Claude Code for normal people: skills, voice mode, and how to collaborate with AI

频道: How I AI
视频: https://www.youtube.com/watch?v=o_eg2TtXAO0
原文语言: en
统计: 共 76 轮 · Grace 36 · Claire 25

[0:00]

Prompt engineering is dead, but intent engineering is where we need to be focusing our time. I do a [music] lot of work when I'm walking and on the go. And I spoke maybe for two or three minutes and just said, "Here's my problem." Claude came back to me [music] and said, "I think we need to be creating an interactive artifact that's password protected." That is a prompt going from a hyperengineered chunk of text to a conversation that's 3 or 4 minutes. Cloud needs to get the prompt out of you. You have another example at the personal level as a fellow email hater. What did Gmail do to us?

prompt engineering 已经死了，真正该花时间的是 intent engineering（意图工程）。我很多活儿是边走路边干的。我大概讲了两三分钟，就说：我的问题是这样的。Claude 回过来跟我说：我觉得我们该做一个带密码保护的交互式 artifact。这就是 prompt 的变化——从一大段精雕细琢的文字，变成一场三四分钟的对话。得让 Claude 把 prompt 从你嘴里掏出来。你还有一个更偏个人层面的例子。作为同样恨邮件的人我想问：Gmail 到底把我们害成什么样了？

[0:35] Grace

What did Gmail do?

Gmail 干了什么好事？

[0:37]

I know it was once upon a time amazing. It is like everybody has a key to your front door and can come leave things in your house. No more. I want to show you what I built out of pure frustration. Send in Gmail. This will push it to Gmail a draft and then hopefully open up a browser window and get it the extra mile. This gets me pretty much where I want to go because I can go back in here and work with Claude and chat and have the reply drafted or I can just send it and never have to deal with it again. My job is to sit over your shoulder and smack your hands every time you open Slack or Gmail or Google Calendar and say, "No, Claude. People need to feel the benefits in order to keep using it." That's actually not true. Instead, people [music] mostly need to understand that we're going to learn to collaborate. The real hump to get over is defaulting to this and building the muscle memory of simply opening an app. Welcome back to How I AI. I'm Clarvo, product leader and AI obsessive here on a mission to help you build better with these new tools. Today I have Grace Clark and she's going to show us Claude for normal people. But not really for normal people. For people who want to show up to their clients and to the many, many, many people in their inbox with a level of personalization and professionalism that only an AI agent can provide. Let's get to it. This episode is brought to you by Bolt.new, the AI app builder for people who have ideas and want to ship them. Most AI tools spit out code that looks great in a demo and falls [music] apart the second you try to do anything real with it. Or they lock you into their own platform with no real way out. Bolt [music] is different. You describe what you want to build, a startup MVP, a landing page, an internal tool, a side project, and Bolt generates productionready code in minutes. Connect Stripe [music] or any other MCP, hook up your domain and deploy it live. Founders

我知道它当年确实惊艳过。但现在就像人人都有你家大门钥匙，随时能进来往你屋里扔东西。到此为止。我想给你看看，我纯粹是被逼疯了才做出来的东西。「在 Gmail 里发送」——它会把草稿推到 Gmail，顺利的话还会打开一个浏览器窗口，把最后一步也走完。这基本就到位了：我可以回到这儿跟 Claude 聊，让它把回复起草好；也可以直接发出去，从此再不用管。我的工作就是趴在你肩膀后面，你每次打开 Slack、Gmail 或者 Google Calendar，我就打你的手，说：不行，去找 Claude。都说人得先尝到甜头才会一直用下去——其实不是这么回事。更关键的是，大家得明白：我们要学的是怎么跟它协作。真正要迈过去的那道坎，是把它变成默认动作，把「打开这个 app」练成肌肉记忆。欢迎回到 How I AI。我是 Claire Vo，产品负责人、AI 狂热爱好者，我的任务是帮你用好这些新工具、把东西做得更好。今天请到的是 Grace Clarke，她要给我们讲「普通人怎么用 Claude」。但其实也不只是给普通人的——是给那些希望在客户面前、在收件箱里那一大堆人面前，拿出只有 AI agent 才做得到的个性化和专业度的人。我们开始吧。本期由 Bolt.new 赞助，这是一款面向「有想法、想赶紧做出来」的人的 AI 应用搭建工具。大多数 AI 工具吐出来的代码，demo 里看着挺唬人，一上真活立刻散架；要么就是把你锁死在它自己的平台上，根本没有退路。Bolt 不一样。你只要描述想做什么——创业 MVP、落地页、内部工具、业余项目——Bolt 几分钟内就能生成可上生产的代码。接上 Stripe 或者任何别的 MCP，绑好域名，直接上线。创业者……

