设计 Grok Bot：从 Cursor 到消费级产品

欢迎来到 Dive Club。我是 Rid，这里是设计师永不停歇学习的地方。Grok Bot 这周一直在病毒式传播。所以今天的节目请到了 John B，他是 Cursor 纽约办公室的第一位设计员工。我们将聊一聊他早期做的一些探索，深入他的设计流程，也看看他在自己的实践中是如何使用 Grok Bot 的。那么，话不多说，我们开始吧。

Welcome to Dive Club. My name is Rid and this is where designers never stop learning. Grok Bot has been going viral all week. So today's episode is with John B who was the first design hire in the New York office for Cursor. And we're going to talk through some of the early explorations that he was doing, get into the weeds of his design process, and also look at how he's using Grok Bot in his own practice. So, without further ado, let's dive in.

在 Cursor 的整个期间，我们一直在做 Cursor 应用本身，也就是桌面应用，而我长期负责增长方面的工作。其中很多与新手引导有关，努力让新手引导流程更容易上手。其实在我加入 Cursor 之前，我就强烈感觉到这是一个强大的工具，但像我这样的非技术用户很难适应。很多非常工程化的概念，让一个可能习惯用 Figma 之类的设计师更难把它作为主要工具来使用。

During my entire time at Cursor, we have been working on the Cursor app itself, the desktop app, which I was working on growth for a long time. A lot of that had to do with onboarding, trying to make the onboarding flow more accessible. Before I even joined Cursor actually, I felt very strongly about this being a powerful tool that the nontechnical user like myself had a hard time getting acclimated to. A lot of these concepts that are very engineering focused make it harder for a designer who's maybe used to using Figma or something to enter into this as the primary tool.

在这个过程中，我们对于这个桌面应用能做什么，遇到了一个类似十字路口的抉择。第一种是努力让非技术用户更容易上手。围绕这个我们做了很多举措，也跑了很多实验。公司内部还有另一个阵营，他们强烈认为我们必须从头开始。

Along the way we had kind of like a crossroads approach towards what we could do with this desktop app. The first one was try to make it more accessible to the nontechnical user. There were a lot of initiatives around that, a lot of experiments that we ran. There was another camp within the company that felt very strongly that we just had to start fresh.

其实我觉得有点好笑的是，我当初提的一个方向，恰好就是 Codex 后来用切换开关做的事情。我当时想，如果我们有两种模式会怎样？我们有协作模式，你知道的，显然，或者我们只有编码模式。我们内部把它叫做编码与任务。

I actually kind of find it funny that one of the directions that I was pitching was kind of exactly what Codex ended up doing with their toggle. And I was like, what if we just had two modes? We have the co-work mode, you know, obviously, or we just have the coding mode. We were calling it internally like coding versus tasks.

团队里大多数人觉得它背负了太多包袱。Cursor 这个品牌，虽然备受尊敬和喜爱，但可能让任何非工程师都不想下载尝试，因为它太技术化了。所以，虽然某些实验在让企业团队、市场或销售人员用它完成特定任务方面表现不错，但 UI 的设计方式，本质上仍然非常技术化。

Most of the team felt like it just contained way too much baggage. The Cursor brand, while very respected and beloved, probably felt too technical for any non-engineer to want to even download and try. And so, while certain experiments did perform well at getting maybe an enterprise team, their go-to-market or salespeople using it for specific tasks like that, the UI, the way it was designed, was still at its core very technical.

最终发生的是，当我们还在推动非技术用户的采用时，有一个自上而下的指令，要求尝试一些新东西。而这个非常有趣。团队里最顶尖的工程师，一些最资深的人，他们自己进了一个房间，开始探索这可能是什么样子。我想早期在公司内部，有很多怀疑，或者说不理解，为什么采用这种方法，以及它是否真的能成功。但我觉得他们逐渐开始发布，最终成为 Grok Bot 的初始版本，基本上就是 Cursor Glass 的精简版，也就是我们的智能体窗口。

What ended up happening was while we were still pushing on adoption for the nontechnical, there was a mandate kind of top-down to just try something new. And with this one, it's super interesting. You get some of the most cracked engineers on the team, some of the most OGs, and they go into a room on their own and start exploring what this could look like. And I think early on, even within the company, there was a lot of skepticism or maybe lack of understanding about why this was the approach and whether it would work out at all. But I think gradually they started shipping and the initial version of what would eventually become Grok Bot was essentially like a paired down version of Cursor Glass, which is our agents window.

聊天功能看起来和感觉上也大不相同，因为它更像 iMessage。你可以看到，这是让 Grok Bot 对普通用户更易用的重要因素之一。虽然 AI 重度用户非常习惯流式文本、思考状态，但大多数日常使用消息工具的人，觉得 iMessage、WhatsApp、Messenger 交互起来最舒服。我认为这最初就是那种秘诀，让我们内部知道工程师们发现了什么。但我认为在很长一段时间里，它最终会变成什么形态，以及为什么以后会有用，仍然非常不清楚。

The chat functionality looked and felt much more different as well because it was much more iMessage-like. And you can see that kind of being one of the big factors in making Grok Bot feel more accessible to the average user. While AI power users are very used to that streaming text, that thinking state, most everyday people that use messaging tools find iMessage, WhatsApp, Messenger most comfortable to interact with. And I think that initially was kind of that secret sauce that let us know internally that the engineers were on to something. But I think it was still very unclear for a very long time what shape this would take on and why it would be useful later on.

我认为这也是设计非常重要的地方，而且内部再次出现了分歧。一种方法是把正在构建的东西打磨一下，尽量让它可用，然后发布这个本质上就是侧边栏加主聊天的版本。显然侧边栏里还有计算机使用功能。但当我们这样推进时，我觉得很多设计团队成员都感到好奇，或者想探索一些不同的东西，而这就是我在整个过程中的方法。我很乐意分享我们经历的一些探索。但大部分情况下，一种方法是尽量保持与工程师构建的相似。另一种方法是看看我们能做出多大的不同。最终两者融合，真正结合成了两全其美的东西。

I think this is also where design is super important and once again there was a divergent path internally. One approach was to take what was being built and apply some polish to it. Try to make it as usable as possible to ship this version of what is essentially sidebar and main chat. Obviously there's also the computer use in the side pane. But as we were approaching it that way, I think a lot of the design team felt kind of a sense of curiosity or wanting to explore something different, and that was very much my approach in this whole process. And I think I'm happy to share some of the explorations that we went through. But for the most part, one approach was keep it as similar to what was being built by the engineers. The other approach was see how different we can make it. And eventually the two converged and really married on something that is the best of both worlds.

你把它做得有多不同？

How different did you make it?

说实话，早期的一些东西非常超前。我最初的疑问之一是，为什么这需要是另一个基于聊天的界面，为什么界面仍然需要是左侧面板列出智能体，中间面板进行主要聊天，然后右侧是你深入查看的任何额外细节。对我来说，非常值得探索一些更环境化的东西，也许是一些感觉更贴近实际操作系统的原生体验。

Some of the early stuff honestly was way out there. One of the first questions that I had was why does this need to be another chat-based interface and why does the interface still need to be the left pane with your list of agents, the center pane with the primary chat taking place, and then the right side being any additional details that you dig into. To me it was very worth exploring something more ambient and maybe something that felt a lot more native to the actual operating system.

快速插播一条消息，然后我们继续。我提前看到了 Desen 新的私有测试版。我跟你说，他们把无限画布提升到了一个全新的水平。这不仅仅是随意的模型或一次性的原型。你可以直接在画布上运行真实的产品页面。想想这为并行探索解锁了什么。你可以随意分支。让智能体并排探索多个想法，这对于团队对齐非常完美，因为你有一个空间来审查原型、留下评论，甚至让智能体根据你收到的反馈采取行动。这很重要。但问题是，Desen 每周只接纳几个团队。但如果你访问 dive.club/desen，你可以和他们的联合创始人 Gab 预约聊天，告诉他们 RID 让你来的，就可以插队。所以去 dive.club/desen。就是 deesn。自从 Paper 通过他们的 MCP 推出令牌以来，我能用智能体设计和探索的质量绝对飙升了。我的意思是，画布上出现的所有东西看起来都和成品无异。现在你可以在 Paper 内部直接创建和编辑令牌了。

Real quick message and then we can jump back into it. So, I got a sneak peek of Desen's new private beta release. And let me tell you, they are taking the infinite canvas to a whole other level. It's not just random mock-ups or one-off prototypes. You can run real product pages directly on a canvas. I mean, just think about what that unlocks for parallel exploration. You can branch as much as you want. Get the agent to explore multiple ideas side by side, which is perfect for aligning the team because you have one space to review prototypes, leave comments, and even get the agent to act on the feedback you receive. It's a big deal. But here's the thing. Desen is only onboarding a few teams every week. But if you go to dive.club/desen, you can book a chat with their co-founder, Gab, and tell them RID sent you to jump the queue. So head to dive.club/desen. That's deesn. Ever since Paper launched tokens via their MCP, the quality of what I'm able to design and explore with agents has absolutely skyrocketed. I mean, everything that lands on the canvas looks indistinguishable from production. And now you can create and edit tokens directly inside of Paper.

而且他们把用户体验做得很好。比如你甚至可以在文件之间复制粘贴 token。这也是 Paper 成为我首选设计工具的又一个原因，我强烈推荐它。今天就前往 dive.club/paper 开始设计吧。现在进入正题。

And they kind of nailed the UX. Like you can even copy and paste tokens between files. It's just another reason why paper has become my go-to design tool and I cannot recommend it enough. Head to dive.club/paper to start designing today. Now on to the episode.

