受欢迎的 GPT 之一，有大约 500 万次对话，简直疯狂，而它只是一个用来制作网站落地页的 GPT。

Yeah. Because for me it's like, I don't like when designers leave this like a static world, right? You need to see how things feel to make sure the experience is correct. And so when I was at Brex, basically we got access to the GPT-3 API and we were probably one of the first companies. This is before the ChatGPT craziness, the LLM frenzy. And I built a demo of what a CFO at Brex could look like using this API. And I made things like returning graphs using SQL as an output. Those are all things I just tried and it worked. It was kind of mind-blowing. So we launched Brex AI, it was pretty successful, and then I was like, you know what, I think I have a pretty good hint how this AI thing works. And actually, since the early days, my biggest experience up to four years ago was actually about how can we make beautiful websites using AI. And when the GPT marketplace opened, I launched this GPT called Designer GPT, which to this day is one of the most popular GPTs on the platform. It has like 5 million chats or something insane, and that's just a GPT to make website landing pages.

随着我对 AI 作为协作者这一理念越来越得心应手，我注意到，我在 Anthropic 工作了一段时间，做不同的产品，然后我心想：这是怎么回事？为什么我都没打开 Figma？为什么一年来，我这个设计师几乎没碰过这个我用了一辈子的工具？所以我意识到，我不再去 Figma 文件，而是直接开始用 AI 做原型。可能是用 Cursor，或者我自己写的一些脚本。我做了个叫 Cloud Engineer 的东西，它可以说是最早的智能体式终端编码智能体，远早于后来那些智能体。它实际上是 Claude Code 的灵感来源，这很酷。这也是我去 Anthropic 工作的原因。最终我意识到，这里发生了什么事？我在一月底发了一条推文：再也没有理由用 Figma 了。

And so as I was getting really comfortable with this idea of AI used as a collaborator, what I noticed is that, you know, I was working at Anthropic for a while, working on different products, and I was like, what's going on here? Why haven't I opened Figma? Why for a year me, a designer, haven't opened this tool that I've used my entire life almost, right? And so what happened was I realized, instead of going to a Figma file, I would just start AI prototyping. That might be Cursor or something I built on my own, like some sort of script. I built this thing called Cloud Engineer, which was kind of like the very first agentic terminal coding agent, way before all those agent things. It was actually the inspiration for Claude Code, which is really cool. That's why I went to work at Anthropic. Eventually, what I realized was, you know, what's going on here? And I tweeted at the end of January: there is no reason to use Figma anymore.

我们都记得。

We remember.

那条推文疯传了。我是说，大约有 500 万人互动，简直疯狂。但后来很多人开始攻击我，因为他们把这看作是对设计领域的攻击，而不是说，“嘿，看，现在有更好的做事方式，你不这么做就错过了。”就连 Figma 的人，比如副总裁，也开始在评论里攻击我。我想，“这是怎么回事？”而不是，“嘿，也许这家伙有点东西。他确实展示了这个 AI 工具的一些力量。我们为什么不听听他说什么？”所以那时我想，“好吧，我就做给你看。”基本上，出于人们认为可能的事和我认为可能的事之间的这种反差，我创立了 Magic Path。我们上线还不到两个月，现在是第七周。我们有 5 万活跃用户，每周都在疯狂增长，设计师们喜欢它，因为我们真的设计出了一种方法，让这些模型基本上能成为出色的中级到高级设计师。

It went so viral. I mean, I'm talking about like 5 million people reacting, which is insane. But then a lot of people started attacking me because they saw that as an attack on the design field, versus saying, 'Hey, look, there's a better way of doing things today and you're missing out if you don't do that.' Even Figma people, like the VP, started attacking me in the comments. I was like, 'What's going on here?' Instead of, 'Hey, maybe this guy has something to say. He definitely has shown there's some power with this AI tool. Why don't we hear what he has to say?' And so at that point, I was like, 'You know what? I'm just going to show you.' And basically out of this sort of contrast between what people believed was possible and what I believed was possible, I founded Magic Path. We've been out for less than two months. So it's our week seven. We have 50,000 active users. We grow like crazy every week and designers love it because we really engineered a way to make those models kind of like amazing mid-level to senior-level designers basically.

快速插播一条消息，然后我们继续。好的，Framer 刚刚发布了两个杀手级 AI 功能，我必须告诉你。第一个是 Wireframer。你可以与 AI 聊天，跳过空白画布，激发各种想法。没有比这更简单的方法来快速生成一个带有结构和可编辑起始内容的响应式页面了。第二个功能是 Workshop。它可以用自然语言创建高级组件。所以你可以快速生成各种新的视觉效果，无需编码。这只是我向所有人推荐 Framer 的又一个理由。你现在就可以开始使用这些功能，只需访问 dive.com/framer。我设计已经很久了，但不知怎么的，今年才注册 Mobbin。老实说，我不敢相信花了这么长时间。它简直就是一个作弊码。能够为几乎任何模式、任何产品、任何流程找到灵感。它很快，很漂亮。我已经无法想象没有 Mobbin 的设计了。它是我寻找灵感的第一个地方，我强烈推荐。所以请访问 dive.comclub/mobbin 开始使用。拼写是 m o b b i n。好了，回到节目。

Real quick message and then we can jump back into it. All right, Framer just released two killer AI features that I have to tell you about. The first is Wireframer. You can chat with AI to skip the blank canvas and spark all kinds of ideas. And there's no easier way to spin up a responsive page with structure and starter content that's ready to edit. Now, the second feature is Workshop. It's a way to create advanced components using natural language. So, you can spin up all types of new visual effects. No coding required. It's just another reason why I recommend Framer to absolutely everyone. And you can start using these features today. Just head to dive.com/framer. I've been designing for a long time. But somehow I only signed up for Mobbin this year. And honestly, I can't believe that it's taken me this long. Like it's literally a cheat code. Being able to find inspiration for basically any pattern, any product, any flow. It's fast. It's beautiful. And I already cannot imagine designing without Mobbin. It's the first place that I look at for inspiration and I can't recommend it enough. So head to dive.comclub/mobbin to get started today. That's m o b b i n. Okay. Now on to the episode.

如果你是一名设计师，你的角色是从事 AI 工具、AI 应用或任何 AI 相关的工作，那么你的角色已经发生了根本性的变化。你在摆弄这些模型的过程中学到了什么，你认为设计师可以从中受益？

If you're a designer that the role is to work into AI tools, AI application, any sort of AI thing, your role has fundamentally changed. What have you learned about tinkering with these models that you think designers could benefit from?

这不仅仅是关于视觉，也不仅仅是关于交互。它实际上是关于想法、合作伙伴和各种不同的事物，把 AI 当作一个头脑风暴伙伴，而不是仅仅“让这个落地页变漂亮”。在公司内部，不分享原型几乎会显得不好看，你知道，我们正在进入这样一个世界：如果你分享一个静态的 mockup，别人会觉得“呃，我们这是在干什么？”

It's not just about the visual. It's not just about the interaction. It's really about the ideas and the partners and the different things and using this as a brainstorming partner versus, you know, just a 'make this landing page pretty.' It would almost be a bad look internally to not share a prototype, you know, like we're entering that kind of a world where if you share a static mock, it's like 'uh, what are we doing here?'

如果你能把它作为工作流的一部分用在设计中，你将势不可挡。

If you can just like take that and use that as part of your workflow in a comp, you are unstoppable.

那么，当你审视设计工具领域的格局、不同的工作流程以及这些工作流程之间的差异时，你看到了什么？你对当今世界的状况以及幕后正在发生的某些惯性有什么看法？

So when you look at the landscape for design tooling and the different workflows and the disparity between what those workflows might be, what do you see? What's your perspective on the state of the world today and some of the inertia that's happening behind the scenes?

完全同意。基本上，我们在 Magic Path 上做的一件有趣的事情是，与市场上许多工具相比，我们退后了一步。比如当你看到 Lovable、Bolt、v0 这些工具时，它们都在向你承诺一种“文本到应用”的理念，像是“你写点东西，就能得到一个应用”。但不幸的是，如你所知，我是这个星球上最看好 AI 的人，但我认为这个现实其实并不接近。可能还需要几年才能实现，因为直到今天，你仍然需要工程师来把你的产品落地，比如身份验证、后端等等这些东西。所以我们基本上是在向人们兜售一个谎言。我们做了不同的事情。我们说：“等等，这不是要取代角色。我们不是说你要用一个工具来摆脱你的产品经理、设计师、工程师。实际上是要增强现有的工作流程。”所以我的座右铭一直是：下一个 Instagram 可能不是建立在 Lovable 上，但可能建立在 Magic Path 上，因为我们正在构建由 AI 驱动的工作流程和终端来实现这一点。这就是为什么我认为我们能够从嘈杂中脱颖而出，并在不同类型的工具中引起共鸣。因为当你来到 Magic Path，它基本上是一个无限画布，一个创意空间。你实际上拥有专业工具所期望的一切，同时也拥有使用 AI 作为入门套件或用于编辑等所有优势。所以我认为我们将开始看到越来越多的这类工具变得更加专业化。我们确实是第一个设计智能体工具，对吧？基本上 Magic Path 是一个设计智能体。Mascari 是一个你可以使用的工具，而其他所有工具都采取了不同的方法。如果你那样做，结果就是你没有专注于设计师需要的东西，对吧？因为 Lovable 并不专注于一次性做出最美的东西，也不专注于如何构建流程，也不专注于成为设计师意味着什么、构建产品意味着什么。而我们专注于这些东西。所以，尽管市场混乱且充斥，我认为我们将开始看到越来越多这种专业化的工具。我认为 Magic Path 在某种程度上代表了设计领域的第一个此类工具。

Totally. So basically, the interesting thing that we did with Magic Path is that we took a step back compared to a lot of those tools you see in the market. Like when you see things like Lovable, Bolt, v0, right? They're kind of promising you this idea of text-to-app. It's like, "Oh, you write this something and you're gonna get an app." And unfortunately, as you know, I'm the most AI-positive person on the planet. I don't think that reality is actually close. It might actually take a few years to get there because to this day, you still need an engineer to bring your product home, and that might be things like authentication, your backend, all this kind of stuff, right? So we're basically selling a lie to people. We did a different thing. We say, "Wait a second. It's not about replacing the role. We're not saying you're going to use a tool to get rid of your PMs, your designers, your engineers. It's actually about enhancing the existing workflow." And so my motto has been: the next Instagram might not be built on Lovable, but it might be built on Magic Path because we're building the workflows and ends by AI to allow that. And that's why I think we were able to sort of make it through the noise and resonate across different kinds of tools. Because when you come to Magic Path, it's basically an infinite canvas, a creative space. You actually have all the things that you would expect from a pro tool, but you also have all the advantages of using AI as a starter kit or for editing, and so on. And so I think what we're going to start seeing more and more is that those tools will get more specialized. We are truly the first design agent tool, right? Basically Magic Path is a design agent. Mascari does a tool that you can use, and all the others are taking a different approach. If you do that, what happens is that you're not focusing on the things that a designer needs, right? Because Lovable is not focusing on making the most beautiful thing in one shot, or it's not focused on how to build a flow, or it's not focused on what it means to be a designer, what it means to build a product. And we are focusing on this stuff. So as convoluted and polluted as the market is, I think we're going to start seeing more and more of this kind of specialized tool. And I think Magic Path sort of represents the first one of this kind for design.