[2:40]

are using Bolt to build businesses doing real revenue. Product managers are shipping prototypes their teams actually use. [music] Designers and marketers are launching campaigns without waiting in line. Anyone can build. Engineering can ship. Everyone wins. You just need an idea and a weekend. Check it out at bolt.new/howi AI. Grace, welcome to How I AI. Thank you for having me. I am excited about this episode because one of my hypotheses about how AI is going to like impact our work is it's really going to allow us all to differentiate on service a lot better. And I think the bar for service and service quality and customer relationships and human relationships overall will go up because we'll be able to do more and make things more customized for the people that we're working with and the and the people that we collaborate with. And so I love some of the stuff you've built. And I wonder if you could just walk us through what brought you to some of these AI projects you're going to show us and what problem you were trying to solve. I am a big believer that AI is democratizing the way all of us are going to work if we put in the work. And I was thinking about different efficiencies that I could build in my business. I'm an AI teacher and a former marketing consultant. So very relationship driven, very process driven business. And I started teaching myself how to build an open claw in January. And I was posting about it enough on Instagram that people said, "You should teach a class. You seem to be documenting this. There's a process. Would you ever want to get people together in a room and do it?"

……正在用 Bolt 做出真有营收的生意。产品经理用它做出团队真的会用的原型。设计师和市场同学不用再排队等资源，就能把活动推上线。人人都能做东西，工程团队专心交付，皆大欢喜。你只需要一个想法和一个周末。去 bolt.new/howiai 看看。Grace，欢迎来到 How I AI。谢谢邀请。我特别期待这一期。因为我一直有个判断：AI 对工作的影响，很大一块是让我们能在「服务」上做出真正的差异化。服务的标准、服务质量、客户关系乃至人与人的关系，整体门槛都会被抬高——因为我们能做的更多，能为一起合作的人把东西做得更定制化。你做的一些东西我特别喜欢。能不能先讲讲，是什么把你带到今天要展示的这些 AI 项目上的？你当时想解决的是什么问题？我坚信 AI 正在让工作方式民主化——前提是你自己肯下功夫。当时我在想，自己的生意里还有哪些效率可以做出来。我是 AI 讲师，之前做市场顾问，所以这门生意非常依赖关系，也非常依赖流程。一月份我开始自学怎么搭 OpenClaw。我在 Instagram 上发得够多，就有人说：你该开个课。你一直在记录这个过程，这里面是有方法的。要不要把大家凑到一间屋子里一起做？

[4:25] Grace

The process of teaching myself OpenClaw and building a curriculum is now the foundation of what I teach people and it's the reason I have a few core products that I've built that I actually use in my curriculum. So now I've built a few processes and products that run my business that I actually teach to my students and then the teams that I get to train. The most impactful is my pipeline operator. And that's because underneath it is solving a few problems. One, just a deluge of emails. Email is a scourge. Nobody wants to be in Gmail anymore. So my claude is ingesting all of that and correlating it with a bunch of context. The second problem it's solving is that client relations and relationship building is really devoid of any emotion and love. So when we can make visual expressions of that through HTML, interactive HTML and encrypted documents, it's a really warm welcome and it's a much more hospitable way to communicate process. And then third is I simply cannot keep track of 20 open tabs in any given day, much less organize a process where I'm meant to support people and teach them and support them through learning. So going from 20 tabs and 20 hours a week on admin just to be able to teach people was so painful. I wasn't doing the teaching and I just had a hunch that there was going to be a way to solve all these things. The first thing I did was open up Claude Code after having it downloaded for 2 hours and I yapped into it everything I just told you and I sat back and let Claude take me through the process and learned from the jump how to collaborate with Aentic AI and now I have a pipeline that runs once an hour and it's probably the most popular thing I teach. Amazing. So, can we actually see what this looks like? Because I think there's a lot of folks that, you know, have businesses similar to yours that are just looking for practical ways to integrate this into your flow. And so, I'd love just

自学 OpenClaw、顺手搭出一套课程的这个过程，现在就是我教学的地基；也正因为这样，我做出了几个核心产品，而且真的用在课上。所以现在有几套流程和产品在跑我的生意，同时我把它们教给学生，也教给我培训的那些团队。影响最大的是我的 pipeline operator，因为它底下同时解决了好几个问题。第一，邮件多到淹死人。邮件就是一场灾难，没人还想待在 Gmail 里。所以我的 Claude 把这些全吃进去，再跟一堆上下文关联起来。第二个问题是，客户关系和关系维护这件事现在毫无温度、毫无用心。当我们能用 HTML、交互式 HTML 和加密文档把它做成看得见的东西，那就是一份特别温暖的欢迎，也是一种更有待客之道的方式来传达流程。第三，我一天开 20 个标签页根本管不过来，更别说还要组织出一套流程去支持学员、教他们、陪他们学下去。20 个标签页、每周 20 小时耗在行政杂事上，就为了能去讲课，太痛苦了。我根本没在教书。我隐隐觉得，这些事一定有办法一起解决。我做的第一件事，就是在装好 Claude Code 两小时之后把它打开，把刚才跟你说的这些一股脑对着它说了一遍，然后往后一靠，让 Claude 带着我走完整个流程，从第一天就学会怎么跟 agentic AI 协作。现在我有一条每小时跑一次的 pipeline，可能是我课上最受欢迎的内容。太棒了。那能真的看看它长什么样吗？我觉得有很多人做的生意跟你类似，他们就想找些实际的办法，把这套东西接进自己的流程里。我很想……