这个的第一个版本，我的意思是，你知道，不是特别有用或出彩，但它本来是一个任务列表，你把这些任务分配给一个智能体，然后你能实时看到它执行这些任务。

First version of this, I mean, you know, not particularly useful or inspired, but it was meant to be a list of tasks that you assigned to an agent and you would be able to see it perform those tasks in real time.

哦，这就像一个悬停小窗口，能看到智能体。这挺酷的。

Oh, this is like a little hover window into the agent. That's kind of cool.

没错。是的。而且其中很多都只是我在尝试一些听起来很酷的东西。我想知道……对，如果你点进去，它会给你任务的详细信息。你可以点击窗口本身，给它一个后续指令。目前很多功能还有点问题，但有一些想法更像是任务控制中心的视图，每个智能体都在运行自己的电脑，运行自己的任务，你可以看到它们在你的机器上工作。你想让它们做别的事，就再给它一个任务，再加一个空间。我觉得这些都是非常酷、值得探索的东西，但当我向领导层展示这些时，反应有点冷淡。我想大多数人只是认为，那些淡出背景的东西，不如我们想要作为品牌核心的那种工具那么相关或那么突出。

Exactly. Yeah. And like a lot of these were just me trying out different things that sounded cool. I wonder if... Yeah, if you click into it, it would give you details on the task. You'd be able to click into the window itself and give it a follow-up. A lot of it's kind of broken at this point, but there were ideas around maybe more like a mission control type of view where each agent is running their own computer, running their own task, and you can just see them at work on your machine. You want them to do something else, you give it another task, add another space. I think these were really cool, worthwhile explorations, but when I showed these to leadership, the reaction was kind of lukewarm. I think most people just assumed that something that faded out into the background was not as relevant or as front and center facing as the type of tool that we want to put our brand behind.

我能快速确认一下你的流程吗？所以你在网页上运行这个。那么你一开始……你是在 Cursor 里做原型吗？这是你的第一步吗？你基本上是在说：“嘿，让我们在浏览器里模拟一个桌面。”

Can I confirm something about your process really quickly? So you're running this in web. So you're starting off... are you prototyping in Cursor? Is that your first step? And you're basically saying, "Hey, let's simulate a desktop in the browser."

不完全是。我会说对我来说很多这类工作还是先在 Figma 里进行。让你了解一下我是从哪里开始做这类东西的。对我来说……

Not exactly. I would say a lot of this stuff for me still takes place in Figma first. Give you an idea of where I started with something like this. For me...

你确实做了这些模型。好的。

You did mock these up. Okay.

是的。快速做出几个概念仍然快得多。你看到的有些东西，你知道，桌面上的浮动聊天框，或者更小的形态，更紧凑。如果这样会是什么样子，你知道，你刚才看到的，横向滚动、纵向滚动那种形态，它会演变，会呈现不同的形状，也会有不同的外观和感觉。Figma 成了这些想法的收集空间，让我先把它们做得足够好看，以便能向 Cursor 描述。所以，我不会说我一上来就用 Cursor。我觉得我仍然需要先把它可视化，这样我才知道如何反过来把概念传达给 Cursor。

Yeah. It's still much faster to whip up a couple concepts. Some of what you're seeing, you know, a floating chat box on your desktop or perhaps like a smaller form factor, more compact. What would it look like to have this, you know, what you just saw, the horizontal scroll, vertical scroll type of form factor, which would evolve, take on different shapes, take on different look and feel as well. And Figma became kind of a collection space for these ideas, for me to make first look decent enough to be able to describe it to Cursor. So, I wouldn't say I start with Cursor immediately. I think there still is the need for me to visualize it so that I know how in turn to communicate the concept back to Cursor.

是的，这完全说得通。

Yeah, that makes total sense.

我想给你看的一些东西也在逐渐成型。我觉得当某个想法看起来有潜力的时候，你就可以想花多少时间就花多少时间去完善它。而我刚才给你看的那些可能感觉有点太 Raycast 风格了。你必须调用它，而且它藏在很深的背景里，以至于人们可能会忘记它。如果有一个持续存在、随时可访问，但又不仅仅是另一个聊天界面的东西，会是什么样子？这基本上就是我在推特上提到的“刘海”概念的由来。我其实不确定你能不能看到上面那个实际的“刘海”。

Some of the stuff I want to show you also kind of was getting built out. I think the moment it felt like some idea had legs to it. Then you know you can spend as much time as you want fleshing it out. Whereas the what I just showed you maybe felt a little bit too ray casty. You have to invoke it and it lives very far in the background to the extent that perhaps people might forget about it. What would it look like to have something that's persistent and ambiently accessible but not overtly just another chat interface? That's kind of what led to the notch concept which I tweeted about. I'm actually not sure can you see the actual notch up here.

哦，是的，我能看到。哦，酷。

Oh yeah, I can. Oh, cool.

我们有 Figma，然后我们实际上把它放在这里。我从一个草图开始，就在 Figma 里。你知道，如果我们有我们的小家伙会怎样，实际上我想先谈谈这个小家伙。这个小家伙最初是一个受光标形状启发的吉祥物。想法是，因为当时它还是一个非常 Cursor 的产品，我们有一个六边形的形状，把 Cursor 的标志性 logo 变成一个互动角色会是什么样子？所以自然的做法就是给它眼睛，做一个非常简单的形状，让它可爱、有响应性。这就是小 Grok Bot 吉祥物的起源。想法是它会非常有像素感或半色调，非常数字化，让它感觉有点书呆子气，有点嵌入你的桌面，然后头脑风暴它可能出现在桌面上的不同方式，无论是通过这个浮动窗口，还是最终通过“刘海”，或者是一个小小的浮动光标。也许你可以点击桌面上的任何东西，让它与之互动。也许它甚至住在角落里。是的，我认为 Figma 仍然是在你开始真正做原型之前，围绕一个想法产生并建立视觉信念的合适场所。

We got Figma and then we got it actually over here. And I started off with a sketch which was just in Figma. It was you know what if we had our little guy and actually I want to talk about the little guy first. This little guy was first going to be a cursor shape inspired mascot. The idea was since it was at the time still a very cursor product, we have a hexagonal shape, what would it look like to turn the cursor iconic logo into an interactive character? And so the natural direction was just give it eyes, make it a very simple shape, make it cute, make it responsive. That's where the little Grok Bot mascot originated from. The idea was that it would be very pixely feeling or halftone, very digital, like make it feel a little bit nerdy, a little bit embedded on your desktop and brainstorming different ways that it could show up on the desktop, whether it be through this floating window or eventually the notch, whether it be a little floating cursor. Maybe you could click onto anything on your desktop and have it interact with it. Whether it maybe lived even in the corner. Yeah, I think Figma is still the appropriate place to generate and build some visual conviction around an idea before you start committing to actually prototyping it.

我觉得这些探索之所以让我印象深刻，是因为我们作为设计师在迭代和探索时，大部分时间都是视觉性的。你知道，我们只是在移动像素，然后思考我们如何呈现这些信息？而这充满了新颖的交互概念，你知道，没有太多先例。是的，有一些，比如 OpenAI 收购的 Sky，还有 Raycast，以及那种无处不在的 UI，但这里面有很多新鲜的东西，伙计。这真的很酷。

I think where I find these explorations so impressive is so much of the time that we spend as designers where we're iterating and exploring is visual. You know, we're just kind of moving pixels around and it's how do we present this information? And this is just full of novel interaction concepts, you know, like there's not a ton of precedent. Yeah, there's some like I've you've got like the sky which OpenAI bought and you have Raycast and some of that kind of omnipresent UI, but there's a lot of fresh stuff in here, man. This is really cool to see.

是的。而且我觉得我刚才给你看的那些，也首先出现在这些探索中。那就是如果我们把事情分解成任务、搜索、协作、已经发生的单个动作，会怎样。

Yeah. And I think even what I had just shown you took place first in these explorations. It was what if we broke things down into tasks, searching, collabing, individual actions that have taken place.

所以一旦我觉得某件事值得投入实际代码，我就去做了。这就是我们走到 Notch 那个角色的原因。一旦你开始构建，我想说，想法就会源源不断。你知道，很多东西你得先和它交互，新的想法才会在此基础上不断涌现。所以最初的想法就像是一个网格，或者你活跃应用的光环，你可以打开它，它现在可能不太管用了，但哦，好吧，它确实管用——当时的想法就是选一个应用，告诉机器人去里面干活，它就会去做。而且大概，你知道，你仍然需要某种聊天界面，这样你可以回去查看它正在处理的所有事情。仍然有一个持久的聊天界面让你能和它交互。

And so once I felt like something was worth committing to actual code, then I went for it. And that's kind of what brings us to the Notch guy. Once you start building, I would say the ideas start flowing. You know, a lot of these things you need to interact with before new ideas start piling on top of that. So what started off as like a grid or a halo of your active apps—you can open it, and it's probably not going to work now, but oh, okay, it does work—the thinking was just choose an app, tell the bot to work in it, and it'll do it. And presumably, you know, you still need some sort of chat interface, so you can go back and see everything that it's working on. Still having that persistent chat surface for you to be able to interact with it.