听到你如此刻意地谈论“工作流程”这个词，很有意思。你认为设计的未来实践和工作流程正在如何演变，这又如何影响你对 Magic Path 在技术栈中位置的思考，以及最终的产品策略？

It's interesting to hear you talk about the word workflow so intentionally. What do you believe about the way that the future practice of design and that workflow is evolving that's then influencing how you think about where Magic Path fits into the stack and ultimately what the product strategy is?

完全同意。我们经常被问到的一个问题是：“你们打算如何取代 Figma？”这是一个经常被问到的问题，对吧？我的回答是：这不是要复制并一对一地做 Figma 所做的事情。而是要理解人们今天在 Figma 中做什么。从工作角度来看，它们代表了一些重要的东西，你可以增强它们或将它们转化为 AI 时代的东西，对吧？所以当你看到我们在做什么时，我们真正专注于界面设计。为什么？因为我对进入品牌、插画或绘画领域没有兴趣，这些都是你今天可以在 Figma 中做的事情，对吧？所以当你思考工作流程时，举个例子，我们最近推出的一项对我们非常有利的功能——它带来了订阅量的大幅增长——就是设计系统，对吧？当你看看今天的 AI，AI 就像你永远不知道会得到什么。它只是一个黑箱。你输入一些东西，得到一些东西。我们实际上能够通过我们的设计系统功能来约束这种生成。所以很多公司都喜欢 Magic Path，因为他们来到这个平台，定义设计系统，然后他们就可以生成看起来像是他们团队制作的东西。这是一件大事，对吧？因为当你回顾所有那些工具，比如之前的 Sketch 和 Photoshop，以及之后的 Figma，它们只是一个事实来源。归根结底，Figma 并没有做什么其他工具做不到的特殊事情。它们只是一个非常好的事实来源。所以问题是：你如何将这种成为良好事实来源的信念，并使其以 AI 为先？这就是我们正在做的。

Totally. So, you know, a question that we get a lot is like, "How are you guys going to replace Figma?" That's a question you get a lot, right? And my answer to that is: this is not about copying and doing what Figma does one-to-one. It's about understanding what are the things that people are doing today in Figma. They represent something important from a work perspective that you can enhance or translate into the era of AI, right? And so when you look at what we're doing, we're really focused on interface design. Why? Because I have no interest in going into brand or illustration or drawing, which are all things you can do in Figma today, right? And so when you think about workflow, for instance, one of the things that we launched recently that was really good for us—it created a huge uptick in subscription—was design systems, right? So when you look at AI today, AI is just like you never know what you're going to get. It's just a black box. You put something in, you get something out. We are actually able to contain that generation via our feature which is design system. And so a lot of companies are loving Magic Path because they come to the platform, define design systems, and then they can just generate things that look like they were made by their team. And that's a huge thing, right? Because when you look back at all those tools like Sketch and Photoshop before, and Figma after, they're just a source of truth. At the end of the day, there's nothing that Figma is doing special that no other tool can do. They're just a really great source of truth. And so the question is: how do you take that belief of being a good source of truth and make it AI-first? And that's what we're doing.

成为事实来源到底有多重要？这是我最近一直在问自己的一个问题，因为我正在开发一个非常依赖 AI 的新产品。所以我正在以更偏向工程的方式构建东西。顺便说一下，我是终极的事实来源型设计师。我的 Figma 文件非常整洁。我有所有的库、图层，一切都完美。但我最近有点放松了对这方面的掌控。

How much does it even matter to be the source of truth? And that's a question I've been asking myself a lot recently as someone who's working on a new product that's very AI heavy. So I'm building things in ways that are a little bit more engineering. And for context, I am the ultimate source of truth designer. I mean my Figma files are pristine. I have all of my libraries, my layers, like everything's perfect. And I've kind of been relaxing my grip on that a little bit recently.

那么你怎么看这个问题？设计工具是否仍应作为事实来源，还是说它纯粹是我们要抽象和优化的正确工作流？

So how do you think about that? Like should a design tool still try to be the source of truth or is it purely what are the right workflows that we should abstract and optimize for?

有趣的是，在这个时代，成为事实来源反而更加重要。我的意思是，回顾过去，设计师进入 Figma 把规格做到像素级完美，但随后总会出现翻译偏差，工程师去实现时会有出入。我无法告诉你我在 Facebook 花了多少个夜晚，把所有边距和尺寸都标注好，然后看到产品上线时却偏了五个像素。我当时想，这怎么会发生？那是因为人性使然，我们会犯错。但有了 Magic Path，长期愿景是这成为设计与工程之间的完美交接。如果你拥有事实来源，未来的设想是，在不远的将来，设计师可以进入 Magic Path，更新某些内容，然后这些更新会直接推送到生产环境。所以从某种意义上说，你甚至不用看代码就能完成工程工作。这让我非常兴奋，因为它回到了我在 Facebook 的日子，那时我把 Origami 放在手机上，制作交互原型，感受动画、手机的振动和触觉反馈。现在我们能做到这一点了。我觉得这很美。如果我能回到 10 年前的 Pietro 身边，对他说，嘿，你将创造一个能让你所有梦想成真的工具，我会欣喜若狂。我们生活在这个时代真是太酷了。

The funny thing is that in this era it's actually even more important to be the source of truth. What I mean by that is, if you look at the past, designers go into Figma to make things spec perfectly down to the pixel, but then there's always been this loss in translation where the engineer will go and implement that. I could not tell you the amount of time I would spend at Facebook nights, speccing out all the margins and sizes, and then I would go see the product shipped and it would be five pixels to the left. I was like, how does this even happen? That's because it's human nature, we make mistakes. But with Magic Path, the long-term vision is that this is the perfect handoff between design and engineers. If you have your source of truth, the idea is that in the future, not so far from here, a designer can go into Magic Path, update something, and as they update, that gets pushed into production. So in a way, you are doing engineering work without even looking at the code. To me, that's very exciting because it goes back to those Facebook days where I would put Origami on my phone, prototype the interaction, feel the animations, the vibration of the phone, the haptic feedback. Now we can do it. I think it's beautiful. If I could go back to Pietro 10 years ago and say, hey, you're going to make a tool that's going to allow all your dreams come true, I would be ecstatic. It's so cool that we live in this era.

这真的很酷。确实如此，因为我现在已经接受了这样一个事实：在 Cursor 里工作对我很有价值，在代码中探索对我很有价值，贴近输出对我很有价值，尤其是在使用 AI 的时候。但在这个领域探索对我来说还是有点慢。所以我仍然在 Figma 里探索。但当我到了某个点，觉得这个有意思时，我基本上要从头开始在 Lovable 或 Cursor 里做，虽然有 MCP 之类的，但体验并不好。有很多手动工作，感觉很慢，没有创意感。但这是做出本质上功能性的东西的代价。而你直接跳到了功能性，同时又让我在前端层探索后能快速连接起来，无需重复劳动。这个价值主张对我来说很有意义。

It's really cool. It really is because I'm someone who's now really come to terms with the fact that it's valuable for me to be in Cursor. It's valuable for me to play in code. It's valuable for me to get really close to the outputs, especially when you're working with AI. And yet it's still a little bit slow for me to explore in that arena. And so I still explore in Figma. But then when I get to the point where I'm like, okay, this is interesting. Now I'm basically starting from scratch either in a Lovable or Cursor and yeah there's MCP and everything but it's not amazing. There's a lot of manual work, it feels slow, it doesn't feel creative. But that's the price of making something that is inherently functional. Whereas you're jumping right to functionality but then also making it so that okay I've explored just at the front end layer but now I can hook it up quickly without having to do redundant work. That value proposition makes a lot of sense to me.

对吧？因为我们想成为你 GenAI 体验的起点。你来到这里，只有当你准备好了，只有当你定义了端到端流程并对结果满意，对你的编辑、手动编辑等一切满意时，我们下周还会添加图层视图，这让我非常兴奋。我的意思是，我们要把它变成真正任何人都能使用的工具。而这正是 Magic Path 的美妙之处。你 10 年前构建的东西，非常干净漂亮，但你可能失去了实现，因为你的团队可能不在了，你失去了与团队的联系。无论情况如何，把那个截图放进去，让它工作。这太棒了。

Right? Because we want to be the starting journey of your GenAI experience. You come here, and only once you're ready, only once you define your end-to-end and you're happy with your result, you're happy with your edits, your manual edits, whatever you're doing, and we're going to add layer view next week, which I'm very excited. I mean, we're going to turn this into truly a tool that anybody can use. And that's actually the beauty of Magic Path. You built something 10 years ago, it's really clean and beautiful, but you lost the implementation because your team might not be around anymore, you lost the connection with the team. Whatever the case may be, put that screenshot there, make it work. It's awesome.

好的，那么放大来看，考虑到你已有的成果以及你对未来的愿景，你认为这些新功能、新技术以及你带来的所有可能性，将如何影响设计师作为更大软件团队中一员的运作方式，以及团队协作的方式？

Okay, so zooming out, given what you have but also given your vision for where you want to take this, how do you think this new functionality, technology, all these possibilities that you're bringing to the table, how do you think that's going to impact the way that designers function as a unit of a larger software team and the way that teams collaborate?