[6:40] Grace

some inspiration on how this thing actually works. You don't need to run a whole process like this in order to have cloud or gentic AI support you on demand. It can generate a proposal without all of this underneath. It's something that looks like this. At the core of this is a skill file that tells Claude to create a proposal in HTML. And the outcome is something that looks like this.

……从它实际怎么跑的里面得到点启发。其实你不用搭这么一整套流程，也能让 Claude 或者 agentic AI 随叫随到地帮你。底下没有这些东西，它照样能生成一份提案。大概就长这样。核心是一个 skill 文件，告诉 Claude 用 HTML 生成提案。产出大概是这样的。

[7:07] Claire

Yep.

对。

[7:08] Grace

Beautiful branded reflects all of the context from conversations I've had with someone, puts it into a process that I actually get to talk to and iterate. And then we get to gift the client this wonderful welcome to the way that we're going to work. And it acts as a bit of an advertisement for what they're going to be able to do because immediately they're typing a password into a site that looks like them, that feels like them, that they can interact with, and I get to say, "This is what you're going to be able to build by the end of our time together."

很漂亮，带着我的品牌感，把我跟这个人所有对话里的上下文都吸收进去，变成一套我可以真正对话、反复打磨的流程。然后我们就能把这份「欢迎来到我们的合作方式」当礼物送给客户。它同时也是一个预告片，告诉他们将来能做出什么——因为他们一上来就要在一个长得像他们、气质像他们、还能互动的页面里输入密码。我就可以说：等我们合作结束，你也能做出这样的东西。

[7:42]

[snorts]

（笑）

[7:43] Grace

It also spits out interactive pre-work that gets someone prepped for our time together [snorts] automatically. We'll pull updated documentation from any of the large labs that we're going to be working with. You can click around on the back end. This is communicating to me where people are in their progress. So, I can nudge them if they're not quite ready for our training. And my favorite part that I just added on to this skill is the generation of a questionnaire before I have a session with people so I can understand themes and where they're at, where they're stuck, what have you started using Claude for since our first session, what's challenging. Seeing themes here is really helpful and there would be no way for me to have gathered this through Slack DMs or one-on-one conversations with a 30 person team. That's just not going to happen. So underneath this is the same logic. It's a Google form, but it is much more branded, much more fun to make and students and clients will say, I need to learn how to do that myself. Teach me what I tell people when they are trying to drive AI adoption internally. So I like I I get to work and talk to a lot of executives and they're trying to teach their teams to use AI and give them good reasons to use AI. And I tell them if every touch point that you do is not AI first, then how can you convince people that every touch point they should do should be AI first?

它还会自动生成交互式的课前作业，让人提前为我们的课做好准备。我们会从要用到的那几家大模型厂商那里抓最新的文档。后台可以点着看，它会告诉我每个人的进度到哪儿了，如果谁还没准备好，我就能去推一把。我最喜欢、也是刚加进这个 skill 的功能，是在开课前自动生成一份问卷，让我了解共性的主题、大家的水平、卡在哪里——比如第一次课之后你开始拿 Claude 做什么了？哪里觉得难？能看到这些共性特别有用。面对一个 30 人的团队，靠 Slack 私信或者一对一聊，我根本收集不到这些，压根不可能。底下的逻辑是一样的。它本质上就是一个 Google 表单，只是品牌感强得多、做起来也有意思得多。学生和客户会说：这个我自己也得学会，教教我。这也正是我对那些想在公司内部推动 AI 落地的人说的话。我有机会接触和交流很多高管，他们想教团队用 AI，也想给团队一个非用不可的理由。我就跟他们讲：如果你自己每一个触点都不是 AI 优先的，你凭什么说服别人，让他们每一个触点都 AI 优先？

[9:10]