我们当时的原型内部叫 Sand，我用 Sand 来帮我构建这个 Notch 概念。所以我是在用实际产品来构建我对这个产品理想化愿景的想象。在这个过程中，我意识到现有的产品形态其实非常强大。现在网上有很多讨论说我们需要摆脱聊天界面，而且你知道，即使我抱怨三栏式布局，我想在我使用它的过程中，它变得很清晰，这仍然是我们与一群本质上是智能体的东西交互的正确方式，这些智能体执行事先定义好的任务。尤其是因为，你知道，在我构建这个的时候，我实际上最终把它和真正的 Sand 应用连接起来了。所以我亲自试了试。我试着把这个作为 Sand 的主要形态，发现很难跟踪上下文，尤其是当它不断消失到 Notch 里的时候。快速原型设计的一部分乐趣就在于你早点学到这些教训，然后用它们来指导下一步。

The prototype that we had was called Sand internally, and I was using Sand to help me build this Notch concept. So I was using the actual product to build my idealized vision of what the product could be. And in the process, I realized that the existing product form factor was actually very powerful. There's a lot of discourse online these days about like we need to move away from the chat interface, and you know, even if I complain about the three-column approach to it, I think as I was using it, it became clear this is still the right way for us to interact with a fleet of what are essentially agents carrying out tasks that are defined to them ahead of time. And especially so because, you know, as I was building this, I actually ended up hooking it up with the actual Sand app. And so I tried for myself. I tried using this as the main form factor for Sand and found that it was hard to keep track of context, especially when it kept dismissing into the notch. Part of the game with prototyping quickly is that you learn these lessons early and you use that to inform the next step.

我觉得这两个概念之间另一个有趣的分歧是，一个是给智能体命名，给它一个永久的家，而另一个更像是做所有事情的智能体，它更以工具为中心，而不是以身份为中心。这是不是那种紧张关系的核心之一？

I guess another split that I find interesting between the two concepts is one is very much so name an agent, give it a permanent home versus this, which is more like the agent that does everything and it's more tool focused rather than identity focused. Like, was that kind of one of the main things that was at the heart of that tension?

我想用这个原型捕捉的是，有一个拟人化的智能体，一个工人的概念，它可以在你的机器上做你通常会做的事情。这就是它想要捕捉的。但我认为这也是我们今天试图传达给 Grok Bot 的主要叙事。我认为你说得对，现在这个有点太以应用为中心了。现实是，你知道，机器人可以进入浏览器，做你今天花大量时间做的事情。某些事情比如需要点击进入地图——我不知道，比如我打开了它，但我并没有积极使用它。当我去用它的时候，很多时候我自己做反而更快。

What I was trying to capture with this prototype was that there is this personified agent, this concept of a worker that can do on your machine things that you typically would do. And that's what this was meant to capture. But I think that's also still what we're trying to communicate as the primary narrative for Grok Bot today. I think you're right in the fact that this right now is a little bit too appentric. The reality of it is, you know, the bot can go into the browser and do stuff that you spend a ton of time doing today. Certain things like needing to click into maps—I don't know, like I have it open, but like I'm not actively using it. And when I go use it, there's often times I can do it myself faster.

是的，我觉得这也是一个教训，你知道，这种应用优先的方法可能也不太适合这个。

Yeah, I think a lesson learned as well, you know, that this app-first approach maybe wasn't the most appropriate for this as well.

工程师们打下的原始构建中，智能体有脸吗？比如，你是想把它拟人化吗？那是当时做出的设计决定吗？

Did the original build that the engineers laid the foundation for have agents with faces? Like, were you trying to personify it? Was that a design decision that was made?

Notch 概念的价值主要在于为它设计一个小角色。早期的迭代并没有他。让我看看我是否有一些模型图。我认为早期的迭代——我们看到很多队友给他们的智能体起名字，上传你知道的照片、表情包、AI 生成的缩略图。所以有一个种子，嘿，我想定制或赋予这个为我做任务的角色生命。我认为这最终变成了我们需要默认给它一个角色。你知道，显然在今天的 Grok Bot 中，你仍然可以上传自定义照片作为它的头像，但我认为早期除了你知道，一个简单的圆形首字母之外，没有关于它长什么样的概念。

The value of the Notch concept primarily was in having a little character for it. Early iterations of it did not have him. Let me see if I actually have any mock-ups here. I think early iterations of it—we saw that many of our teammates were giving names to their agents and uploading you know photos, memes, AI generating a thumbnail for it. And so there was a seed of hey, I want to customize or give life to this character that is doing tasks for me. And I think that eventually turned into we need to give character to it as a default state. You know, obviously in Grok Bot today, you still can upload custom photos for it to have as an avatar, but I think early on there was no notion of what it looks like beyond, you know, a simple set of initials in a circle.

好的。所以你到了那个点，意识到这个大致形状至少在方向上是正确的。在那个时候，你会想，好吧，现在是什么样的东西。我觉得有趣的是，你提到的一些事情，你知道，这是一个有点集中的空间，就像我们在一些基本模式上达成一致，我经常在 Twitter 上看到人们抱怨，哦，一切看起来都一样，但这也可能是件好事。话虽如此，你在哪些地方有意地寻找机会去做一些略有不同的事情？

Okay. So you get to the point where you realize this general shape is at least directionally correct. And at that point, you're like, okay, now what kind of thing. I think it is interesting because some of the things that you were hitting on where you know it's a little bit of a concentrated space like we are aligning on some of the fundamental patterns and I see all the time on Twitter people complaining about like oh everything kind of looks the same and it's like well that might be a good thing too. That being said, where were you intentionally circling opportunities to do things slightly differently?

我们现在一致认为现有的形态仍然非常有价值和有用，但我想我还没有完全准备好放弃它。你可以看出，我仍然在推动一些我认为可能有用的想法，除了有一个矩形的聊天框，你知道，那种形态。我认为其中一些想法是，Sand 角色能不能，你知道，出现在你屏幕的角落，对吧？即使你有生命。

We're in agreement right now that the existing form factor is still very valuable and useful, but I think I wasn't quite ready to give that up just yet. And you could tell like I was still pushing on some ideas that I had that felt like they could be useful in addition to having the chat box in a rectangle, you know, form factor. And I think some of those ideas were like, could the Sand character, you know, show up in the corner of your screen, right? Even if you have lives.

是的。这很像 Clippy 那种风格。

Yeah. It's like very Clippy-esque.

而且我认为当时行业里很多人都在探索这个，我自己也是。你知道，感觉这是一个非常合理的时机把它带回来。我在 Figma 里提到的其他事情：如果你有一个角色，你可以直接问它关于窗口的任何问题，对吧？就像我认为仍然有需求，也许市场上有一些空间，让一个本质上就是 Clippy 的变体或特定用例版本住在你的桌面上，无论是否被调用。

And I think a lot of people at the time in the industry were exploring it, and myself as well. You know, it felt like a very reasonable time to bring that back. Other things as I mentioned in the Figma: what if you had this character that you could just then ask any questions about a window to, right? Like there's still, I think, an appetite and maybe some space in the market for a variant or a use-case-specific version of what is essentially Clippy to live on your desktop, whether invoked or not.

你展示的很多东西都非常新颖。比如那些卡片弹出的方式，窗口上的选中状态——你触及的模式，我觉得不是那种——这不是设计师每天都会看到的东西。跟我谈谈你自己的构思过程，比如当你像一张白纸一样思考我们该做什么？那个行为是什么样的？就像帮我们做一只墙上的苍蝇，如果你只是坐下来探索一些这些东西。

A lot of the things you're showing are really fresh. Like even the way that those cards popped up, the selection state on the window—like you're reaching for patterns that I think are not the—it's just not the things that you would see every day as a designer. Talk to me a little bit about your own ideation process, like when you're just kind of blank slate thinking about what do we do? What does that act look like? Like help us be a fly on the wall if you were just kind of sitting down to explore some of this stuff.

我认为事实是，当你日复一日地被限制在一个已经非常健壮的现有界面上——在这种情况下是 Cursor——我认为你会寻找机会来表达它的反面。Cursor 作为一个工具，几乎就像一件精密仪器，你知道，它的设计方式非常响应迅速，非常以任务为导向，所以没有太多空间容纳乐趣或表现力。我认为这就是你在这里看到的——很多这些探索更多是关于我插入动感的机会。愉悦。

I think the truth of it is when you're restricted day-to-day to working on an existing surface that is often times very robust already—in this case Cursor—I think you look for opportunities to express the opposite of that. Cursor as a tool is like almost like a precision instrument, you know, it was designed in a way to be very responsive and very much task-oriented, and so there's not a whole lot of room for fun or expressiveness. And I think that's kind of what you're seeing here—a lot of these explorations are more about opportunities for me to insert motion. Delight.

也许这个词现在有点老套了，但我认为当你的目标受众是那些期望得到某种东西、并且愿意为非常具体的东西付费的工程师时，探索阶段就成了设计师尝试一些更一次性概念的机会。我认为你需要抱着完全抛弃它们的预期去尝试，但要去试，要玩得开心。我觉得早期阶段非常适合这类工作。一旦你们在方向上达成一致，跟我聊聊团队形式吧，比如都有谁在做这个？你们是怎么合作的？给我们讲讲作为设计师实际参与这个项目的背景。

Maybe that's a played out word at this point, but I think when your target audience is engineers that expect something, they pay for something very specific. The exploratory phase becomes the opportunity for a designer to try out some of these more throwaway concepts. And I think you need to go into it expecting to fully throw them away, but to try it and to have fun with it. I think the early phases are very ripe for that type of work. Once you all were aligned on the direction, talk to me a little bit about the team form like who all is working on this? How did you work together? Give us the context for like what it was like actually as a designer being staffed on this project.