我给任何进入设计领域、即将成为设计师的人，甚至是已经做了一段时间的人的最大建议是：你需要学习 AI。原因在于，正如你所见，AI 已经在这里了。它非常真实。人们正在用 AI 构建东西。他们可以比你过去快 10 倍地行动。这非常重要，因为你的 Figma 技能或其他技能可能无法很好地转化，除非你学会如何使用这些技能。社区里有些人确实学会了如何使用这些技能。因为归根结底，最重要的是想法和品味。而像 Magic Path 这样的应用，它能做的就是缩小你的想法、构思与实际尝试之间的差距。更进一步，我认为设计师的角色正在改变。它将彻底改变。我认为设计师的角色将变得更加重要，因为他们基本上会成为体验的守护者，也就是你的应用或产品应该实现的目标。从某种意义上说，他们真的会融合进来，甚至不再是设计工程师了。它几乎可以称为我所说的设计架构师，你实际上是在与一个能为你做事的智能体合作，遵循你的品味、你的技能、你的可用性，它让你能够比过去快得多地为公司进行原型设计和构思。背景是，在创办 Magic Path 之前，我在 Anthropic 工作了 10 个月。我是作为研发人员被引入的，负责为那些模型设计体验，因为模型非常好，但体验仍然落后。我在公司内部极力推动的一件事，后来几乎自成体系，那就是我极力主张先做原型。我极力主张产品周期因为 AI 而根本性地改变，我们需要适应。这有点好笑，因为你期望一家 AI 公司已经弄清楚了这一点，但没有人弄清楚，因为所有这些东西都是新的，对我们和对他们来说都一样是未知的，这真的很酷。因此，你成为了那个既负责体验，又展示原型、展示端到端、展示演示的人，真正了解想法的本质，这样你就能让你的团队比以前更快地达成一致。过去你有一个 Figma 文件，你和设计团队开第一次会议，共享屏幕。

My biggest recommendation for anybody that is either entering the world of design, approaching to be a designer, but even people that have been doing this for a while is that you need to learn AI. The reason is because as you can see, this is here already. It's very real. People are building stuff with AI. They can move 10x faster than you could move in the past. And it's very important because your Figma skills or other skills that you have might not translate well unless you learn how to use those skills. Some people in the community did learn how to use those skills. Because at the end of the day, what matters the most is ideas and taste. And an app like Magic Path, what it allows you to do is just reduce the delta between your idea, ideation, and actually getting to try something. And going even farther than that, I think the figure of the designer is changing. It will change completely. I think the designer's role is going to be even more important because they're going to be basically the keeper of the experience, the things that your app or product is supposed to accomplish. And in a way, they're really going to merge into this, it's not even like the design engineer anymore. It's almost like what I would call the design architect, where you are actually working with an intelligence that can do things for you, following your taste, following your skills, following your availability, and it allows you to have this prototyping, this idea for your company much faster than you would do in the past. For context, before I started Magic Path, I was working at Anthropic and I worked there for 10 months. I was brought in as the guy in R&D to figure out experiences for those models because the models are really good but experiences are still lacking behind. And the one thing that I really pushed inside of the company, that then kind of took a life of its own, was that I was really pushing for prototyping first. I was really pushing that the product cycle is fundamentally changing because of AI and we need to adapt. It's kind of funny because you expect an AI company to have that figured out, but nobody has figured it out because all those things are new and as reason to us as they are to them, which is really cool. So because of that, you now become this person that is in charge of the experience but also shows the prototype, shows the end-to-end, shows the demo, really gets a sense of what ideas are so that you can get your team aligned way before, way faster than you could do before. In the past you had this Figma file, you have the first meeting with your design team, share the screen.

好，这个屏幕要放到那里，这个团队要做这个。也许有人会做个原型，但原型其实不怎么样，很难分享，在手机上体验也不好，诸如此类，对吧？所以设计师你才是掌控者。其实挺有意思的，人们害怕 AI，我就想，怕什么？用 Magic Path 把自己提升 10 倍啊。你明白我的意思吗？既然你让这件事看起来这么简单，在公司内部不分享原型反而说不过去了。我们正在进入这样一个世界：如果你只分享一个静态稿，那感觉就像——

Okay, this screen is going to go there. This team's going to do this. Maybe some people will do a prototype, but the prototype is really not that great. It's tough to share. It's tough to feel it on the phone. Like all those kind of things, right? And so you a designer are in control. And actually, it's funny. People are afraid of AI. I'm like, what? Use Magic Path and 10x yourself. You know what I mean? Given how easy you just made it look, it would almost be a bad look internally to not share a prototype. You know, like we're entering that kind of a world where if you share a static mock, it's like

我们这是在干什么？

what are we doing here?

这其实是我想强调的另一个重点。我最近跟很多公司交流，也跟在这个领域认识 15 年的朋友聊天，发现最成功的设计师都是采用 AI-first 方法的人。这也是我在创办 Magipath 之前一年的做法，也正是这个推动我做了这个产品。因为就像你说的，你可以用不同的工具分别做这些事，但如果有一个统一的地方来处理所有事情——变化、创意——那就太好了。你刚才提到了“对齐”这个词，这几个月来这个词一直在我脑海里打转：这一切到底走向何方？当你可以一键看到完整的流程和所有变体时，未来做设计师意味着什么？你几乎每天都能拿出无限多的产物。做设计师到底意味着什么？我不断回到“对齐”这个词，因为到那时你不仅拥有更大的一块拼图——从一个想法的萌芽一直到可能整个前端的实现——

That's actually the other thing that I think it's important touching here. The most successful designers that I've seen recently talking to companies and friends that I know in the space for like 15 years are the ones that are adopting this AI-first approach. And that was like my approach a year before I started Magipath. And that's what pushed me to make this product because I was like, to your point, you can do this type of stuff separately in different tools, but it will be nice to have a unified place to do all the stuff: the variation, the creativity. You mentioned the word alignment earlier which has kind of been the key word that's been bouncing around my brain the last few months where it's like where does all this head? What does it mean to be a designer in the future when you can literally just hit the button and see full flows and all of these variations? You almost have an infinite amount of artifacts that you can bring to the table on a daily basis. What does it mean to be a designer? And I keep coming back to that word alignment because it's like you're the one then that is not only owning a larger part of the puzzle where it's all the way from the nugget of an idea to potentially the entire front-end implementation.

对。

Yeah.

而且你也是那个提出建议、掌舵方向、帮助别人理解你到底在做什么的人：你在拉动哪些杠杆？这些决策背后的推荐方向是什么？想想看，这是一个令人兴奋的世界。这没什么好怕的。这是很大的影响力，纯粹的影响力。

But you're also the one who's making the recommendations and trying to steer the ship and helping people understand what are you actually playing with here? What levers are you pulling that are behind these decisions? What is the recommended direction? And that's kind of an exciting world to think about. That's not something to be scared of. That's a lot of impact, you know, that's pure impact.

纯粹的影响力。我最喜欢的一句话是：我最好的想法就是拥有很多想法。我深信这一点。而这基本上就是它能让你做到的：尝试很多想法。因为想法的问题在于，除非你真正去测试，否则很难知道效果如何，对吧？实际上，我再给你讲一个更有趣的轶事。现在可以说了，因为已经公开了：我在 Anthropic 做深度研究时，尝试了各种不同的功能。这远在我开始考虑 Magikipath 之前。我一直把 Claude 当作合作者，从不把它仅仅看作一个工具。它对我来说就是合作者。我们在做深度研究时，我试图想出一些点子，如何在模型开始可视化搜索之前，让我感觉作为提出请求的人仍然有控制权，而不是让模型自由发挥、疯狂乱跑。我和 Claude 一起头脑风暴：“你觉得这个怎么样？那个呢？”它给出了一个想法：“如果你在提出请求时，我们展示预览卡片，列出这项研究可能分支出的所有不同可能性、所有不同的走向，然后我们点击我们想要的那个，再深入那个领域。”我当时想：“天哪，这太棒了，对吧？”然后我们一起构建了它。我们内部当然没有用 Cursor，但我们用 Claude 驱动的工具构建了它，做出了漂亮的 UI，团队非常喜欢，他们说：“这真是个好主意。”所以，你就像……这是我最想强调的一点：这不仅仅是视觉，不仅仅是交互，而是关于想法、合作伙伴和不同的事物，把它当作头脑风暴的伙伴，而不是“把这个落地页做得漂亮点”。因为过去设计师的主要抱怨是什么？“我不只是一个像素搬运工。我不是坐在那里推像素的猴子。”这就像那个梗，对吧？一直以来的梗就是他们对你说：“把 logo 放大一点。”

Pure impact. One of my favorite quotes is that the best idea I had was to have a lot of them, and I really really believe that. And this is basically, you know, it allows you to do that: just try a lot of ideas because the problem with ideas is like until you actually test the thing, it's just really hard to know how things are going to perform, right? And actually, I'll give you an even more interesting anecdote. So one of the things that I can talk about this now because it's out: I was working on deep research at Anthropic. I was trying out different kinds of functionality and things. This is way before I even started to think about Magikipath. I always use Claude as a collaborator. I never saw Claude as just a tool. It was a collaborator for me. And we were working on deep research and I was trying to come up with ideas around how can I visualize the searches before the model actually starts visualizing the search in a way that makes me feel I still have control as the person that asks the request versus the model just go free and go crazy. I started brainstorming with Claude: "What do you think about this? What do you think about this?" And he comes out and gives you this idea: "What if when you ask your request, we show a preview cards of all the different possibilities that this research can branch off with all the different ways that this research could go through, and then we click on the one that we want and then we kind of go deeper into that field." And I was like, "Holy, that's so good, right?" And we built it together. We didn't use Cursor internally, of course, but we built whatever tool we were using, powered by Claude, put this beautiful UI, and the team loved it. They were like, "This is such a good idea." And so it's like, you're most like... and this is the part I really really really want to stress: it's not just about the visual, it's not just about the interaction, it's really about the ideas and the partners and the different things, and using this as a brainstorming partner versus just "make this landing page pretty." Because what was the main complaint that designers had back in the days? "I'm not just a pixel pusher. I'm not a monkey that sits here and pushes pixels." That was like the meme, right? It's always been the meme that they say to you, "Make the logo bigger."

从某种意义上说，人们用这些工具就是在做同样的事。所以我通过 Magic Puff 教给人们很多东西，比如，在聊天中，试着跟它聊一聊，像头脑风暴一样。抛出一些想法，因为你可能会发现，在你真正去做之前，还有更多宝藏。这简直就是超能力，伙计。如果你能把它作为工作流程的一部分，你在公司里就势不可挡了。没有什么能阻止你，因为你成了参考对象，然后你就变得重要了。还有别的事，你知道的。我在 Anthropic 的时候，变得“重要”了。大家会说：“好吧，这件事我们怎么做？问问 Pietro。”而答案总是 Pietro 加 Claude：我的大脑加上它。

And in a way that's what people are doing with these tools. And so a lot of the stuff that I'm trying to teach people via Magic Puff is like, hey, in the chat, try to chat to it a little bit like brainstorm with it. Bounce off some ideas because you might find out that there are even more nuggets that you left even before you go and do the thing. And that's just a superpower, man. If you can just take that and use that as part of your workflow in a company, you are unstoppable. There's nothing going to stop you because you become a reference and then you become important. And that's other things, you know. When I was at Anthropic, I became quote-unquote important. It was like, "Okay, how do we do this thing? Let's ask Pietro." And it was always Pietro plus Claude: my brain plus that.