So, I'm like, if your agendas, you know, you're not doing the best job doing these like cool HTML sort of like custom agendas for your meeting, if you're not if you're not using an agent, if you're not customizing them, then you can't demonstrate to others how you can change how you work. And so I love this from the moment you're trying to I would say like incept a client to really see the AI future, you're you're setting a vision by saying if you use AI the way I am just implicitly using AI, you're going to get this level of quality asset and just imagine how that could could go to your customers. I mean the other thing that I always tell people is I try to get out proposals very fast. I'm sure you can get proposals out very fast. And I'm like, if you if you like improve your AI use, you too could get proposals out in like 45 minutes if you really wanted to. And so I I I think like setting the standard of like speed, quality, personalization is is a good way to just get people to start seeing the the impact of AI and how they can imagine it in their life or their work. The number one thing that I get asked when I'm teaching is how can I build the muscle of going to AI of going to ChachiBT desktop of going to claude and underneath it I need to support them in building a muscle. So teaching them that everything we're looking at is AI is not sufficient which is one of my big lessons from teaching now. Instead there are two things that I really focus on doing even for friends who are just asking how they can get into this. One is to build a forcing function in your environment that gets you to open up Claude. And the easiest thing is to set a Slack reminder or a Google calendar alert that says whatever you're doing, screenshot it and put it into Claude.

所以我会说：如果连你的会议议程都没做好——没做成那种很酷的 HTML 定制议程，没用 agent，没做个性化——那你怎么向别人示范工作方式是可以改变的？所以我特别喜欢这一点：从你想给客户「植入」一个 AI 未来的念头那一刻起，你就在设定一个愿景——你在说，如果你像我这样把 AI 用在暗处，你能拿到的成果就是这个质量；再想想这些东西交到你自己客户手里会是什么效果。我还常跟人说：我出提案很快，我相信你也很快。我就说，如果你把 AI 用得更好，你真想的话，45 分钟就能出一份提案。所以我觉得，把速度、质量、个性化这三条标准立起来，是让人开始看见 AI 的影响力、并且能想象它怎么进入自己生活或工作的好办法。我教课时被问得最多的一个问题是：我怎么才能养成「遇事先找 AI」的习惯——先打开 ChatGPT 桌面版、先打开 Claude？说到底，我得帮他们把这块肌肉练出来。光是告诉他们「我们眼前看到的这些都是 AI 做的」根本不够——这是我教到现在最大的心得之一。我真正会做的是两件事，哪怕对方只是个来问「我该怎么入门」的朋友。第一，在你的环境里造一个强制机制，逼你去打开 Claude。最简单的办法是设一个 Slack 提醒或者 Google Calendar 提醒：不管你手上在干什么，截个图，扔给 Claude。

[10:58] Grace

Get the whole window and just ask Claude, "Could you help me with this?" We are just trying to build the muscle of deferring to Claude and asking, "Could you be my shadow and help me here?" With a screenshot is no text. Claude can infer. It shows people the magic power of inference and that Claude can come to you with a prompt that is just an image. The second thing I tell people is let's actually build a skill together. So, I host virtual co-working [snorts] sessions now and we'll build a voice guide and underneath it people will be able to look at their markdown and I make an effort to teach people technical terms even if in the near future we don't use them because that builds confidence and it eradicates this view of not being technical. That is the biggest hump to get over is that people think I'll never know how to do this so they don't have a chance to get the value out of it. So teaching people this muscle memory and then getting them excited about the technical aspects of this turns everyone into this excited adopter. It's it can be a really slow process. So show us how how this actually works behind the scenes because what I'm seeing is like customized content, customized branding. You have this whole pipeline like kind of what goes into building something like this.

把整个窗口截下来，然后问 Claude：这个你能帮我看看吗？我们要练的就是这块肌肉：习惯性地交给 Claude，问它「你能不能当我的影子，在这儿搭把手？」一张截图，一个字都不用打，Claude 自己就能推断。这能让人亲眼看到「推断」的魔力——你的 prompt 可以只是一张图。我跟人说的第二件事是：我们一起真的动手做一个 skill 出来。我现在会办线上的共创（co-working）场次，一起做一份 voice guide，做完大家能看到自己的 markdown。我会刻意教一些技术术语，哪怕过不了多久这些词可能就不用了——因为这能建立信心，也能打掉「我不是技术人」这种自我设限。最难迈的那道坎就是：人觉得「这我一辈子也学不会」，于是根本没机会尝到它的好处。所以先教肌肉记忆，再让他们对技术那部分兴奋起来，人人都会变成积极的使用者。这个过程可能真的很慢。那给我们看看幕后到底怎么跑的吧。我现在看到的是定制内容、定制品牌，还有你这一整条 pipeline——做出这样一个东西，都需要什么？

[12:15] Grace

It's really just three steps. Not to simplify it too much, but underneath it is some standards that you document with Claude. For me, it's a voice guide and it is an explanation of what a proposal is for me. Defining those two things, putting it on a timer and then having everything pub to netleify. So, I want to show you what it actually looks like. Give you the road map. Yep. One thing I teach all my students is the importance of verbalizing an SOP, especially when we're building workflows, whether it's one simple task or many chain together. So, one of our exercises is yapping into Claude and having it make us a road map that we can understand are these steps right? It's as if we're training a new employee. So, underneath I'm going to show you the skills. It's a pipeline operator that wakes up every hour and checks a couple things for me and knows if we need to move clients through different steps or make something. It's my proposal maker which I'll show you and it's my voice guide, my how I think and sound guide. extra cherry on top is sometimes I will run a proposal through my board of directors and on that board right now is Cat who's the head of applied AI at anthropic Ben Thompson from strategy Jamie Diamond who is going to give me a really solid CBA costbenefit analysis and then a skeptic an investor and a founder and I will run everything at the end through them but I want to show you underneath what it actually looks So, proposal maker, I updated this one yesterday. I like to teach students to version their documents and add naming conventions to it so they feel more connected to their tech.