如果你说的“在方向上达成一致”是指我承认那些更有趣的概念失败了，那么是的，绝对如此。但话说回来，我愿意认为这些早期的探索和其中的乐趣最终融入了最终产品。大约在这些探索被分享、我在观察哪些有效哪些无效的时候，每个对应用某个角落感兴趣的设计师都自然地各就各位了。事实上，到最后，总共有五位设计师在积极投入，冲刺把它带过终点线。当我尝试一些更新颖的形式和交互时，Pong、Keith、Tyler、Mamuso 他们加入进来，发挥各自的优势。我觉得当我在摆弄这些其他东西时，其他设计师在努力打磨最终成为应用外壳的部分。而且，你知道，促成其成功的最大因素之一可能是品牌团队。我们的品牌设计师做了大量的迭代，我觉得分享他们一路走来的所有探索不太合适，但很长一段时间我们都不知道这个应用会叫什么名字，所以我们一直在添加占位符。对我来说，有占位符这件事让整个过程变得非常有趣。

If what you mean by like a line on a direction, you mean me admitting defeat to some of these more fun concepts, then yes, absolutely. But once again, I like to think that some of these early explorations and the delight made their way into the final product. Around the time that these explorations were being shared and I was observing what was sticking and wasn't, each individual designer that was interested in pursuing some corner of the app kind of fell into place organically. In fact, towards the end, there were a total of five designers actively working on it, sprinting on it to get it across the finish line. While I took the approach of trying out some of the more novel forms and interactions, we have Pong, Keith, Tyler, Mamuso who stepped in and essentially played to their strengths. I think while I was kind of tinkering with these other things, the other designers were working hard at polishing what eventually became the final shell of the app itself. And also, you know, probably one of the biggest factors that led to the success of it was the brand team. Our brand designers had a ton of iterations and I wouldn't feel right sharing all the explorations that they went through along the way, but for the longest time we didn't know what the app was going to be called and so we just kept adding placeholders. For me, I think having placeholders was something that made it really fun.

我来分享一些本质上是一系列引导流程概念的东西。让我从第一个版本开始，它叫 Astra，你会看到不同的 logo，因为每次品牌团队做了新的修订，我都会想，哦我喜欢那个，我要把它放在这里，以防它成为最终版本。第一个概念是，嗯，让我们从“你是谁”开始。对我来说，比如我是 John。也许这个工具会查一下你，做一些背景调查，然后告诉你，嘿，我们知道你是谁，这些智能体可以这样帮你。你是这种类型的人。如果不对你可以改，但先认识一下你的智能体。这有点回到你之前的问题，就是这些机器人和智能体是否有更多身份？这是我们想表达它的一种方式，那就是如果我们直接给你看所有用例会怎样？你得到

I'll share a little bit about what was essentially a series of onboarding concepts. Let me start with kind of the first version which was called Astra and you'll see a different logo because every time the brand team did another rev, I was like, oh I like that one, I'm going to place it here in case it becomes the final thing. And one of the first concepts was, yeah, let's just start with like who who are you? And so for me, like I'm John. Maybe the tool will look you up, do a little background digging and then let you know, hey, we know who you are and this is how the agents can help you. You're this type of persona. Maybe you could change it if it's incorrect, but meet your agents. And this was kind of going back to your question about, you know, was there more identity to these bots and agents? This was one way that we wanted to express it, which was what if we just show you all the use cases right? You get the

是啊，这很酷。

Yeah, this is cool.

你有点像在组建一个团队，如果你想浏览所有智能体，就像这样，给你，对吧？这些是占位符，或者你可以创建自己的。我们有些想法没有进入最终版本，但比如，你知道，如果你想非常务实和详细，连接一些技能然后开始，对吧？

You're kind of assembling a team if you want to browse all of them like here you go, right? These are the placeholders or you can create your own. Some of the ideas we had didn't make the final cut, but it's like, you know, what if you wanted to be really pragmatic and verbose and connect some skills and then get started, right?

连那个占位符对小头像来说都很有趣。

Even that placeholder is fun for the little avatar.

我喜欢它。

I love it.

这是一个想法。另一个是，我们能不能为你量身定制一个智能体，而不是向你展示所有可用机器人的广度或菜单？如果你连接 Google，分享一些个人数据，然后它就像，嘿，我们正在了解你。这也是我想引入更多动效的地方。这背后的想法是，也许我们不知道所有信息，所以让我们慢慢让它进来，这样 AI 就像在思考和处理，

This was one idea. Another one was, can we tailor an agent for you instead of showing you the breadth or the menu of bots that are at your disposal? What if you connect with Google, share your personal data a little bit, and then it's like, hey, we're learning about you. And this is also where I wanted to introduce a lot more motion. The idea behind this was that maybe we don't know all the information, so let's slowly have it come in so that the AI is like thinking and processing,

争取时间。

Buying time.

那很好。那很好。

That's nice. That's nice.

最后也是一样，但就像，嘿，这来了，就像他在依附你，对吧？他成为 Maya 的助手，从工具中学习，也许你会在收件箱里收到某些事情的通知。所以就像，嘿，我们又一次在争取时间，这个设计是作为一个单一的智能体，你将从这个智能体开始，或者你可以，你知道，再次创建一个，一旦你创建了，它就会是你与之互动的第一个东西。

And same thing in the end, but it's like, hey, this is coming, like he's attaching to you, right? He's being Maya's assistant, is learning from the tools, maybe like you get notifications in your inbox for certain things. So it's like hey, we're buying time once again, and this one was meant to be a single agent that was going to be the one that you started off with, or you could, you know, once again create one, and once you did, it would be the first thing that you interact with.

哦，那真的很好。

Oh that's really good.

这是一个叫 dot 的迭代版本，非常相似，我认为很多这些设计都取决于我们先了解你，但我觉得这个版本更侧重于解释这个应用是如何工作的。

This one was an iteration called dot, where very similarly, I think a lot of these hinged on the fact that we would learn about you first, but I think in this one it was more about explaining how this app works.

嗯。

Mhm.

你可以看到，你知道，仍然有这些插画式/AI 生成的头像，但它们正在被命名，对吧？我们让你对这些东西如何工作有所了解。而且我在动效上玩得很开心，对吧？每个机器人都在使用自己的电脑。最后，你知道，它们可以在自动驾驶模式下运行这些任务。

And you can see, you know, there's still kind of these illustrated/AI generated avatars, but they're being named, right? We're giving you an inkling of how these things work. And I'm having tons of fun with the motion, right? Each of these bots is using their own computer. And finally, you know, there are these tasks that they can run on autopilot.

关于那个动效画面，快速问一下。你是先在 Figma 里把它做出来吗？把它带进来，然后描述你希望东西如何移动，就这样吗？

Quick question on that motion screen. Are you just bringing like, did you shape that in Figma first? Bring it in and then describe how you want things to move and that's that.

这个我主要用 Figma 来起草故事板。这是第一轮，就像首先你告诉我们你做什么。我们会了解你，然后我实际上在 Figma 里布置了动画概念，然后我把它喂给了当时的 Sand。我把它喂给了 Grok Bot，只是描述了我想怎么做，它就在 cursor 里为我做了，最后我完全跳过了 cursor，直接让它在浏览器里生成选项，这个我一会儿可以详细说。所以我当时做的，是的，基本上就是布置故事板，你知道，每一轮都在变得更精细。我觉得方法一直在演变。你会看到更多，比如有一个阶段品牌团队在做这种太空侵略者风格的头像，所以那成了引导流程的一部分。你知道，如果所有这些都出现会是什么样子，这些都是 Grok Bot 在 Figma 里为我生成的。我只是给了它一个 notion 文档，然后说，现在把它们都布置出来。同样，你知道，你会看到这些角色开始在这个流程中成为核心。有些现在坏了，但是的，甚至到选择你的第一个队友的第一个版本。

This is one where I would kind of use Figma mainly for drafting like storyboards. This is round one, which was like first you tell us about what you do. We're going to learn about you and then I actually laid out kind of the animation concept in Figma and then I fed it to Sand at the time. I fed it to Grok Bot and just described how I wanted to do it and it did it for me first in cursor and then eventually like I skipped cursor altogether and just had it generate options within the browser and I can speak to that in a second. So what I was doing, yeah, essentially was laying it out having storyboards, you know, they were being more refined with every round. I think like the approach just kept evolving. You'll see even more like there was a point at which the brand team was doing these kind of, you know, space invadery type of generated avatars and so that was becoming a part of the onboarding flow. You know what might it look like to have all of these and these were all generated by Grok Bot in Figma working in Figma for me. I just gave it a notion doc and was like here now lay them all out. Likewise, you know, you see these characters start to become central in this flow. And some of these are broken now, but yeah, even down to like the first version of what it looks like to choose your first teammate.

我从这次对话中得到的一个收获是，我觉得人们很容易假设，在非常前沿的 AI 公司工作的人会第一个说：“是啊，我不再用 Figma 了。”但你其实用了很多 Figma，你知道，这挺酷的。我觉得这真的是一个我很高兴能捕捉到的视角。

One of my takeaways from this conversation is that I think it's easy to assume somebody working at very AI forward companies would be the first in line to say, "Yeah, I don't really use Figma anymore." But you're doing a lot of Figma, you know, like this is kind of cool. Like I think that's like a really it's a it's a perspective that I'm glad to capture.

我觉得 Figma 最近名声不太好。我认为它仍然是一个非常有用的草图工具。而且我觉得尤其是现在智能体能够操作它，很多烦人的步骤，更不用说重命名图层了，比如布置每一个变体，你知道，有些东西需要大量打字，我不需要再那么做了，因为我只要把链接给我的机器人，它就会按照我描述的方式去做。甚至这个其实这里坏了。我看看能不能找到另一个。

I think Figma gets a bad rap these days. I think it's still a very useful sketching tool. And I think especially now with agents being able to operate it, a lot of the annoying steps, like not even mentioning renaming layers, but laying out every variant of, you know, some of the stuff that you have to do a lot of typing for, I don't have to do that anymore because I just give the link to my bot and it'll do it according to how I describe it. Even this one actually is broken here. I'm going to I'm going to see if I can find another.

嗯，这些看起来都像是……好吧，我觉得这些只是浅色模式，但我基本上说了同样的话。我说：“嘿，从 Notion 文档里把所有在线 logo 拉出来，填满这个组件网格。”

Well, these are all like Okay, it was I think these are just like in a light mode, but I I basically said the same thing. I was like, "Hey, pull up from a notion doc all the logos online and fill in this this grid of components."