对。这就是那个等式，对吧？真的很有趣。

Right. It's the equation, right? It's just really fun.

真的很有趣。

It's just really fun.

好，我要说得非常具体了，因为我觉得你触及了一个极其重要的问题。我看到很多人抨击 AI 的输出，而实际上你利用 AI 的能力有一个无限宽的频谱，可以从中获得真正有用的东西。你刚才两次区分了工具和合作者。

Okay, I'm going to get really specific here because I think you're hitting on something that's super important. And I see so many people bash the output of AI when you have this infinitely wide spectrum of your ability to tap into AI and actually get something useful out of it. And so you've now twice made this distinction between tool and collaborator.

是的。

Yes.

我想我大概知道合作者是什么样的。也许深入探讨这个问题的一个有趣方式是：你能谈谈只把 AI 当作工具使用是什么样子吗？你可能会看到设计师有哪些行为，在你看来是信号，表明“嘿，我们得拓宽视野了，你能做的远不止这些”。所以，帮助人们理解这个频谱的不同侧面，理想情况下让设计师更自在地把 AI 当作合作者。有一阵子有个大梗叫“AI 垃圾”。AI 是垃圾，不好。

I think I kind of have a sense of what the collaborator looks like. Maybe the interesting way to dig into this would be, can you talk a little bit about what it looks like to only use AI as a tool? What are some of the behaviors that you might see designers using that in your mind would be a signal of like, hey, we got to broaden the horizons here. Like there's so much more that you could be doing. So help people understand the different sides of that spectrum as a way to ideally get designers more comfortable using AI as a collaborator. One big meme for a while was AI slop. AI is slop. It's not good.

它不如人类做出来的东西好，这在某种程度上是事实。让一位经验极其丰富的设计师去做一件事，跟模型一次生成的流水线相比，人类设计师当然表现更好。我想让设计师明白的是，当你跟这些模型对话时，它们拥有整个人类知识的压缩包。如果我现在跟你聊设计，然后突然问“你对合成这个分子有什么看法？”，你会一头雾水。AI 就是这样工作的：所有上下文都融合在一起。要获得最佳结果，你需要深入挖掘那些神经元激活，把模型变成优秀的设计师、优秀的经纪人或者股票交易员。这就是我想让大家理解的。当我说“用 AI”时，意味着学会如何跟 Claude 对话以获得好结果。Magic Path 通过消除摩擦做到了这一点——即使你问一个简单的问题比如“给我一个登录卡片”，它也会很漂亮，因为我们像做手术一样打开了模型的大脑，激活了负责设计的神经元。所以不要气馁。当人们说“哦，这太垃圾了”，我觉得他们只是试了一次，没得到想要的结果，就断定这玩意儿很蠢。有趣的是，跟人类交流也一样。如果我走到我的设计师 Jacob 面前说“你把这个马上改一下”，他会说“什么？”你需要上下文：目标、客户、用户、为什么这么做。这些东西激活你的大脑，也以同样的方式激活模型的大脑。这就是为什么很难把它们叫做工具——我想称它们为协作者。Anthropic 发布过一张图：先是聊天时代，现在是队友协作阶段，很快会进入先锋时代，模型能做出比我们独自完成更好的工作。如果你抓住这条曲线，随着模型越来越好，你会飞得越来越高。而且探索还能帮你激活其他方面的神经元，比如如何让家人感到被尊重。那也很重要。

It's not as good as something a human can make, which is true at some level. Take a super experienced designer and have them do something compared to a one-shot pipeline from a model—of course the human designer will perform better. What I want designers to understand is that when you talk to these models, they have a compressed zip file of all human knowledge. If I were talking to you about design and then asked, "What's your take on synthesizing this particular molecule?" you'd be confused. That's how AI works: all contexts are merged. To get the best result, you need to dig deeper into the neuron activations that turn the model into a great designer, or a great broker, or stock trader. That's what I want people to understand. When I say "use AI," it means learning how to talk to Claude to get good results. Magic Path does that by removing friction—even if you ask something simple like "give me a sign-in card," it will be beautiful, because we've surgically opened the model's brain and activated the neurons in charge of design. So don't get discouraged. When people say "Oh, it's slop," I think they tried it once, didn't get what they wanted, and decided it was dumb. Funny enough, the same happens with humans. If I walk to my designer Jacob and say "Just redo this on the fly," he'll be like "What?" You need context: goals, customers, users, why we're doing this. Those things activate your brain the same way they activate the model's brain. That's why it's tough to call them tools—I want to call them collaborators. Anthropic released a chart: first the chat era, now the teammate collaboration phase, and soon the pioneer era where models do work better than we could alone. If you ride that curve, you'll fly higher as models get better. And exploring helps you activate neurons for other things, like making sure your kids feel respected. That's important too.

我们能聊聊模型吗？我从来没问过别人这个问题，但我看到你一直在发推文说摆弄各种模型，而且你对每个模型的优缺点似乎有不错的把握。我想大多数听众现在的水平大概是：我知道什么是 deep research，我知道开始新对话时有个默认模型，也许我还知道 03 是用来思考的，但仅此而已。你在摆弄这些模型的过程中学到了什么，能让设计师受益？我特别想知道，如果你想有效利用 AI，不仅是在创建界面方面，而是在职业生涯中，你需要达到什么程度？

Can we talk about the models for a second? I've never asked anyone this question, but I see you tweeting all the time about tinkering with different models and you seem to have a decent grasp on pros and cons of each. I think the majority of people listening right now, I'm going to go ahead and assume that their level of proficiency or understanding of the models is somewhere around like I understand what deep research is and I understand that there's like a default when I start a new chat and maybe I know 03 is for thinking, but that's like it. What have you learned about tinkering with these models that you think designers could benefit from? And I'm very specifically interested in where's the line where it's like okay you probably should get to this point if you want to be able to effectively take advantage of AI not just in creating interfaces but like in your career.

有趣的是——我鼓励大家去玩每个模型的原因是，你可以把自己也叫做一个模型。某种程度上，我光看回复就能盲猜出是哪个模型。我知道是 DeepSeek、GPT 系、Anthropic 还是 Gemini，就凭回复。因为我花大量时间跟这些工具对话——可能比跟我女朋友聊得还多，我们还住在一起。这很疯狂，但你会积累知识，了解模型的品味。第一阶段：人们说“模型气味”。听起来好笑，但确实有模型气味。对于 3B 或 7B 参数的小模型，我能闻出那种小模型气味——它听起来有点卡通化，过于兴奋。比如“你为什么这么兴奋？我只是问纽约的建筑。”而大模型听起来像上过高中和大学的人在跟你聊天。这是第一部分：理解内在洞察和品味。第二部分是知道每个模型擅长什么。所有模型都在大量数据上训练，然后在特定子集上微调。Claude 3.5 之所以风靡世界，是因为它是一个编程模型——通用但专精于编程，这是 Anthropic 的秘诀。我们发现，当模型擅长编程时，它也是好的设计师。这并不总是成立，因为可能过拟合。看看 Grok：我把它放进 Magic Path 的流水线测试它的设计能力——很糟糕。为什么？因为他们在科学、数学、医学和太空方面深耕。有趣的是，在现实生活中，这类人通常也不是好设计师。这反映了现实。

A funny thing—the reason I push people to play with every model is that you can call yourself a model. In a way, I can tell you what model it is just by the response, blind. I know if it's DeepSeek, a GPT-based model, an Anthropic model, or a Gemini model just from the response. That's because I spend so much time talking to these tools—probably more than I talk to my girlfriend, and we live together. It's crazy, but you accumulate knowledge where you understand the model's taste. There's this first phase: people talk about "model smell." It sounds funny, but there is a model smell. For a small model like 3B or 7B parameters, I know that small model smell—it sounds almost cartoonish, too excited. Like, "Why are you so excited? I just asked about buildings in New York." Then bigger models sound like a person who went to high school and college, talking to someone. That's the first part: understanding the inner insight and taste. The second thing is knowing what each model is good at. All models are trained on large data but fine-tuned on specific subsets. Claude 3.5 took the world by storm because it's a coding model—general but specialized in coding, that's Anthropic's secret sauce. And we found that when a model is good at coding, it's also a good designer. That doesn't always happen because you might overfit the model. Look at Grok: I pulled it into Magic Path's pipeline to see how it performed as a designer—it was horrible. Why? Because they really deepened it on science, math, medicine, and space. Funny enough, in real life, those kinds of people are also not good designers. It mirrors reality.