其实就三步。不是我故意说得简单：底下无非是一些你跟 Claude 一起写下来的标准。对我来说就是一份 voice guide，加上一份「提案对我意味着什么」的说明。把这两件事定义清楚，挂上定时器，然后让所有东西发布到 Netlify。我想让你看看它实际长什么样，给你一张路线图。——好。我教所有学生的一点是：把 SOP 用嘴讲出来非常重要，尤其是在搭工作流的时候——不管是一个简单任务，还是好几个串在一起。所以我们的练习之一，就是对着 Claude 一顿说，让它给我们生成一份路线图，然后我们来判断：这些步骤对不对？就跟带一个新员工一样。下面我给你看这些 skill。一个是 pipeline operator，每小时醒来一次，替我检查几件事，判断要不要把客户推进到下一个环节、或者需不需要生成点什么。一个是 proposal maker，待会儿给你看。还有我的 voice guide——也就是「我怎么思考、我说话什么调性」的指南。锦上添花的一步是：有时候我会把提案丢给我的「董事会」过一遍。现在这个董事会里有 Anthropic 负责 applied AI 的 Cat，有 Stratechery 的 Ben Thompson，有 Jamie Dimon——他会给我一份非常扎实的成本收益分析（CBA），再加上一个怀疑论者、一个投资人和一个创业者。最后所有东西我都会让他们过一遍。不过我想先给你看看底下实际长什么样。这就是 proposal maker，我昨天刚更新过。我喜欢教学生给自己的文档做版本、定命名规范，这样他们跟自己的技术之间会更有连接感。

[14:03] Grace

But this outlines exactly what I need to do. It has a change log at the top and then it has rules about my philosophy for teaching, which is where we define the difference between teaching and consulting. has different steps, different reference points. It will also explain how it to itself how it needs to lay out this document, has different workflows, confirms the deal shape, on and on and on and on. And how did you how did you make this skill? Just like what's your general like skill making process? Did you write this by hand? Did you use um our favorite the Yappers API to get it out? like h how did you actually build this skill? I'm a big believer that prompt engineering is dead, but intent engineering is where we need to be focusing our time. Technically, this was me on a walk opening up the cloud mobile app. I do a lot of work when I'm walking and on the go. So, the mobile app is incredible for getting things out of your head in an unfiltered way. And I spoke maybe for two or three minutes and just said, "Here's my problem, and here's what I think I want. First, do not interview me ever. I hate this have Claude interview you approach. The work should never be on you. Claude has immense amounts of context if you've connected it right. It should be studying you and then coming back with a really strong idea that you can react to. So, I always put the pressure on Claude and I said, "Come back to me. What do you think this process could be? Is this even possible?" And Claude came back to me and said, "I think we need to be creating an interactive artifact that's password protected. What do you think?

这里面把我要做的事写得清清楚楚。开头是变更日志，接着是我教学理念的规则——我们在这儿定义了「教学」和「咨询」的区别。再往下有不同的步骤、不同的参照点。它也会自己对自己解释这份文档该怎么排布，里面有不同的工作流，会确认这单生意的形态，等等等等。那这个 skill 你是怎么做出来的？你做 skill 的一般流程是什么？是手写的吗？还是用了我们最爱的那个「碎碎念 API」（Yapper's API）把它说出来的？你到底是怎么搭起来的？我坚信 prompt engineering 已经死了，真正该花时间的是 intent engineering。严格说，这东西是我散步时打开 Claude 手机 app 做出来的。我很多活儿是边走边干的，手机 app 特别适合把脑子里的东西不加过滤地倒出来。我大概讲了两三分钟，就说：我的问题是这样，我大概想要的是这样。第一条：永远不要反过来采访我。我特别讨厌那种「让 Claude 来采访你」的做法。担子不该压在你身上。只要上下文接对了，Claude 手里的信息量是巨大的。它应该去研究你，然后拿一个足够有力的想法回来，让你来反应。所以我总是把压力压给 Claude。我说：你回来告诉我，你觉得这个流程能长什么样？这事到底可不可行？Claude 回过来说：我觉得我们该做一个带密码保护的交互式 artifact，你觉得呢？

[15:41] Grace

And we went back and forth. That is a prompt going from a hyperengineered chunk of text to a conversation that's three or four minutes. Claude needs to get the prompt out of you. So, this was 10 minutes of talking and then an hour of Claude creating this thing and me giving it feedback, actually creating an HTML right inside Claude that I was pressure testing and throwing screenshots in of what I liked and didn't like. This was the very first thing I ever made when I was making the pipeline operator and I pretty much did it mostly on my phone.