嗯，这种事情，你知道，以前你实际上必须——我不知道其他人怎么做，但我会去 Google 搜索 logo，希望它是高质量的 SVG 之类的，然后把它拖进 Figma，调整大小，放到合适的位置，现在你完全不用做这些了。

Um it's type of thing where, you know, before you had to actually what I would do, I don't know about the rest of people, but Google the logo, hope it's like quality SVG or something, and then just drop it into Figma and then resize it, put it in place, and now you don't have to do any of that.

有一个问题我忍不住一直问自己：如果公司反过来申请和你对话，而不是你去找他们，会怎样？这个问题正是全新 Dive Talent Network 的基础。而且它正在奏效。比如现在，我正在帮助我认识的最令人兴奋的初创公司，雇佣那些收听本节目的设计师和开发者。所以，如果你好奇可能有什么机会，或者想加入我的名单，甚至可能在寻找你的下一位设计人才，请访问 dive.club/talent 今天就加入。

There's one question that I can't stop asking myself. What if companies applied to talk to you rather than the other way around? And that question is the foundation for the all-new Dive Talent Network. And it's working. Like right now, I'm helping many of the most exciting startups that I know to hire the designers and builders who listen to this show. So, if you're curious what might be out there and maybe you want to get on my list, or maybe you're even looking for your next design hire, head to dive.club/talent to join today.

我想现在深入探讨的一个话题是，我们能不能从原则层面聊聊 onboarding？你显然在深入思考什么造就了优秀的 onboarding 体验，你在 Cursor 也谈过这个。你现在有大量探索，而且都很好。这真的是很有意思的工作。我一直在思考这个张力：所有 AI 工具都把你的所有账户前置，但你也需要传达价值，而你把这些融入一个步骤的方式我觉得非常迷人。所以跟我们聊聊，当你坐下来设计 onboarding 流程时，你追求的高层原则是什么，这些原则如何塑造了你展示的这些成果？

I think a rabbit hole that I want to duck down now is can we just like talk at a principal level about onboarding? Like, you're obviously thinking so deeply about what makes for a an excellent onboarding experience and you talked about it with cursor. you're now tons of explorations and they're good. Like it's really interesting work. I I I'm very much so thinking about this tension where all of the AI tools went all of your accounts up front, but you also kind of got to like communicate the value and the way that you kind of wound those into a single step I think is very fascinating. So talk to us about like what are the highlevel principles that you are reaching for when you're sitting down to design an onboarding flow that shape some of this output that you're showing?

你知道，我喜欢开玩笑说，我在 Cursor 的整个时间都在设计 onboarding。这是真的。实际上这是我加入 Cursor 时做的第一个项目。节奏变成了，一开始是我的全部注意力，思考如何让用户上手 Cursor。当我开始的时候，就在我们发布 GlassS 一周后，我们必须让用户上手这个新的智能体窗口，那些以前主要在 IDE 里的用户，我们必须解释为什么智能体窗口是更好的工作方式。我们做了大量调查，比如让我们想想，为什么这比 IDE 更好用。最终变成了一系列实验，我有很多不同的概念，关于我们如何讲述为什么智能体窗口强大，既要讲给现有的工程师用户，他们真的不想要太多改变，也要讲给过去的我——一个对 Cursor 好奇的设计师，听说过很多但不太知道如何开始。我认为带着这个想法，我清楚地意识到，你一开始讲述的故事极其重要，而且你大概只有一两次机会给用户留下第一印象。你知道，我以前认为你只有一次第一印象，但现在人们会说：“哦，我回到了 Cursor。”所以，你知道，这不完全正确，但第一印象非常重要。

You know, I like to joke that my entire time at Cursor, I've been designing onboarding. And it's true. It was actually the first project that I did when I started at Cursor. The cadence became like it started off as my full attention. Uh thinking of ways to onboard people onto cursor. When I started like the week after we launched GlassS and we had to onboard people onto this new agents window, people that were once in the IDE primarily, we have to explain the value of why the agents window is a better way to work. And it took a lot of investigating us to like let's think about actually why is it better to use this compared to the IDE. It eventually became a lot of a series of experiments where I had a bunch of different concepts around how might we tell the story of why the agents window is powerful both to an existing engineer user who you know really doesn't want too much change and for myself in the past a designer that just is curious about cursor heard a lot about it but don't quite know how to get started and I think with that in mind it became clear to me that like the story that you tell up front is immensely important and you also kind of only get one maybe two first impressions for a user. You know, I used to think you only get one first impression, but these days people are like, "Oh, I came back to cursor." So, you know, that's not really true, but the first impression matters a ton.

不是一定要做比较，但大约在这个时候，Buzz 正在发布，我经历了那个 onboarding 流程，我觉得他们试图传达太多概念了。我给其中一个参与设计的设计师发短信说，我不确定我能完成这个 onboarding 流程，不是因为它步骤太多，而是因为它试图做太多事情。这是我内心深处相信的一点：如果 onboarding 步骤有吸引力，并且能让你清楚地看到你所 onboarding 的东西的价值，人们是会愿意走完这些步骤的。这就是为什么我一直告诉人们，重点不是添加跳过按钮，也不是让你一路点击直接进入产品，因为现在有了 AI 工具，你进入产品后完全困惑，不知道从哪里开始。Cursor 就是这种情况。如果你跳过 onboarding 流程，你会被丢进一个空状态，只有一个空白的 composer，这不是一个好地方。更重要的是，在你看到第一个起始屏幕并遭遇行动瘫痪之前，先确切了解这个工具能做什么以及如何使用它。

Not to like draw comparisons necessarily, but it was around this time that um Buzz was launching and having gone through that onboarding flow, I think they tried to communicate way too many concepts. I was I was texting one of the designers that worked on it and I was like, I I don't know that I'm going to get through this onboarding flow, not because it has too many steps, but because it's trying to do too much. That is one of the things that I kind of believe in my core which is people will go through onboarding steps if it's engaging and it makes the value of what you're onboarding onto abundantly clear upfront which is why I tell people all the time like it's not about you know adding a skip button. It's not about letting you just click through click through and jump into the product because these days with AI tools you jump into the product and you're totally confused. You don't quite know where to start. And that was the case with cursor. If you skip the onboarding flow, you're dropped into an empty state with just a blank composer, which is not a good place to be. It is much more important for you to learn exactly what this is capable of and how to use it before you're introduced to that first starting screen and you get hit with, you know, action paralysis.

我喜欢你用引入的概念数量而不是步骤数量来定义 onboarding 的规模。这很迷人。

I like that you defined the size of onboarding by not the amount of steps, but the amount of concepts that are being introduced. That's fascinating.

对我们来说，核心概念就是这三个步骤，这也是我们今天 onboarding 的步骤。我们可能会打磨它。我在这里的整个时间都会致力于 onboarding。但想法就是，是的，你有一个团队。不只是一个智能体，不只是一个机器人在做事。而是一个团队，每个成员都有自己的特定任务。我认为与其用一个冗长的标题来说明这一切，不如直接展示。你只要说“智能体团队”，然后通过命名展示它们有特定任务。你一开始就介绍 artifacts、时间安排，甚至群聊。这个很迷人。我觉得这有点难展示，但这时候动效就特别有用，机器人自己移动鼠标，窗口出现又关闭，甚至可能显示一个 logo。这样你就明白，无论这个智能体是什么，它都在做事情。最后，这个更多是关于你可以自动化这些事情。对我来说，这是最不有力的信息，但我觉得这也是我们在内部看到的最常见的用例之一。

The concept for us at its core was these same three steps, and there are the steps of our onboarding today. We're we're probably going to polish it. I'm I'm going to be working on onboarding my entire time here. But um the idea was just Yeah. Yeah. You have a you have a team. It's not just one agent. It's not just one bot doing things. It's a team that each has its own specific task. And and I think instead of one long headline that says all that, you can show it. You just say team of agents and you show that they have specific tasks by the naming. You introduce them off the bat to artifacts, timing, and maybe even group chats. This one is fascinating. And I think like it's a little bit hard to show, but this is where motion comes in especially in handy, which is the bots are kind of moving the mouse on their own and windows are appearing, windows are closing, maybe even a logo shows up. And so you get the idea that whatever this um agent is, it's it's doing something. And finally, like this one was more about you can automate these things. I think to me this was the least powerful of the messages, but I think it's also one of the most common use cases that we were seeing internally.

所以在让他们进入那个空屏幕之前，值得先强调一下，你知道，哦，这个没被选中，但它是“介绍一下你自己”，这样我们

And so it was worth highlighting before jumping them into, you know, oh, this one didn't make the cut, but it was tell us about yourself so that we

酷。

Cool.

是的。这样你进入第一个空屏幕时就能为成功做好准备。

Yeah. Set you up for success once you're in that first empty screen.

有没有一些你很想添加、但为了保持简洁而最终砍掉的东西？

Were there things that you felt a temptation to add that you ended up cutting in order to preserve that simplicity?

就是这个。这是一个屏幕。

This was it. This is one screen.

这是其中之一。我想我仍然强烈地觉得，如果我们能通过做一些前置集成来向你学习，那可能会更强大。但说实话，我觉得这两个（被砍掉的功能）各有原因。第一个是，这个我觉得有点复杂，而且我们还没有建立信任，所以我们不确定人们会连接起来。他们可能只是——显然有后备方案，但你懂的，我理解。对于这个，诱惑可能更大，因为那周正好是 ChatGPT 和 Claude 都发布了语音模式，所以我们就想，酷，我们也应该做这个。最后，它并没有特别有用。人们不知道该说什么，因为他们没用过，也不知道该让它帮忙做什么。所以你可以想象，它的位置完全不对。

This was one of them. I think I still feel strongly that if we were to learn from you by doing a little, you know, integration up front, it could be more powerful. But, you know, honestly, I think these two for various reasons. The first one is this one I think was a little bit more complicated and we hadn't established trust and so we weren't confident people would connect. They might just get obviously there's fallbacks for that, but you know, I understand that. For this one, the temptation was probably more so this was the week that both ChatGPT and Claude were launching their voice mode and so we were like, cool, we should do that, too. In the end, it did not feel particularly useful. People didn't really know what to say because they hadn't used it and they didn't really know what to ask it for help with. So you can imagine the placement of it was just all wrong.