就像，哦，你是科学家？那你肯定不擅长做史莱姆，对吧？但我打赌你很擅长写星际关系和黑洞模拟之类的东西，Grok 4 在这方面特别强。这是一种除非你花大量时间否则无法获得的技能。你问这个其实挺有意思的，因为今天我要发布一个链接。你知道玩电子游戏时有时会看到那种星形图吗？显示这个角色擅长什么，比如攻击强、防御强。我现在对模型也做类似的事。当你审视一个模型的能力时，我这样总结：有思考能力——模型内部思考的能力如何？从想法到行动的能力如何？有生成质量——模型输出的质量如何，无论是否经过思考？还有工具调用能力。我们社区现在意识到，这比什么都重要。原因在于，如果一个模型擅长工具调用，它就能被应用于很多场景。到目前为止，Claude 3.5 Sonnet 及其后续版本一直是工具调用的王者。Claude Code 之所以可能，是因为 Claude 的训练方式让它能进入一个循环，不断调用它知道自己有权访问的工具，而不会迷失。它会说：“哦，我想深入了解 Pietro Schirano。”然后它有一个搜索工具、一个抓取工具，还有一个打开页面阅读的工具。不知为何，Claude 在这方面平均比其他任何模型都做得好，比 o3、o4 mini、Gemini 都好。Gemini 在工具调用上其实很糟糕，我从来不在生产环境用 Gemini，因为它会迷失方向。它说“我要用网页搜索工具”，然后实际上没用。这也是为什么中国出了个新模型叫 Kimi K2。不知道你听说过没有？

It's like, oh, are you a scientist? For sure you suck at making slimes, right? But I bet you are good at writing down interstellar relationships and black hole simulations and all that kind of stuff, which Grok 4 is really good at. This is a skill you can't acquire unless you just spend a lot of time. It's actually funny you asked me this because today I'm going to publish this link. Do you know when you play video games sometimes there's a graph? It's almost like a star graph and you see what the character is good at. Oh, this guy has good attack, good defense. I'm doing that for models. When you look at the abilities of a model, I summarize it in this way. You have thinking abilities. So, how good is the model at inner thinking? How good is it at going from its thoughts to doing something? You have the quality of generation. So, how good is the generation you get from the model, even with or without thinking? And then the other thing is how good is tool calling. This is actually what we in the community are realizing is more important than anything else. The reason why is because if you are a good model that's good at tool calling, you can be employed in a lot of experiences. And so far, Claude 3.5 Sonnet and all the versions after that were the king at tool calling. The reason why Claude Code is possible is because Claude, due to the way it's trained, can go into a loop of running tools it knows it has access to and never lose itself. It says, "Oh, I want to know Pietro Schirano deeply." Well, I have a search tool, a fetch tool, and also a tool to open a page and read it. And somehow, for some reason, Claude can do that better than any other model on average. Better than o3, better than o4 mini, better than Gemini. Gemini actually sucks at tool calling. I never use Gemini in production because it just gets lost. It says "I'm going to use the web search tool" and then doesn't use it. This is also the reason why there's this new model from China called Kimi K2. I don't know if you're familiar with it.

没有，我没听说过。

No, I'm not.

它席卷了世界，至少是我的世界，因为它在工具调用上太厉害了。我发过一个演示，它简直疯狂，跟 Claude 4 不相上下。现在我们有史以来第一次有了一个免费且跟 Claude 4 一样好的模型，你可以把它装到一台足够强大的机器上——当然，因为它是一个万亿参数的模型。但有很多提供商，比如 OpenRouter、Groq 等等。这些都是非常重要的东西。你会发现，如果你在做像 Magipath 这样的应用，你确实需要一个 Anthropic 的模型，因为它们擅长工具调用。这基本上就是模型要成为一个好的设计师所需的所有流程。而这些事情你只需要反复测试。你构建自己的 Python 工具。我所有的工具都是自己写的。基本上，我选择的测试方法是：我要求做某件事，让它循环 10 分钟，然后看它迷失了多少时间、回退了多少时间、实际完成了多少。那些模型就是你认为真正擅长工具调用的。所以当 Kimi 2 上市时，我发了一条推文，视频里 Kimi 进入了这个工具循环，我说：“天哪，我们有了一个家庭版的 Claude 4。”这太疯狂了。

It took the world by storm, or at least my world, because it's incredible at tool calling. I actually released a demo where it just goes crazy. It's comparable to Claude 4. Now for the first time ever, we have a model that is as good as Claude 4 for free, that you can install into a machine with enough power, of course, because it's a one trillion parameter model. But you have a lot of providers like OpenRouter, Groq, and so on. These are the things that are really important. And what you realize is, if you're doing an application like Magipath, you do need an Anthropic model because they're great at tool calling. That's basically all the things, all the pipeline the model needs to do to be a good designer. And those are the things you just need to test and test. You build your own Python tool. I build all my tools. Basically, my test of choice is: I ask for something, I let it loop for 10 minutes, and see how much time it gets lost, how much time it goes back, how much time it actually does the thing. And those are the models you realize are really good at tool calling. So when Kimi 2 came to the market, I tweeted a video of Kimi going into this tool loop and I was like, "Holy, we have Claude 4 at home," which is crazy.

你刚才说的这些，你觉得设计师需要掌握其中多大比例的知识？还是说我们会走向一个世界，我们使用的软件只是动态地选择最佳方案，真正需要理解其工作原理的只有那些想构建自己工具的人？

The stuff that you're talking about, how much do you think that designers are going to be responsible for some percentage of that knowledge? Or do you think that we move more into a world where the software that we use is just dynamically using what is best at all times and the only people that really have to understand how it works are the ones that want to build their own tools?

我喜欢这个问题，这正是我接下来要说的。这也是我在 Anthropic 时深度参与的事情之一。如果你是一名设计师，工作涉及 AI 工具、AI 应用或任何 AI 相关的东西，你的角色已经发生了根本性的变化。过去，如果你在做 SaaS 应用，你的全部关注点和目标就是优化流程，确保客户点击正确的按钮，确保客户不会迷失方向等等。现在，不同之处在于，你不仅要与客户打交道，你实际上是在为一项任务雇佣一个人。你雇佣了一个所谓的“实体”来完成任务。所以作为设计师，你的主要工作应该是运用你的设计敏感性——那种让你与众不同的敏感性——并理解如何雇佣这些模型。当我在 Anthropic 做所有这些体验设计时，比如客服智能体、深度搜索等等，我 90% 的工作就是如何让 Claude 的行为不仅提供良好的体验、符合我的愿景，还要确保 Claude 拥有成功所需的一切。这是一种技能，如果你作为设计师能掌握它，你的价值将不可估量。有一件有趣的事我从未公开说过：当我拿到我公司 Magipath 的投资意向书时，同一周，我也收到了 Anthropic 的全职 offer，去领导一个叫“额外体验团队”的实验室团队。我晚上都睡不着，因为这不是在 Facebook 和 Google 之间选择，也不是在 OpenAI 和 Anthropic 之间选择。这是两种截然不同的生活，完全不同的生活道路。我认为我之所以能得到那个职位，是因为我最终成为了一个所谓的“模型沟通者”。你必须具备敏感性，懂得如何运用这种智能去做正确的事情。这就是为什么我们刚开始对话时，我用了一个非常具体的词：你是设计架构师。因为几乎就像你掌控着整个蓝图，而那个工人不知为何在错误的位置钻孔或敲锤子，你会说：“你到底在干什么？你应该在这里敲锤子。”只有设计师能做到这一点，因为我们拥有别人没有的敏感性和执着。我认为这就是设计的特别之处，也是设计会一直存在、不会消失的原因。

I love this question and this is where I was going to go next. This was also one of the things I was really involved in at Anthropic. If you're a designer whose role is to work with AI tools, AI applications, any sort of AI things, your role has fundamentally changed. In the past, if you were working on a SaaS application, your entire focus and goal was to optimize the flow, make sure the customer clicks the right button, make sure the customer isn't lost, and all that. Now, the difference is that you're not only dealing with customers, but you actually employ someone for a task. You employ a quote-unquote entity to do a task. So as a designer, your main job should be taking your design sensibility, that sensibility that makes you different from anybody else, and understanding how to employ these models. When I was working on all these experiences at Anthropic, like a customer support agent, deep search, and all that, 90% of my job was how can I make Claude behave in a way that not only provides a good experience and fits my vision, but also ensures Claude has everything it needs to succeed. That's a skill that, if you can acquire it as a designer, your worth is just incalculable. One funny thing I've never talked about publicly was when I got the term sheet for my company Magipath. The same week, I got a full-time offer from Anthropic to lead a team called the Extra Experience Team, like a lab team. I couldn't sleep at night because it's not like choosing between Facebook and Google, or between OpenAI and Anthropic. This is two different lives, fundamentally different life paths. And I think the reason I was able to even get that position is because I eventually became this person who was basically a model whisperer. You have to have the sensitivity to understand how to employ this intelligence to do the right things. That's why initially when we started this conversation, I used a very specific word: you are a design architect. Because it's almost like you're in control of the entire blueprint, and there's that worker who for some reason is drilling in the wrong position or hitting the hammer in the wrong place, and you're like, "Why the heck are you doing this? You should hit the hammer here." Only designers can do that because we have the sensibility and the obsession that no other people have. I think that's what makes design special, and that's why design is going to stay and it's not going anywhere.

这也是为什么当人们把我“不要用 Figma”的说法当作对设计师的攻击时，我真的很受伤。我是最狂热的设计师粉丝，我真心相信设计师的未来是被这个实体赋能，而不是被削弱。

And that's also why I was really hurt when people took my comment about 'don't use Figma' as an attack on designers. I'm the biggest designer fanatic and fan. I truly believe in the future of the designer as someone who can be empowered by this entity, not diminished.

这说得太好了，因为“架构师”这个词和概念在这个节目里经常出现，但通常是从理解材料的角度出发，这显然非常重要，也是我们讨论一切的前提。但你又往前推了一步，提出“不，不，不。作为设计师，要想最成功，你要去指挥员工、指挥别人为你做事。”哇，这已经远远超出了画矩形的范畴。

It's so good because 'architect' as a word and concept comes up on this show, but it's very much through the lens of understanding the materials, which obviously is very important. It's the precursor for everything we were talking about. But you're taking it one step further with this idea of 'no, no, no. As a designer, to be the most successful, you are going to be commanding employees and people doing things on your behalf.' And wow, that exists outside the box of drawing rectangles pretty far.

因为那样你的头脑就能去思考以前根本不会去想的事情。是的。

Because then that enables your head and your mind to think about things you wouldn't even think about before. Yeah.

所以我们很快会推出一个 Magic Path 的新功能。我知道一旦发布，它就会成为热议话题，因为没人想过这个。而且这之所以能实现，是因为我让那个员工待在房间的角落里，专心致志地做那件事，对吧？这需要战略思维，需要架构师思维，而不仅仅是像素。很多时候我看到设计师痴迷于像素，这很好，但像素并不能让产品成功，它只是让产品感觉不错，但这只是故事的一部分。其他所有事情是：这有用吗？人们在使用吗？你如何改变人们的生活？我收到大公司的邮件说：“天哪，因为 Magic Path，我们正在重新思考产品管线和我们构建产品的方式。”对我来说，这才是影响力。

So, we have a feature coming up soon for Magic Path. I know that once we release it, it's going to be the talk of the town because once you release it, everybody's going to be like, 'How the...' and nobody's ever thought about this. And two, that only comes because I told this employee to just be in this corner of the room and literally focus on that, right? And that takes this strategic mind, this architect mind, not just the pixel. Many times I see designers obsessed with the pixel, which is good, but the pixel is not what turns a product into a successful product. It's what makes it feel like a good product, but it's just one part of the story. All the other things are: is this useful? Are people using it? How are you changing people's lives? I get emails from big companies like, 'Oh my god, we're rethinking our product pipeline and how we build product because of Magic Path.' To me, that's the impact.