然后我们来回聊了几轮。这就是 prompt 的转变——从一大段精雕细琢的文字，变成一场三四分钟的对话。得让 Claude 把 prompt 从你嘴里掏出来。所以整件事就是：10 分钟说话，然后一个小时里 Claude 在做这东西、我在给反馈——直接在 Claude 里生成 HTML，我一边压力测试，一边把喜欢和不喜欢的截图丢进去。这是我做 pipeline operator 时做出来的第一个东西，而且基本上是在手机上完成的。

[16:14] Grace

I I love it. I love it so much. And how often do you like do you read this really unless you're teaching or you're like that's Claude's business. That's not my business. And as long as the output's good, I'm happy. I think the world is right that HTML is the new markdown, which I know is an easy thing to throw around, but it's true. I actually don't read markdown anymore unless I am teaching. Occasionally I'll go in here and point to certain things, but I want to see my markdown visualized.

我太喜欢了，真的太喜欢了。那你多久会看一次这个？除了备课的时候，你真的会读它吗？还是说「这是 Claude 的事，不归我管，只要产出好我就满意」？我觉得大家说得对，HTML 就是新的 markdown——我知道这话很容易被随口一说，但确实是真的。除了讲课，我现在真的不看 markdown 了。偶尔会进去指一指某些地方，但我更想看到我的 markdown 被可视化出来的样子。

[16:45]

Y

嗯。

[16:45] Grace

I am not in here anymore. I do look at the markdown for my voice guide, which is another part of this proposal maker. It is a skill that autofires in almost every single thing I do because it's not just a voice guide, it's a think like me guide and it's the best corpus of how I make decisions. So Claude always needs to have that ready to go. So I want to show you that too. All right. Pride and joy and the number one thing I teach in my class because it is a quick win. It forces us to learn the power of ingesting context and it teaches students how to push and collaborate with Claude and then commit something so that it is an invocable skill. What's most interesting about creating this is all it requires is you telling Claude how you think and then having it go study you. So to create this which has all of my philosophy of communication, words I use, words I don't use. This started with me saying, I think I need a guide that will make everything more like me. And that was the very first instruction that I gave Claude. and it had the idea to produce more of a communication and philosophical guide. Only down at the bottom do we start to see words I use and words I don't use, which is where we get we avoid the AI slop of it all. And now my students can take this and repurpose it for themselves. No fluff, no filler.

我基本不再进这里面了。不过我 voice guide 的 markdown 我还是会看，它也是这个 proposal maker 的一部分。这个 skill 几乎在我做的每一件事里都会自动触发，因为它不只是一份「声音指南」，更是一份「像我一样思考」的指南，是关于我怎么做决策的最好语料。所以 Claude 必须随时把它备好。这个也给你看看。好。这是我的心头好，也是我课上教的头号内容，因为它能很快见效。它逼着我们体会「把上下文喂进去」的威力，也教学生怎么去推着 Claude、跟它协作，最后把成果 commit 下来，变成一个可以被调用的 skill。做这东西最有意思的地方在于：你只需要告诉 Claude 你是怎么思考的，然后让它去研究你。这份东西里有我全部的沟通理念、我会用的词、我不会用的词。它的起点就是我说了一句：我觉得我需要一份指南，让所有东西都更像我。这就是我给 Claude 的第一条指令。然后是它自己想到，应该做成一份更偏沟通、偏理念的指南。一直翻到最下面，才开始出现「我会用的词」和「我不会用的词」——这一段就是用来躲开那股 AI 味儿的。现在我的学生可以直接拿去改成自己的版本。不废话，不注水。

[18:15] Grace

[laughter] No, you're absolutely right. Here's the uncomfortable truth. I never want to see these things. And most people don't either, which is why this gets updated. So often I'll just voice note into Claude and say, "I saw this terrible post on LinkedIn. Can you make sure my voice guide [laughter] never touches like this?"

（笑）不，你说得完全对。说个扎心的事实吧——这些话我一个都不想再看到，大多数人也一样，所以这份东西才要不停更新。我经常直接给 Claude 发条语音说：我在 LinkedIn 上看到一条特别难受的帖子，你能不能保证我的 voice guide 永远不碰这种东西？

[18:34]

Yes. I mean, underneath this in the session log is just basically me calling out all sorts of people on LinkedIn who I never want to film. It's just like a moving target, too, because every time one of these new models come out, it like it ships like two or three really obvious tells and then it just starts to like grind my gears every time every time I see it. I think side note, I think on TikTok I could do you know how those I don't know if you know this, how like there are women that will like close their eyes and they have like a blind box and they taste the different diet cokes. They're like, "This is diet coke and ice and this is diet coke and I could do that with slop from a model." Easy. Easy. I could do it. So maybe there's going to be a new how I AI mini series on Claire Blind taste testing slop from.