就我个人的体验而言，我的顿悟时刻是看到那个小电脑。比如每个智能体都有一台电脑，这个概念上我知道那是存在的。但到目前为止，对我来说，所有云电脑的实现都非常不透明，你只能相信那里正在发生什么。但你们对嵌入式窗口和接管按钮的思考方式，以及像说“好了，我完成了”这样的设计，我喜欢那个用户体验。我绝对喜欢那个用户体验。

For what it's worth, my own experience, the thing that was the aha moment for me was seeing the little computer. Like the idea of each agent having the computer and conceptually I knew that that was a thing. But every implementation of cloud computers up to this point for me has just been very opaque where you kind of just has to like trust that something's happening there. But the way you all thought about the embedded window and the takeover button and like the saying like, "Okay, I'm done." I love that UX. I absolutely love that UX.

是的。我也想给你看看它最终的样子，对吧？就像这是

Yeah. And I'd love to show you also kind of the final form it took on, right? Like this is

请

Please

这是版本一，就像你先从一个任务开始，然后是的，同样的东西。所以，你知道，你看到的那个版本是三步的，我觉得我们最终的样子非常相似，只是在这个版本上做了些微调。你知道，我们一开始就告诉你这里有多个机器人。这些五颜六色的角色，你开始使用。它仍然是一个熟悉的界面，你有你的作曲家，一个小角色，当你发送时，它就开始思考。第二步仍然是相同的电脑使用，只是在这种情况下只有一台电脑，因为感觉三台有点杂乱。然后在这里，它分支到各种任务。最后，就像你每天用什么？你知道，这些家伙在周围盘旋，你选择几个工具，它就开始帮你创建你的第一个机器人，这大概就是你开始时交互的流程。

this is the version one where it was like then you start with a task and then Yeah. same thing. So, you know, you saw the version that, you know, was three steps and I think where we ended up took a very similar shape just with mild changes in this one. you know, we're teaching you that there's multiple bots right off the bat. There's these colorful characters and you get started. It's still a familiar interface where you have your composer, a little character, and when you send it, you know, it starts thinking. The second step is still the same computer use and just one single computer in this case because it felt like the three was a little bit cluttered. And then this is where, you know, it branches off into various tasks. And finally, it's like what do you use every day? you know, these guys are kind of hovering around and you select a couple of tools and it gets you started with creating your first bot, which you know, this is probably the flow that you interacted with when you started.

引导探索非常有价值，因为它本质上是在练习如何提前传达价值，然后确保产品的其余部分在整个用户体验中承载这种价值。有一个概念，Colin Dunn，我不知道你是否熟悉他，他是 Visual Electric 的联合创始人，一位伟大的设计师，他谈到了“知情简洁”这个概念，以及当你使用一些感觉非常明显的东西时，实际上需要先广泛探索，尝试所有的事情，才能缩减成某样东西，让用户使用时会说：“是的，当然是这样工作的。”你知道，那里没有太多设计。我喜欢这次对话的一点是，人们打开产品，可能感觉非常明显，但看到你为了塑造和调整产品而拉动的每一条细线，正是我在这个节目中最喜欢捕捉的东西。所以，感谢你揭开帷幕，展示这一切。

The onboarding explorations were super valuable because it was an exercise in essentially communicating the value up front and then ensuring that the rest of the product carried that value throughout the user experience. There's a concept that Colin Dunn, I don't know if you're familiar with him, he was the co-founder of Visual Electric, a great designer, and he talked about this concept of informed simplicity and how when you use something that feels incredibly obvious, it actually takes going really wide first and exploring and trying all of the things in order to whittle down into something where somebody uses it and it's like, "Yeah, of course that's how it works." You know, there's not a lot of design there. And something that I love about this conversation is people open the product, it probably feels very obvious, and yet seeing just every little strand that you are pulling on to help shape and tune exactly what ship is like my favorite thing to capture on this show. So, thank you for pulling back the curtain and showing all this.

很大一部分原因在于团队结构，以及我们这些在设计领域协作的人都在发挥各自的优势。信不信由你，早期有过这样的对话，我说，我们不能有这些，我们不能有和 Cursor 一样的按钮。我们最初的设计是尽可能多地复用 Cursor 的组件。我强烈地觉得，如果我们走向更消费级的方向，像 Cursor 会让这个产品感觉像 Cursor，并吸引同样的受众、同样的用户类型。剩下的用户不会来，因为它看起来像 Cursor，而他们已经会受到这种认知的影响。我们有过关于移除某些功能、引入某些交互的辩论。我们有过 jam，每个人都带来了自己的版本，关于文本是流式显示、淡入还是完全不显示，或者与智能体电脑交互的不同版本。我认为这就是团队和互补技能仍然如此重要的地方。当我去探索引导流程并投入大量时间时，其他设计师在打磨应用的其他方面。这确实表明，这个产品不是由一个孤独的设计师或工程师完成的。即使它可能始于少数几个人，我们从构建内部产品市场契合度中获得的每一个信号都被应用到了最终产品中。我们非常批判性地使用它。我们非常批判性地看待什么有效，什么无效。在 Cursor 内部，我们有很强的“取消发布”文化。我真的很高兴我们有，因为这帮助我们移除很多复杂性，专注于 Grok Bot 应用本身的优势。

A big part of it was very much due to the team structure and the fact that each of us that were kind of designing in the space collaboratively were playing to our own strengths. Believe it or not, there were conversations early on about I was like, we can't have these we can't have the same buttons that we have in Cursor. We started off with a design that was trying to reuse as much of the componentry from Cursor as possible. And I felt very strongly, you know, if we're going to go in this more consumer direction. being like cursor will make this product feel like cursor and will attract the same audience, the same types of users. The rest of the pie will not be coming to this because it looks like cursor and they're already going to be kind of affected by that perception. We've had debates about removing certain functionality, introducing certain interactions. We had, you know, jams where everyone was bringing to the crit their own version of what it looks like to have text either streaming or fading in or not at all or, you know, different versions of interacting with the agents computer. I think that's where the team and having complimentary skill sets is still so important. While I went to explore onboarding and focused quite a bit of time on it, other designers were going to polish certain other aspects of the app itself. And it really shows this is a product that wasn't just a lone designer or a lone engineer. Even though it might have started off in kind of just a few players, every single signal that we had from building internal product market fit was applied in what became the final product. We were very critical about using it. We were very critical about what was working and what wasn't. And within cursor, we have a strong culture of unshipping. And I'm really glad we do because that helped us remove a lot of the complexity and focus on the strengths of the Grok Bot app itself.

我们能更深入地谈谈在 Cursor 做设计师是什么感觉吗？我相信很多像我一样的人都很好奇你们是怎么运作的。你在那里待过。这是你的第一年。所以我想应该还有新鲜感。你注意到哪些事情让 Cursor 的设计实践变得独特？是的，我真的很想了解其他大型实验室，因为我一直告诉人们这个。Cursor 设计团队里可能没有两个设计师有相同的过程，也没有相同的工具应用。你知道，我认识一个设计师，他非常积极地使用纸张。我甚至没试过。但我也知道，即使在 Cursor 内部，甚至在 Figma 内部，我们使用它的方式也非常不同。

Can we go a little bit deeper into what it's like as a designer at Cursor? Like I'm sure a lot of people like myself are curious just how you all operate. You've been there. This is your first year. So there's still a freshness I would imagine. What are some of the things that you noticed that made the way that design is practiced at Cursor unique? Yeah, I'm really curious about some of the other large labs because I tell people this all the time. There's probably no two designers on the cursor design team that has the same process and has the same application of tools in their process. You know, I know one designer who's using paper like pretty actively. I've never even tried it. But I also know even within cursor and even within Figma, the way that we're using it is very different.

有相似之处，但每个人都有自己的流程，用一套非常相似的工具各有各的用法。就像我说的，我花很多时间在 Figma 里画草图，然后放进原型里，再摆弄、微调，真正让它感觉对、看起来也对。但我也认识一些设计师，他们直接从代码开始——直接跳进代码，边做边调，他们很清楚自己想要什么，所以完全跳过 Figma。有时候我会去找 Maya——她是做云智能体的设计师——要 Figma 文件，因为我们需要把它放进某个演示文稿里。我想调整内容好放进文稿，她告诉我没有 Figma 链接，只有一个 Vercel 链接，我可以去跟那个真实原型交互。所以这成了一种既有趣又好玩的协作方式。我也变得非常擅长在 Figma 里快速重建东西。我们的工作方式，以及我们能从彼此流程中学到的东西，让 Cursor 的设计变得非常独特。

There are similarities, but each person has their own process and their own way of working with a very similar set of tools. As I said, I spend a lot of time sketching things out in Figma before dropping it into a prototype, then playing with it and fine-tuning it, really making it something that feels right and looks right. But I know some designers who start with the code—they jump right into the code, making tweaks, and they know exactly what they want to do, so they skip Figma altogether. There are times when I'll go to Maya, a designer working on cloud agents, and ask her for the Figma because we need to put it in a deck somewhere. I want to adjust the contents so we can put it in the deck, and she tells me there's no Figma link—there's only a Vercel so I can go interact with this real prototype. So that's become both an interesting and fun way of working. I've also gotten really fast at recreating things in Figma. The way we work and the lessons we can learn from each other's process make design at Cursor just so unique.