他们怎么说？他们的流程在如何改变？

What are they saying? How is their process changing?

比如他们不用再打开 Figma 文件了。他们直接开始制作 UI 和所有结构。基本上 Magic Path 成了事实标准，然后工程师直接去复制代码。因为设计师的愿景和像素级的完美已经在那里了，没有信息丢失。

Well, things like they don't have to open the Figma file. They just start making their UI and all the structure. Basically Magic Path becomes the source of truth, and then the engineer just goes and copies the code. Because the vision of the designer and the pixel perfection is already there. There's no loss in translation.

说到代码，我想回到模型理解门槛的问题，从另一个角度探讨类似的想法。我在网上经常看到设计师问：技术驱动的门槛到底在哪里？我需要学编程吗？我在 Reddit 上总看到这个问题。很多节目嘉宾都强调 Cursor 的重要性，说要摆弄代码，把 IDE 当作设计工具。你也在代码里做了很多像画布一样的事情。我很好奇你的看法：如果我是一个设计师，可能处于职业生涯中期，或者只做了两三年，想要为未来做好准备，我需要深入到什么程度？你怎么看？

Speaking of the code, I want to return to the threshold of understanding for the models question and approach that similar idea from a different angle. A question I see a lot in the wild from designers right now is: where is the more technical, code-driven threshold in general? Do I need to learn how to code? I see it all the time on Reddit. A lot of guests on the show talk about the importance of Cursor, tinkering with code, and thinking of IDEs as design tools. You're doing a lot of things in code that feel like a canvas. I'm interested in your perspective: if I'm a designer, maybe mid-career, maybe two or three years in, and I want to future-proof myself, how far do I have to go into this code-based world? What is your opinion?

对我来说，最大的梗就是代码会消失，对吧？代码会变成抽象。英语，如果你是英语母语者，会成为你的主要编程语言。我坚信这一点。事实上，我们是唯一一个除非你需要，否则根本看不到代码的工具。免费版的用户直接打开，疯狂尝试设计、探索。他们喜欢探索阶段和设计灵感，但更重要的是能够真正与某物互动，这太棒了。我不认为设计师如果不想学编程就应该去学。我实际上认为这些工具正在降低这个领域的门槛。最初我们说设计工程师会融合成设计架构师。这就是我相信的原因：因为编程作为我们需要说的语言会消失。它只会成为那个蓝图的基础设施。但你不需要关心那个，因为别人——员工、云或 GPT，无论你用哪个模型——都在地下室里。最终我们会变得非常好，代码基本上被解决了。疯狂的是人们没有在谈论这个：前端已经被解决了。前端编码已经被解决了。当然，还有状态、优化、加载应用时重新加载什么、新框架这些东西——它们会一直存在。但它不再是一个角色了。它不再是一个专门的职位。中型到大型公司的前端工程师曾经年薪接近 50 万美元，包括薪酬和股权。那个角色已经被解决了。所以我个人不能告诉设计师去学编程。我能告诉他们的是：理解 AI，因为它让你无需关心代码就能构建你的想法。这就是为什么我认为设计师不用 Cursor 或 Bolt 之类的东西——因为这些工具很吓人。我不想在创意状态中看到一大堆代码。没有经典设计师想那样做。这就是 Magic Path 看到的：一个画布。一切都是可视化的。你永远看不到代码。永远看不到，因为我们让它一直构建、一直工作、永远不会出错。然后如果你愿意，你可以弹出代码，但你不必。你用 Magic Path 分享的链接就足以体验你的愿景，获得初版，也许这本身就足够让你去找投资人，为你的公司融资。但我永远不会说学编程能让你成为好设计师。那不是事实。我会告诉人们：如果你想了解地下室里到底发生了什么，那就学编程。如果你想了解你房子下面那些基础是什么。

For me, the biggest meme has been that code will disappear, right? Code will become abstraction. English, if you're an English speaker, will become your main programming language. I am a huge believer of that. In fact, we are the only tool where you really do not see code unless you need it. People on free plans just pull it up, go crazy, try designs, explore. They love the exploration phase and the design inspiration, but also the ability to actually interact with something, which is huge. I don't think you should learn how to code as a designer if you don't want to. I actually think because these tools are lowering the entry level for this space. Initially we said the design engineer would merge into this design architect. That's why I believe it: because coding will disappear as a language we need to speak. It will just be the infrastructure of that blueprint. But you don't have to care about that because somebody else—an employee, cloud, or GPT, whatever model you use—is in the basement. Eventually we'll get so good that code is basically solved. The crazy thing is that people aren't talking about this: front end is solved. Front end coding is solved. Of course, there are things like states, optimizations, what reloads when you load your app, new frameworks—those things will always exist. But it's not a role anymore. It's not a dedicated role. A front end engineer at a medium to big company used to be paid almost half a million dollars a year, including comp and equity. That role is solved. So I personally cannot tell designers to go learn how to code. What I can tell them is: understand AI, because it allows you to build your idea without having to care about code. That's why I think designers don't use Cursor or Bolt or stuff—because those tools are scary. I don't want to be in my creative mindset and see a ton of code. No classic designer ever wanted to do that. And that's what Magic Path saw: a canvas. Everything is visual. You never see the code. Never see the code because we make it so that it builds all the time, it works all the time, you never get an error. And then if you want, you can pop the code off, but you don't have to. The link you share from Magic Path is enough to experience your vision, get a first take, and maybe that itself is good enough to go talk to an investor and raise money for your company. But I will never say that learning coding makes you a good designer. That's just not true. What I will tell people is: learn code if you want to understand what's actually happening in that basement. If you want to understand the fundamentals you build your house on top of.

我认为理解代码仍然有价值。我经常用这些工具写代码，但我本身也是开发者。有时候我看到 Claude 犯一些愚蠢的错误，比如为什么在这里用 useState？这种情况下你根本不该用 state。但这是否会阻止你实现你的想法？这更像是优化层面的问题。所以我认为，如果你想把产品在优化、后端等方面完全做出来，你仍然需要工程师。但教设计师后端？那太疯狂了。真的疯狂。

I think there's still power in understanding code. I code a lot with those tools, but I'm also a developer. Sometimes I see Claude making stupid errors, like why did you use useState here? You shouldn't have used state in this case. But would that stop you from making your idea come true? That's more like the optimization level. That's why I think you still need an engineer if you want to bring your product fully to fruition in terms of optimization, backend, and all that. But teaching desires backend? That's crazy. Actually crazy.

这让我想起 Intercom 设计副总裁 EMTT Connelly 说过的话。他基本上认为设计会吞噬前端。我非常同意。但当前的世界意味着需要两个工具：你在矢量工具里做设计，然后跳到 Cursor 里实现或做打磨修复，在代码里把像素落地。我对你的方法感兴趣，因为它让我作为设计师不仅拥有像素，还拥有前端的实现。感觉如何？怎么运作？怎么连接？有哪些不同状态？未来，接入数据可能也不难。我可以全部完成，而不必离开我的创意领域，进入工程思维的执行领域。

It's making me think back to something EMTT Connelly, VP of design at Intercom, said. He thought design was going to swallow the front end. I very much agree. The current state of the world means that requires two tools though. You'd make something in vector land, then hop into Cursor and implement it or do polish fixes, landing the plane on the pixels in code. I'm interested in your approach because it allows me as a designer to own not just the pixels but the implementation of the front end. How does this feel? How does this work? How does this connect? What are all the different states? In the future, it probably won't be hard to pipe in data. I can do it all without having to split and leave my creative land and move into engineering-minded execution land.

你猜怎么着？这些工作流程对我们来说一直有效。你把 Magikipad 换成 Photoshop，差不多一回事。Magikipad 让你快 100 倍、更好、真正的交互、真正的实时原型，而不是只活在图片里。自从我们服务人类以来，我们一直在这样做。我们是专业化的物种。我们喜欢在某件事上非常擅长，然后根据技能互相帮助。所以这为什么会消失？这就是为什么奇怪，也是为什么设计师焦虑，因为你不知道往哪走。我的梦想和希望是缓解这种焦虑。来 Magikipad 做设计吧，别为代码焦虑，别为学习 subbase 焦虑。实际上，我要分享这个。你会笑的。你知道 Figma 在 AI 上迷失方向的最大证据是什么吗？我给你看。我去 Figma.ai。它重定向到哪？我再做一次，给后面的人看。

And guess what? Those workflows have been working forever for us. You swap Magikipad with Photoshop. Same thing. What Magikipad allows you to do is move 100x faster, better, true interaction, true realtime prototyping instead of just living in an image. That's what we've been doing since we serve humanity. We are a specialized species. We like to be really good at something and then help each other based on our skills. So why would that disappear? That's why it's weird and why designers have anxiety, because you don't know where to go. My dream and hope is to ease that anxiety. Come design in Magikipad and don't be stressed about code. Don't be stressed about learning subbase. Actually, I'm going to share this. You're going to laugh. You know what's the biggest proof that Figma has lost the plot on AI? Let me show you this. I'm going to go to Figma.ai. What does this redirect? Let me do it again for the people in the back.

不，那太坏了。

No, that's evil.

它跳转到哪？

What does this go?

哦，你这家伙。

Oh, you savage.

你知道我为那个域名付了多少钱吗？一万美元。一万。你知道我是在 Config 前几周买的这个域名。你知道这说明了什么吗？

You know how much I paid for that domain? $10,000. 10k. You know when I bought this domain a couple weeks before Config. You know what this tells you?

哦天哪。

Oh my gosh.

Config 是最仓促的。我不想听起来很疯狂，但我可以向你保证：如果我从未发推说“没有理由再用 Figma”，他们就不会完全重做关于 AI 的 Config 构思。这又是一个证据。如果你看 YouTube 上的公告视频，产品经理上台说：“我来 Figma 才一个月，我们构建了所有这些 AI 工具。”所以像 Figma 这样以执行力和精准著称的公司，会推出像 Figma Make 这样有缺陷的产品，除非他们几个月前看到有大量用户不用他们的产品，对吧？这说明了问题。我没有公开这件事，是因为我不想被起诉。但他们给我发了停止函。我不知道。但我没做任何违法的事。我只是买了 Figma.ai。它是可用的。花了一万。我为什么要买？很疯狂，对吧？

Config was the most rushed. I don't want to sound crazy, but I can promise you: if I never tweeted that there was no reason to use Figma, they would not have redone the entire ideation of Config about AI. It's another proof. If you watch the announcement video on YouTube, the PM goes on stage and says, 'I've only been at Figma for a month and we built all those AI tools.' So a company like Figma, known for execution and precision, would ship something as broken as Figma Make, unless they saw something that got huge traction around not using their product a couple months before, right? That tells you. The reason I haven't been public about this is because I don't want to get sued. But they sent me a cease and desist. I don't know. But I did nothing illegal. I just bought Figma.ai. It was available. It was 10K. Why should I buy that? It's crazy, right?