对。这底下的 session log 里，基本上就是我在点名 LinkedIn 上各种我绝对不想变成那样的人。而且这还是个移动靶——每次出一个新模型，它都会带上两三个特别明显的口头禅，然后我每看到一次就来一次气。插一句，我觉得我在 TikTok 上可以做那种——不知道你看没看过，有些女生会闭上眼睛，弄个盲盒，去尝不同的健怡可乐，然后说「这杯是健怡加冰」「这杯是原味健怡」。我完全可以拿模型产出的 AI 废话来做同样的事。轻轻松松，我真能做到。所以说不定 How I AI 会出一个新的迷你系列：Claire 盲品 AI 废话。

[19:21] Grace

Do you think you could pin pinpoint opus 5?

你觉得你能一口认出 Opus 5 吗？

[19:24] Claire

Yeah, I think I could pinpoint Opus 5.

能，Opus 5 我觉得我认得出来。

[19:25] Grace

I think you probably could.

我觉得你大概真行。

[19:26] Claire

Yeah, I could definitely pinpoint Fable Pinpoint 56 on GC 56.

对，Fable 我绝对认得出来。

[19:34]

Easy. Our sweet 56.

太简单了。我们心爱的 56。

[19:36]

Our sweet 56.

我们心爱的 56。

[19:37] Grace

Our sweet five six. [laughter] So the point really of preserving instruction in skill files has been the biggest teaching for me

我们心爱的 56。（笑）所以说，把指令沉淀进 skill 文件——这件事对我来说是最大的教学收获，

[19:48] Grace

is getting people to understand that you are going to chain these things together and that you can write code. Like Karpathy said years ago, the hottest programming language is English, which is really why I think I have a job and also why I think I won't have a job in a couple months. I don't think I'll be teaching this way, but being able to help people understand how to prompt and then end up with an entire workflow is really powerful.

就是让大家意识到：这些东西是可以一环扣一环串起来的，而且你自己就能写代码。就像 Karpathy 几年前说的，最热门的编程语言是英语。这大概正是我今天有工作的原因，也是我觉得再过几个月我可能就没这份工作的原因。我不觉得我还会用这种方式一直教下去，但能帮人搞懂怎么写 prompt、最后拿到一整套 workflow，这件事的威力是真的大。

[20:13] Claire

I love it. Well, this is I mean I think a bunch of folks can um take inspiration from this. So if you are working with clients and you want to deliver highly customized both like proposal, onboarding, tracking, training experiences and you want them to sound like you and you want it to look like them. I think take inspiration from this and um build your own sort of like SOP. Again, something that I tell people a lot is AI has forced us all to sort of write down the processes we have been doing kind of mechanically in our businesses and actually write down like in an ideal world, how would I do this thing step by step because now you have that ideal world because you have sort of this limitless intelligence and ability to execute on tap which you may not have had a couple years ago. And so I say go down and write the ideal way you would do proposals, the ideal way you would do content marketing, the ideal way you would do software engineering and then that SOP can be executed and automated by AI and then you have an even better performing business. So I think this is a really great example at at the business level. You have another example at the personal level um as a fellow email hater. Hater let's just get rid of it. It's game over. He text me.

我太喜欢了。我觉得很多人都能从这里得到启发。如果你在服务客户，想交付高度定制的提案、onboarding、进度追踪、培训体验，想让它们听上去像你说的话、看上去像客户自己的东西——那就从这里取经，去搭一套属于你自己的 SOP。我经常跟人讲的一点是：AI 逼着我们把过去在生意里靠肌肉记忆机械完成的流程真正写下来，而且是写下来「在理想状态下，我会一步一步怎么做这件事」。因为现在这个理想状态真的到手了——你有近乎无限的智能和执行力随取随用，这在几年前是没有的。所以我的建议是：坐下来，把你做提案的理想流程、做内容营销的理想流程、做软件工程的理想流程都写出来，然后这套 SOP 就能交给 AI 执行、自动化，你的生意跑得比原来还好。所以这是一个业务层面的绝佳案例。你还有一个个人层面的例子——作为同样痛恨 email 的人。真的痛恨。干脆废掉算了，游戏结束，给我发短信就行。

[21:33]

What did Gmail do to us? I know it was once upon a time. Amazing.

Gmail 到底对我们做了什么？我知道，它曾几何时也是挺神的。

[21:37] Claire

It is like everybody has a key to your front door

那感觉就像是，所有人都有你家前门的钥匙，

[21:41] Claire

and can come leave things in your house.