这对你们的协作方式有什么影响？因为我觉得你描述的情况——我自己最近也遇到过——我们以前是大家用完全相同的文件格式，在一个协作式网页环境里，你点一下某人的头像就能看到他在做什么；现在呢，大家都分散在各处。跟我聊聊你们怎么做反馈、怎么提想法、设计评审文化是怎么运作的。Cursor 的设计水平到底是什么样的？

How does that impact the way you guys collaborate? Because I think you're describing a situation that I know even myself—I've faced it too recently—where we've gone from everybody in the exact same file format in a collaborative web environment, where you can just click on a person's face and see exactly what they're working on, to now when everybody's kind of everywhere. Talk to me a little bit about the way you guys do feedback, the way you pitch ideas, the way that crit culture works. What does that level of design at Cursor look like?

我觉得每个想法背后都有一个原型——一个可用的原型，你不仅能在评审时在大屏幕上展示，还能分享链接，让其他人都能交互，自己判断用起来对不对。这个团队的运作方式非常追求高保真。你很少会看到半成品。而作为一个喜欢把很多概念做到半熟的人，这逼着我在分享之前先真正面对这个东西到底行不行。我觉得，不管是好是坏，这已经成了我的流程。我说“坏”是因为人们总说好的文化是尽早分享、经常分享。但我觉得这个行业的门槛整体提高了。静态的走查稿已经不够用了。也许这样更好，因为这样你就不会浪费时间打磨一个烂东西。你一眼就能看出它能不能引起团队的共鸣。

I think every idea is backed by a prototype—a working prototype that not only can you present on the big screen during a crit, but you can share the link and everyone else can interact with it and decide for themselves whether it feels right to use. The way this team operates is very high fidelity. Very rarely will you see something half-baked. And I think as somebody who likes to partially bake a lot of my concepts, it's forced me to really confront whether something works before I share it. And I think, for better or for worse, that has become my process. I say for worse because people always say a good culture is one where you share early and often. But I think the bar in general for this industry has been raised. I think a static walkthrough kind of doesn't cut it anymore. Maybe that's for the better, because then you don't spend time polishing a turd. You can see right off the bat whether it's something that will resonate with your team or not.

所以确认一下，你在 Figma 里展示的那些工作，基本上没有其他人真正在看。它们只是素材、故事板，是你喂给模型用来生成最终成品的东西，然后你分享的是那个成品。是这样吗？

So just to be clear, all that work that you showed in Figma—pretty much no one else is really even looking at it. It's just assets, storyboards, things that you are then feeding to the models to create the artifact that you are then sharing. Is that accurate?

对，就是这样。我觉得这个世界变得完全是一次性的。

Yeah, that's accurate. I think the world becomes totally throwaway.

这太不一样了。

That's so different.

我很喜欢聊这个，也喜欢发推分享一些过程性的东西，因为我会想，嘿，我花了时间在上面。我想让人们看到这个没进最终版本的概念，但也就仅此而已。我不再会说，我们需要——我在这上面投入了时间，它必须以某种方式进入最终版本。我觉得我们正处在一个你能做太多事情的时代，所以很多“太多”都被扔掉了。

I love to talk about it, and I love tweeting and sharing some of the process stuff because I'm like, hey, I spent time on it. I want people to see this concept that didn't make the final cut, but that's the extent of it. I'm no longer going to say, we need—I put my time into this, it needs to make the final cut somehow. I think we're just at a point where you can do so much, so a lot of that so much gets thrown away.

这让我想起 Cleo 的另一句话。他说早期设计就像堆沙堡，你得接受它们经常被冲走。我就想，你刚才带我看了大概 20 个超酷的沙堡。

It reminds me of another quote from Cleo actually. He talked about how designing at an early stage is like making sand castles, and you just have to be okay with them getting washed away frequently. I'm like, you have like 20 really cool sand castles that you just walked me through.

所以我觉得，看到它的一点一滴在最终交付物里发光，才那么重要。就像你说的，最初那个两只眼睛的六边形小人——那不是一个美术方向。那是一颗种下的种子，最终长成了品牌创意的一部分。能参与其中感觉很好。我觉得你不能再对自己的概念太宝贝了。

And I think that's why seeing bits and pieces of it shine in the final deliverable matters so much. Like you said, the initial hexagonal guy with two eyes—that's not an art direction. That was something that was a seed that was planted that made its way into the final brand idea. And it feels good to be a part of that. I think you just can't be precious about your concepts anymore.

你是纽约办公室的第一位设计员工。你们之后又招了，我记得你说大概四五位设计师。我猜以目前的增长速度，你们还会继续招人。那么跟我聊聊，基于现在所有变化——产品覆盖面、你们的工作方式——你会优先看重理想设计协作伙伴的哪些特质？有哪些东西会让你觉得“对，我必须把这个人招进团队”？

You were the first design hire in the New York office. Y'all have hired, I think you said, like four or five other designers since then. I'm assuming at the current growth, you're going to continue to hire. So talk to me a little bit about the traits that you would prioritize in the ideal design collaborator, based off of everything that's changing, the product surface area, how you all are working. What are some of the things that would get you to the point where you're like, "Yeah, I got to have that person on the team."

我觉得在现阶段，基本功比很多其他因素都重要得多。我觉得人们往往倾向于看重一个人干活多快、工具库里有多少种工具、会用多少技能、或者对模型的优化策略等等。但我觉得作为设计师，流程在某些方面非常不同，但在很多方面又非常相似。拿我做的 Grok Bot 那个案例来说——那是很多轮迭代循环才找到产品市场契合点。作为产品设计师，我首先看重的是一个人交付产品的能力，以及他有没有一套体现“寻找产品市场契合”这个循环的流程，因为我们处在一个什么都能造的时代，但并不是所有东西都值得造。重点不在于你有多少想法、你的想法多聪明，而在于它们能不能变成别人愿意用的形态。

I think at this point, the fundamentals matter way more than a lot of other factors. I think people tend to gravitate towards how quickly somebody works, or how many different tools they have in their arsenal, how many skills they utilize, or their optimization strategy for the models, so on and so forth. I think as a designer, the process is in some ways very different, but also very much the same. Using the Grok Bot work that I did as a case study—it was a lot of iterative loops to get product-market fit. And as a product designer, I would say primarily somebody's ability to ship a product and having a process that represents that product-market fit finding cycle, because we're in a time where you can build anything, but not everything is worth building. It's much less about how many ideas you have, how clever your ideas are, but do they take the shape that somebody else would want to use?

对吧？我觉得可以放心地说，你们已经具备产品市场契合的雏形了，因为 Grok Bot 简直是我上周 Twitter 信息流里唯一的东西。我个人也相当上瘾。你确实搞出了点名堂。我的问题是，第一，看到 Twitter 因为你做的东西而爆炸是什么感觉？我肯定那很棒。但第二，老兄，我猜你肯定收到了铺天盖地的反馈、想法和请求。跟我聊聊，你怎么决定忽略什么？什么值得立刻跟进？在内部构建并使用了一两个月之后，现在有哪些指导原则塑造了你解读这些反馈的方式？

Right? I think it's safe to say that you all have the makings of product-market fit, considering Grok Bot is literally like the only thing on my Twitter feed the last week. I'm personally quite hooked too. Like, you're on to something here. I guess my question is one, what's it kind of like just seeing Twitter explode based off of something that you made? I'm sure that's awesome. But then two, man, I got to imagine you are getting an onslaught of feedback and ideas and requests. Talk to me a little bit about how do you even think about what to ignore? What makes sense to jump on? What are some of the guiding principles that shape how you now interpret all of this feedback, after kind of building and using everything internally for a month or two?

嗯，我们仍然非常坚信用户研究和数据驱动的决策。

Well, we still believe very strongly in user research and data-informed decisions.

这两个因素的结合，加上我们手头有如此广泛的定量和定性数据，正在为当前的路线图提供信息。正如你提到的，时间线上有大量的反应和热情，也不可避免地有很多噪音。但我认为，当有深思熟虑的用户时，你真的能看出来，他们对产品体验中哪些有效、哪些无效有更深的欣赏或更深的批评。我们之前聊过一点，你知道，Brian Loveven 他列了一整串他生活中喜欢的东西。我认为拥有那种高级用户作为资源是惊人的。我给他发了私信，要求他给出更有针对性和更具体的反馈。于是他发了一堆截图，然后我让我的机器人把这些编译成一个 Notion 文档，分享给了团队。所以我认为，既有那种对反馈非常深思熟虑的用户，你真的必须重视并利用他们作为资源，但也有大量的数据表明，人们通常使用少于五个机器人，对吧？这对 UI 意味着什么？对应用的 IA 意味着什么？我们知道很多人在认证上卡住了，对吧？我们有哪些方法可以优先简化认证？我们知道角色基于反应能引起人们共鸣。我们有哪些方法可以让 Grok Bot 作为吉祥物更突出，或者在某些互动阶段不那么突出？最后，我认为我们有什么机会超越我们最终确定的明显形态。再次，人们提出了一个 iMessage 版本的想法，一个存在于 iMessage 中的 Grok Bot。我在想，是的，我认为如果我们看到这种情况，而且我们已经看到了一些这样的东西，人们在拼凑工作流程，我认为一旦你看到趋势或模式出现，就应该开始为此进行设计。

The combination of those two factors and having such a wide range of both quantitative and qualitative data at our disposal is kind of informing the road map right now. As you mentioned, there's a ton of both reaction and enthusiasm and invariably a lot of noise coming from the timeline. But I think when there are thoughtful users, you can really tell they're the ones that have a deeper appreciation for or a deeper criticism about what in the product experience is working or is not. We chatted about it a little bit earlier, but you know, with Brian Loveven, he kind of did a whole list of things that he liked in his life. I think having that type of power user is amazing as a resource. I DM'd him and was asking for more pointed and more specific feedback. And so he sent a bunch of screenshots, which I then in turn had my bot compile into a notion doc and shared out with the team. So I think there's both the type of user that is very thoughtful about giving that feedback that you really have to value and utilize as a resource, but then there's also tons of data that we know that people are working with fewer than five bots on the regular, right? What does that mean for the UI? What does that mean for the IA of the app? We know that people are getting stuck with a lot of authentication, right? What are ways that we can prioritize streamlining that? We know that the character resonates with people based on reaction. What are ways that we can have Grok Bot as a mascot become more front and center or, you know, less so at certain periods of interacting with the app? And lastly, I think what are opportunities for us to expand beyond the obvious form factor that we landed on. Once again, people have floated the idea of an iMessage version of Grok Bot that lives in iMessage. I was wondering like, yeah, I think if we see that and we've seen some of this stuff, people are hacking together workflows, and I think once you see the trend or the pattern emerge, it behooves one to kind of start designing for that.