我非常相信视频在解释设计师想法方面的力量。所以当需要给反馈时，我会在 Slack 里丢一个 Loom 链接和一个 Figma 原型链接，然后反馈散落得到处都是。一团糟。所以我正在构建我一直希望存在的产品，叫 Inflight。你可以把它看作异步评审。它让你轻松分享视频演示和交互原型或你设计的任何东西，然后 AI 会采访你团队的人，获取你需要的反馈，并把所有内容组织在一个漂亮的洞察页面上。目前我只对 Dive Club 听众开放。所以如果你想成为第一批使用 Inflight 的人，请访问 dive.comclub/inflight 抢占名额。

I'm a big believer in the power of video to explain my thinking as a designer. So when it's time to give feedback, I'll drop a Loom link in Slack and another link to a Figma prototype, and feedback will be scattered everywhere. It's a mess. So I'm building the product I've always wanted to exist, called Inflight. You can think of it as an Async Crit. It's an easy way to share a video walkthrough along with an interactive prototype or whatever you're designing, and then AI interviews the people on your team to get you the feedback you need and organizes everything in a beautiful insights page. Right now I'm only giving access to Dive Club listeners. So if you want to be one of the first to use Inflight, head to dive.comclub/inflight to claim your spot.

感觉就像你脚下的地面在微微移动。我认为重要的是要记住，我们俩仍然处于所有这些技术的绝对尖端。我不知道行业其他部分的转型会是什么样子，需要多长时间，但有很多人刚刚感受到下游效应。比如，“我的经理说我需要做交互式代码原型。我不知道该怎么办。这是否意味着我不再是设计师了？发生了什么？我对自己的角色未来感到害怕。”我每天都能看到这种情绪。我喜欢你的方法。

It literally feels like the ground you're standing on is shifting a little bit. I think it's so important to remember that both of us still exist at the absolute tip of the spear of all this technology. I don't know what the transition for the rest of the industry looks like, how long it will take, but there are so many people just now getting the downwind effects. Like, 'My manager said I need to be making interactive code prototypes. I don't know what to do. Does this mean I'm not a designer anymore? What's happening? I'm scared about the future of my role.' I see that sentiment every single day. And I like your approach.

我喜欢它如何恰好处于一个很好的中间地带，让你作为设计师不必为了成功过渡到 AI 世界而做出存在主义式的跳跃。而且公平地说，这些本来就是人们过去要做的事情。当你提到你的经理要求设计具有响应式时，你不得不做。但现在，你不用花几个小时，只需说“让这个响应式”或“从这个屏幕到这里”，就能实现。然后你当然会精修：“从这个屏幕到这个部分，但用这种图片。”在讨论设计系统之前，展示这个也很酷。定义设计系统的方式：人们来到这里，这里我有一套 Reddit 的设计系统，因为几天前我和 Reddit 的某个人聊过。你构建设计系统：你的主色、强调色、所有属于设计系统的部分，你的字体、字号。我们很快会允许添加字体，大概几天后。然后是阴影等效果。你可以实时预览这个主题的样子。如果我想让圆角更小或更大，你可以实时看到变化。

I like how it kind of fits into this nice middle ground where it doesn't feel like you're having to make this existential jump as a designer in order to successfully navigate this transition into an AI world. And to be fair, those are already things that people used to do. When you were saying your manager asks for the design to be responsive, you had to do that. But now instead of spending hours on it, you can just say 'make this responsive' or 'from this screen go here' and you get there. Then of course you finesse it: 'from this screen you go to this part, but with this type of image.' This is also cool to show you before I talk design systems. The way you define a design system: people come here, and in this case I have a Reddit design system because I was talking to somebody at Reddit a couple days ago. So you build the design system: your main color, primary color, accent, everything that is part of the design system, your typography, your scales. We're going to allow adding fonts here pretty soon, probably in a couple days. And then effects like shadows. You can in real time preview what this theme might look like. If I want the border radius to be smaller or larger, you can see in real time how things change.

但我们还有规则，某种程度上这就像光标规则，但用于设计。所以这里我可以写“我的动画必须使用这种贝塞尔曲线过渡，每次元素进入页面需要 0.4 毫秒，但离开页面时是 0.6 毫秒”，因为这是你做交互设计时学到的东西。交互设计的核心你可以直接放在这里，然后每次使用设计系统生成时，你不仅得到看起来来自设计主题的东西，还实际拥有这些规则。

But then we also have rules, and in a way it's almost like cursor rules but for design. So here I can say things like 'my animation always has to be a bezier curve of this transition, every time an element comes into the page it needs to have 0.4 millisecond, but once it leaves the page it's 0.6 millisecond' because of the stuff you learn as a designer doing interaction design. The core of interaction design you can just put them here, and then every time you use a design system to generate, you don't only get things that look like they're coming from the design theme, but you also actually have the rules.

你看到人们把这一点推到什么程度了？我很好奇自定义设计规则。我的意思是，你可以在里面放任何东西。你注意到了什么？什么有效？你看到什么有趣的用例了吗？

How far have you seen people push this? I'm really curious about the custom design rules. I mean, you could put anything in here. What are you noticing? What's working? What's maybe an interesting use case that you've seen?

是的，人们大量使用它来做动画。这基本上是主要用途之一。对于无障碍性，很多人去那里写“对比度永远不能低于这个或高于那个”。所以基本上他们用这个来实现精确性，这正是 AI 目前缺失的，而我们正在解决这个问题。我们做的另一件很酷的事情是让它能与任何公司使用的标准全局 CSS 兼容。例如，有一个叫 Twixden 的工具非常流行。人们来到这里，有很多设计系统。比如说，有一个叫 CloudForms 的，非常有趣。我们的做法是，你甚至可以访问另一个网站，或者从你的公司导入你自己的全局 CSS。你只需复制它，进入 MagikPath，点击导入，粘贴到这里，砰，设计系统就出现了。

Yeah, so people are using it a lot for animation. That's basically one of the main things. For accessibility, we have a lot of people going there and saying things like 'the contrast never needs to be less than this or more than that.' So there's a lot of using this for precision, basically, which is what's missing with AI right now, and we're basically solving that. The other really cool thing we did is we made it work with any standard global CSS that you would use in any company. For example, there's this tool called Twixden that's very popular. So people come here and there are a bunch of design systems. For example, let's say there is CloudForms, which is very interesting. The way we did it, you can even come to another website or import your own global CSS from your company. You can just copy this, go to MagikPath, do import, paste it here, and boom, now the design system is here.

太酷了。这正是我的问题：我如何轻松地从 Cursor 中的东西转到 MagikPath？

That's cool. That was a question I had: how easily can I get from something that I have in Cursor to MagikPath?

非常容易。除此之外，你还可以立即预览深色模式。因为我们构建的方式是深色模式和浅色模式是嵌入在一起的。所以实际上，你进入深色模式，有时可以看到深色模式的设计，如果我把我的应用设为深色模式。另一件我想给你看的事情回到了整个想法：我应该害怕使用代码吗？到什么程度？因为我认为人们对代码的另一个担忧是认为它会限制他们的创造力。这种恐惧是：“天哪，它太死板了。”所以你看，这是我们最近与 Contra 合作的一个比赛。我们下周还要宣布获胜者，但我们和 Contra（一个设计社交平台）做了一个比赛。这就是设计师在 MagikPath 上做的东西。你告诉我这看起来像 AI 做的吗？

Very easy. And on top of that, you can also preview dark mode right away. Because the way we built it, dark mode and light mode are embedded together. So actually, what happened is that you go into dark mode, sometimes you can see a dark mode design if I make my app dark mode. The other thing I want to show you goes back to the whole idea: should I be scared of using code? How much? Because I think another thing people are worried about with code is they think it's going to limit their creativity. That's the fear: 'Oh my god, it's rigid.' So look, this was from a recent contest we did with Contra. We still have to announce the winner next week, but we did a contest with Contra, which is a design social platform. And this is the stuff that designers are making on MagikPath. You tell me if this looks AI to you.

是的。

Yeah.

没错。这就是我要说的。你可能会付给一个代理公司一大笔钱来制作这个网站。而现在这个人，他真的很棒。如果你看他的页面，那就是他的美学风格。所以他能够把那种美学带回 MagikPath。比如 Zuyu，他做的一切都非常干净、简约、漂亮。他是个很棒的设计师，他正在使用一个工具让他输出效率提升 10 倍，因为现在他不再只是展示一个屏幕，而是给客户发送他们可以感受和尝试的东西，而且是响应式的等等。所以这也是有趣的一点：不要害怕这会限制你的创造力。你的创造力只受限于你投入这些工具的精力，这真的很有趣。

Right. And that's what I'm saying. It's like you would pay an agency a considerable amount of money to make this website. And now this guy, he's just really good. So if you look at his page, that's his aesthetic. So he was able to move that aesthetic back into MagikPath. Like Zuyu, everything he does is very clean and minimal and pretty. He's a great designer, and he's using a tool that allows him to 10x his output because now instead of just having this screen, he's actually sending clients things they can feel and try, and it's responsive and all that kind of stuff. So that's also the interesting thing: do not be afraid that this is going to limit your creativity. Your creativity is only limited by the amount of effort you're trying to put into these tools, which is really interesting.

而且这个设计之所以出色，正是因为它非常有活力。有很多动效，元素如何淡入。我可以看到，对于网页设计来说，在 Framer 或 Webflow 中一个人拥有所有这些可能更现实。但如果你在做产品设计，通常那部分不是设计交付物的一部分，而是留给工程师去做。所以我认为这是一个很好的例子，这个设计的很大一部分在于它的感觉，而现在设计师很容易就能拥有交付物的那部分。

And even just what makes this design is that it's very alive. There's lots of motion, how things fade in. And I can see maybe for web design, it's a little bit more realistic for someone to own all of that in a Framer or a Webflow. But if you're doing product design, it's very typical for that to not be part of the design deliverable and that's left up to the engineer. And so I think it's a great example of so much of what that design was was how it felt, and now it's easy for the designer to own that part of the deliverable.

而代码是实现这一点的途径，甚至你都不用看到它。

And code is the way to get there without you even seeing it.