在你离开之前，也许我们可以谈谈你个人是如何使用 Grok Bot 的，因为我猜很多听众会想，“好吧，酷。”但这有点模糊。你知道，他们还没有经历过你那世界级的优雅的入门流程。他们不太确定这如何融入他们的设计实践，或者老实说，甚至他们的生活。比如我现在有大约七个机器人，其中只有一个跟设计有关。所以你可以朝任何方向说，但你现在最兴奋的、在自己生活中使用 Grok Bot 的用例有哪些？

Before I let you go, maybe we could just talk a little bit about the way that you were personally using Grok Bot, because my assumption is that a lot of people listening are like, "Okay, cool." But it's a little hazy. You know, they haven't gone through your world-class elegant onboarding flow yet. And they're just not exactly sure how this fits into either their design practice or honestly even just their life. Like I got about seven bots and only one of them is design right now. So you can take it in any direction you want, but what are some of the use cases that you're most excited about for your own life right now using Grok Bot?

一个用例是代表你收集数据，既在你的机器上，也在你分散在网络上的身份中。一个非常具体的例子是绿卡申请。现在我的妻子正在申请绿卡，每次填写这些表格时，你需要调出大量的文件。比如银行对账单、租房历史。特别是那个，我当时想，我不记得我过去四年的情况。我在纽约。你每年都搬家。嗯，你知道，我以前每年都搬家。现在我的位置比较稳定，但他们说，你必须提供过去 5 年的记录。对我来说，那是三个不同的公寓。我不记得我搬出的确切日期，但 Grok Bot 可以进入我的收件箱，扫描所有内容，无论是我和旧室友的通信，还是实际的租赁协议本身。所以很多那种翻找挖掘的工作，它非常有价值。另一个例子是派它去做研究。另一个例子还是关于租房的。我在找公寓，我想，你知道典型的工作流程是上 StreetEasy，按条件筛选，但情况在不断变化，所以除了每周刷新页面，你怎么跟上？放一个机器人上去，给它具体的位置，甚至精确到，嘿，我想坐蓝线去上班。所以，找靠近蓝线的地方，然后排序。你知道，你也可以在很多其他工具里做这个。但我认为，有一个已经建立上下文和特定任务的东西，意味着它给我的结果可能比这些其他工具更好。

One use case is in general gathering data on your behalf, both on your machine and across your identity that is scattered across the web. A very specific example is green card application. Right now my wife is applying for her green card and there's a ton of documentation you need to pull up anytime you fill in one of these forms. They're like bank statements, history of rent. And that one in particular, I was like, I don't remember my last four. I'm in New York. You move every year. Well, you know, I used to move every year. Now I'm kind of stable position, but they're like, you have to have the last 5 years. And for me, that's three different apartments. I don't remember the exact date that I moved out, but Grok Bot can go into my inbox and just scan everything, whether it's my correspondence with my old roommate or the actual lease agreement itself. And so a lot of that like packing around digging type of work it's super valuable. Another example is sending it off to do some research on your behalf. Another example here is with rent once again. I was looking at apartments and I was like, you know the typical workflow is you go on to StreetEasy and you filter by something and things change on a consistent basis, and so how do you keep up with that beyond just refreshing the page every week? Put a bot on it, you know, give it specific location even down to like, hey, I'd like to ride the blue line to work. So, find something near the AC, and then stack rank it. You know, you can do this in a lot of other tools, too. But I think having something that has context built up and a specific task means that the results it gives me is probably better than some of these other tools.

最后，我想我之前提到的那个，就是代表我在 Figma 中工作。我认为有很多繁琐或重复的任务，你只是反复做，很花时间，你不想做。找那个标志，你知道，映射出卡片的每个变体，并让它跨不同工具工作，这非常有用。让它从 Slack 拉取数据，上传到 Notion 数据库，然后变成 Figma 中的设计，这意味着如果我们有一个反馈渠道，它就会不断更新设计，而我所需要做的就是检查它的工作。所以，真正那种跨平台、跨应用、跨表面的能力，就是我在工作方面使用它的方式。我目前最大的收获是，我应该假设我能做任何事，因为我还没有遇到障碍。我已经在很多方面拉伸了它，比如试着说，老兄，你能做这个吗？就像，是的，没问题。天哪。

And finally, I think the one that I pointed to earlier, which is working in Figma on my behalf. I think there's a lot of tedium or the types of tasks that you would just do on repeat and it just takes time and you don't want to do it. Looking for that logo, you know, mapping out every variant of a card and having it work across different tools is super useful. Having it pull data from Slack to upload into a Notion database to then turn into designs in Figma means that if we have a feedback channel, it's just updating designs constantly and all I need to do is check its work. And so just really that ability to work across platform, app, and surface is kind of what I use it for as far as work goes. My biggest takeaway at this point has been I should just assume that I can do anything because I haven't hit a wall yet. And I've stretched it in many ways like trying to like, man, can you do this? Like yeah, no problem. Oh my gosh.

我要分享两个用例。一个有点另类，我热爱梦幻足球。我参加了几个高风险的梦幻足球联赛，大家太认真了。我要构建，我已经在构建一个最不可思议的梦幻总经理机器人，它抓取所有东西。它有一个动态的电子表格模型，用于拍卖，而 Sleeper 上的所有交易数据都是公开的。我的意思是，我能做的事情简直难以置信。我对今年以及 Grok Bot 解锁的东西感到非常兴奋。第二件事是我看到一条推文，有人说，“哦，我用它来 QA 我的产品。”我想，“那太天才了。”所以我们正在为 Inflight 做一个大版本发布。我刚刚设置了一个新的 GitHub 和 Vercel，然后说，“嘿，就用它。把每个流程都走一遍。”我让另一个代理快速制定了一个 QA 大纲，然后它就按清单执行，现在它把它设置为每日任务，它运行 QA，创建工单，如果工单达到一定的置信水平，它就立即把它扔给一个 Cursor 代理。而我昨天在纽瓦克机场的行李提取处坐着，我的行李丢了。

The two use cases I'll share. One a little bit out there is I love fantasy football. I'm in like a bunch of high-stakes fantasy football leagues that take it way too seriously. I'm going to build, I'm already building the most unbelievable fantasy GM bot that just scrapes everything. Has this dynamic like spreadsheet model that it's using for auctions and all the transaction data on Sleeper is public. I mean, it's just unbelievable what I'm able to do. I'm so excited for this year and what Grok Bot unlocks. And then the second thing is I saw a tweet and somebody was like, "Oh, I'm using it to QA my product." I'm like, "That's genius." And so we're working on like a big release for Inflight. And I just set up like a new GitHub and Vercel and just said, "Hey, just use it. Just go through every single flow." I had another agent whip up like a QA outline and then it just went down the list and now it sets it up as a daily job and it just runs QA, creates tickets and then if it reaches a certain confidence level on the ticket, it just immediately tosses it to a cursor agent. And I'm sitting there in baggage claim at the Newark airport yesterday and my bag is lost.

我还有大约一个小时的时间，电脑没电了，因为我在飞机上工作，所以我只能用手机发布修复。我修复了一个之前用旧模型两次都没能修复的 bug，而这次我就在行李提取处用手机搞定了。你会有种感觉，哇，这可是件大事，你知道吗？这真是件大事。

I also had like an hour and my computer was dead because I was working on the plane, and I'm just on my phone shipping fixes. I fixed a bug that I was unsuccessful in fixing two other times with older models, and I just did it on my phone sitting in baggage claim. You're kind of just like, dang, this is a big deal, you know? This is kind of a big deal.

是的，我觉得我们还会发现更多的使用场景。这件事令人兴奋的一点是，我的使用场景在整体来看其实非常狭窄。人们会分享他们的技巧、窍门、或者日常习惯，我也会开始采纳这些。看到人们把它用到你没想到的方向，这非常令人兴奋。

Yeah, and I think we're going to uncover way more use cases too. I think part of the exciting thing about all this is that my use case feels very narrow in the grand scheme of things. People are going to be sharing their hacks, tips, tricks, or routines. I'm going to start adopting those as well. It's a very exciting thing to see people take it in directions that you didn't even think about.

感谢你今天来做客，给我们分享了一些幕后故事。我是说，很多我们本来永远不会看到、也不会知道背景的事情。看到你探索不同想法的方式、你的工作方式，真的非常有趣。我想代表我自己和所有听众感谢你，今天为我们揭开了帷幕。

I appreciate you coming on today and giving us a little bit of a behind the scenes. I mean, just a bunch of stuff that we would never otherwise see and have the backstory for. Really really enjoyable to see how you explore different ideas, the way that you work, and just want to say thank you on behalf of myself and everybody listening for coming on and pulling back the curtain today for us.

非常愉快，谢谢。

It's been a blast. Thank you.