我想确保我们覆盖了所有基础。我们讨论了很多不同的事情，我仍然对设计架构师这个概念以及这个角色在未来是什么样子非常着迷。所以我想确认我们是否已经谈到了所有内容。有没有什么我们还没讨论到的，你认为会造就一个成功的设计架构师，或者帮助人们从今天的位置过渡到未来那种角色？

I want to make sure that we're covering our bases. We've talked about a lot of different things and I'm really still fascinated with this idea of a design architect and what that archetype of the role looks like in the future. So I just want to make sure that we've talked about everything. Is there anything that we haven't talked about yet that you think will make for a successful design architect or help people go from where they're at today to moving into more of that type of role in the future?

即使你的工作可能因为速度更快而变得更轻松，你的责任实际上却更高了。因为作为设计架构师，你现在不再只是负责“哦，这个看起来不错，回家吧，放松一下，我完成了。”

Even if your job might get easier because you're moving faster, your responsibility actually goes higher. Because as a design architect, now you're not just in charge of 'oh this looks good, let's go home, let's relax, I'm done.'

现在，你不再需要把工作交给工程师、交给前端人员去处理动画、过渡效果，比如元素如何淡入淡出、侧面板怎么放置、如何移入移出。你完全掌控一切。而当你完全掌控时，你需要关注多得多的细节。所以，我想强调，你现在比以前拥有更大的权力，因为你过去一直局限在非常孤立的环节里。这几乎就像 AI 给了你一次晋升。AI 把你从一个像素搬运工——也许是有经验的、非常资深的产品设计师、一个真正有战略眼光的人——提升为体验的守护者和架构师。因为现在，在产品开发史上第一次，你掌控了体验中进出的所有东西。这以前从未发生过。我的意思是，以前也有过，但那需要你花几个月的时间去打磨，与团队沟通、开会、讨论，甚至争吵。

Now, you're not sending that to an engineer, to somebody else, a front end person to work on the animation, the transition, like how the things fade, where the side panel goes, how it moves in, how it moves out. You are in full control. And if you are in full control, there's so much more details that you have to care about. And so that's why, in a way, I want to stress that you have more power now than before because you were living in this very isolated stuff. It's almost like AI just gave you a promotion. AI just promoted you from a pixel pusher, maybe experienced, really like a very experienced product designer, somebody that's really strategic, to actually now be the keeper, the architect of the experience. Because now, for the very first time in history of product development, you are in charge of everything that comes in and that comes out on that experience. And that's never happened before. I mean, it has happened, but it would take you months of refinement, working with the team, chat and conversation and meetings and yelling.

这也是一种独角兽技能组合。没有多少人能在这两方面都做到顶尖。

It's a unicorn skill set too. Not that many people can be spike on both of those.

有多少人拥有 Figma 那样的产品？有多少人拥有 Google 那样的产品？这是一个非常罕见的角色，对吧？所以对我来说，有趣的地方在于：你可以休息一下。不是说“哦，我现在有 AI 了，所以我可以放松了”。不，你作为设计师的责任一夜之间增加了 10 倍。如果你能掌控这一切，你就能拿到 Anthropic 的 offer，拿到 Google 的 offer，拿到你梦想工作的 offer，他们会给你很高的薪水，因为猜怎么着？现在没人能做到这一点。没有人能像你一样擅长这件事，这是一个需求极高的角色。产品设计师转变为产品架构师和设计架构师，同时具备这种与模型对话的感知力。我认为这将成为所有公司中最重要、最昂贵的角色，我可以向你保证，这将是任何公司中薪酬最高的角色。

How many had the product of Figma exist? You know how many had the product of Google exist? It's a very rare role, right? And so that's the interesting part to me. It's like you can rest. It's not like 'oh I have AI now so I can just chill.' No, your responsibility as a designer is just 10x overnight. And if you can control that, you can get the offer at Anthropic, you can get the offer at Google, you can get the offer at your dream job and they can pay you really well because guess what? Nobody can do that right now. There's no one that is as good at this stuff as you could be and it's a very high demand role. The product designer that becomes a product architect and design architect and also has this sensibility and talking to the model. I think that's going to be the biggest most expensive role for all those companies and I can promise you that's going to be the highest pay role in any company.

我想我同意。也许一年半、两年前，连我自己都感到有些紧张。那时我们刚有了 UI 生成的东西，而代码能力还有点滞后，我当时想‘哦，天哪，会发生什么？’然后突然之间，一夜之间，就像‘哇，AI 其实很擅长写代码’。我觉得我能写代码了，也许一些设计方面的工作……我对这个说法有点保留，因为我认为你刚才分享的一些内容确实很好。但有一个时刻，我心想‘哇，现在 UI 反而滞后了，代码却变得很好。’那时我很庆幸自己拥有设计师的技能，因为我能掌控这个方程中更多的部分。人们带入这个方程的核心基本技能和视角是什么？

I think I agree. And maybe a year and a half, two years ago, even I felt a little bit nervous. It was when we first had the UI generation stuff and code was actually lagging a little bit, where it was kind of like 'oh man, what's about to happen?' And then all of a sudden overnight, it was like 'whoa, AI is pretty good at code actually.' I feel like I can code and maybe some of the design stuff was... I'm kind of asterisk by this statement because I think some of the stuff you just shared was pretty good. But there was this moment where I was like 'whoa, actually now UI is lagging and code is really good.' And it was like I'm pretty grateful to have this skill of a designer because I can own so much more of this equation. And what are the core fundamental skills and vantage points that people bring into that equation?

是的。

Yeah.

当软件在很大程度上被民主化，至少前端及以上的部分，我很庆幸自己是一名设计师。

When software is largely democratized, at least front end and up, I'm pretty grateful to be a designer.

完全同意。有趣的是，这个角色的入门门槛也在越来越低。因为如果你行动不够快，团队里可能有个 PM，因为他们有自主权——在 AI 时代，自主权非常重要——那个人可能在周末就用 AI 把整个端到端流程跑通了，可能获得产品批准，甚至可能把你排除在外，如果你不明白这些工具能多快、多迅速地使用的话。但如果你能那么快地行动，并且能融入你作为设计师的愿景——从根本上说，设计师的愿景平均可能比 PM 更好——那么真正的魔力就发生了。那时你基本上是不可替代的。

Totally. And what's interesting about this is the entry level for that is also going lower and lower. Because if you don't move fast enough, there might be that PM on the team that, because they have agency — agency is a really important thing in the age of AI — that person in the weekend might spit the entire end to end using AI and might get product approval and might even cut you out if you don't understand how fast and how quickly you can move those tools. But then if you can move that quickly and you can embed your vision as a designer, which is fundamentally could be better on average than a PM, then is where the true magic happens. And that's when you're irreplaceable basically.

而且你可以用完全相同的逻辑来应对不同级别的资历和职业位置。比如，如果我是一名非常资深的设计师，还在手工拼凑静态模型，花三天时间做一个原型，而你刚从大学毕业，你就能快速做出一个交互式的东西，扔到 Slack 里，而我还在第三屏上。

And you can use that exact same logic for different levels of seniority and career position too. Like if I'm a very senior designer and I'm still hand assembling noodles of static mocks and spending three days on a prototype, and you just graduated from college and you just crank out something that's interactive and drop it in Slack and I'm on screen three.

你提到这个很有趣。你知道我收到的最多的邮件请求是什么吗？是学生折扣。

It's funny you mentioned that. You know what's one of the biggest requests I get over email is college discounts.

哇。

Wow.

我认为这一代从大学出来的新设计师，把 AI 作为他们工作流程的一部分，他们会大杀四方。你根本无法竞争，对吧？就是无法竞争。这几乎就像印刷和网页，对吧？你怎么竞争？

I think this new generation of designers coming up from college that use AI first as part of their flow, they are gonna crush it. You just can't compete, right? It's just you can't compete. It's almost like print and web, right? How do you compete?

几十年来，我们在每一个行业垂直领域都看到了这一点，自从我们成为技术社会以来。你可以看到清晰的代际界限，区分谁是在什么水平的原生技术中长大的。我们即将看到 AI 原生一代进入劳动力市场。这会变得疯狂。Z 世代之后是什么？Alpha 世代。我不知道。

And we've seen that across every single industry vertical for decades, like ever since we became a technological society. You can see the clear generational lines between who grew up with what level of native technology. And we're about to see the AI native cohort enter the workforce. And it's going to get crazy. What's after the Gen Z? Gen alpha. I don't know.

我不知道，我从来记不住这些。我不是故意说的。

I don't, I never can keep track of these. I didn't say it intentionally.

随便吧。某世代。这些世代的孩子。这些世代的孩子会碾压我们。这就是现实。

Whatever. Gen something. The Gen kids. The Gen kids are going to smoke us. That's the reality.

他们会大放异彩。

They're going to cook.

Pedro，这太有趣了。我真的被你所做的、你思考问题的方式所启发。你关于将哪种未来带入当下、以及这对设计师意味着什么的总体看法，与我的想法非常一致。这真的很令人兴奋。所以，感谢你今天来做客并分享。

Pedro, this has been super fun, man. I'm really, really inspired by what you're doing, how you're thinking about things. Just your general take on the type of future that you want to bring into the now and what that all means for designers really aligned with it. It's really exciting. So, thank you for coming on today and sharing with us.

非常感谢你的邀请。

Thank you so much for having me.

在你走之前，我想花一分钟给你介绍一下我最喜欢的产品，因为我经常被问到我的工具栈是什么。Framer 是我建网站的工具，Genway 是我做研究的方式，Granola 是我在评审时记笔记的工具，Jitter 是我为设计添加动画的工具，Lovable 是我用代码实现想法的工具，Mobin 是我寻找设计灵感的地方，Paper 是我像创意人一样设计的工具，而 Raycast 是我每一步的快捷方式。我精心挑选了这些公司，这样我就能全职做这些节目了。所以，到目前为止，支持这个节目的第一方式就是去看看它们。你可以在 dive.comclub/partners 找到完整列表。

Before I let you go, I want to take just one minute to run you through my favorite products because I'm constantly asked what's in my stack. Framer is how I build websites. Genway is how I do research. Granola is how I take notes during crit. Jitter is how I animate my designs. Lovable is how I build my ideas in code. Mobin is how I find design inspiration. Paper is how I design like a creative and Raycast is my shortcut every step of the way. Now, I've hand selected these companies so that I can do these episodes full-time. So, by far the number one way to support the show is to check them out. You can find the full list at dive.comclub/partners.