随时能进来往你家里丢东西。

[21:45] Grace

Yes. No more. We cannot work like that. I want to show you what I built out of pure frustration that started as a rant into cloud code and has now become a recreation of Gmail that lives in my cloud. So, I haven't intentionally open Gmail in a month out of muscle memory. I think we might for a while, but we can get out of that slog of emails and not just get out of it, but do one better. We can actually be training our AIs as we respond and communicate through Gmail. If we stay in Gmail, all that learning and all that writing and all that work actually doesn't go anywhere. It doesn't compound. It's just locked away. So now rebuilding Gmail is a project that I think everyone can do in a half hour, maybe a little bit longer if we want to get fancy with it. But I want to show you what it looks like and then talk you through how I built it.

对。受够了。这样根本没法工作。我给你看看我纯粹出于抓狂做出来的东西——一开始只是对着 Claude Code 一通吐槽，现在已经变成一个活在我 Claude 里的 Gmail 复刻版。我已经整整一个月没有主动打开过 Gmail 了，只有肌肉记忆偶尔会把我带过去。我猜这惯性还会持续一阵子，但我们完全可以从邮件的泥潭里爬出来——而且不只是爬出来，还能更进一步。我们在回复和沟通的同时，其实是在训练自己的 AI。如果一直待在 Gmail 里，那些学习、那些写作、那些劳动全都留不下来，不会复利累积，就那么被锁死在里面。所以现在重建一个 Gmail，我觉得是人人都能在半小时内做完的项目——想玩得花一点可能要久一些。我先给你看看它长什么样，再讲讲我是怎么做出来的。

[22:41] Claire

Amazing. This episode is brought to you by Hyper Agent, the platform for deploying always on agents that actually run your business. With Hyper Agent, you build agents in the cloud and deploy them where your work already happens, like Slack, Telegram, or email. [music] An agent will scan your inbox and draft replies to vendor follow-ups. Another monitors [music] competitors and spins up rich ad kits and landing pages. A third notices a deal going cold in Salesforce and writes the save email with full account context. These aren't chat bots waiting for a perfect prompt. They're proactive, [music] learning your preferences, retaining your playbooks, and getting better with every run. One user [music] built four agents to run an outbound sales pipeline, prospecting, outreach, follow-ups, CRM updates, all in a single afternoon. No local setup, no VPS bills, no fragile permissions on your laptop, just powerful agents with full control over skills, tools, and guard rails. Hyper Agent was built by the team behind Air Table and How I AI listeners get $1,000 in free inference to start building. Claim yours at hyperagent.com/howi AI. So functionally this is an artifact in Claude co-work cannot get more simple than this. And the process was talking to Claude code and having it write a markdown file that I saved to my computer that I brought into co-work.

太棒了。本期节目由 Hyper Agent 赞助——一个部署常驻 agent、让它们真正替你跑生意的平台。用 Hyper Agent，你在云端搭 agent，然后把它们部署到你工作真正发生的地方，比如 Slack、Telegram 或者邮箱。一个 agent 扫你的收件箱，替你起草回复供应商的跟进邮件；另一个盯着竞品，自动生成完整的广告素材包和落地页；第三个发现 Salesforce 里某个单子快凉了，就带着完整的客户背景写一封挽回邮件。这些不是等着你喂完美 prompt 的聊天机器人，它们是主动型的——会学你的偏好、记住你的打法，每跑一次就更好一点。有个用户用一个下午就搭了四个 agent，跑通了整条外呼销售 pipeline：找客户、触达、跟进、更新 CRM。不用本地配置，不用付 VPS 的钱，也不用在你笔记本上折腾一堆脆弱的权限，就是一批强大的 agent，skill、工具、护栏全由你掌控。Hyper Agent 出自 Airtable 背后的团队，How I AI 的听众可以领 1000 美元免费推理额度开始搭。去 hyperagent.com/howi AI 领取。好，从功能上讲，这就是 Claude Cowork 里的一个 artifact，不能再简单了。整个过程就是跟 Claude Code 聊，让它写一个 markdown 文件存到我电脑上，我再把它带进 Cowork。

[24:10] Grace

One thing that took me a long time to actually accept was that it is easy to move ideas and conversations between sessions or between products. So, the Claude desktop app co-work underneath its code. But moving things back and forth across that partition is just as simple as asking Claude, "Can you write me a markdown file for another session? Can you write me a markdown file? I can bring this conversation elsewhere." So, I went back and forth with Claude saying, "I hate Gmail. Absolutely never want to be in it. It's a bane of my existence because I avoid things that I don't want to do." And it creates this emotional experience for me. I don't really care about inbox zero. I don't go there. I just want to have a better relationship with my work. Claude was like, let's rebuild Gmail and let's have it look the way you want. Let's choose the colors that you like and I'm going to mock it up for you and pull in some of your real information and we're going to go from there. Back and forth, back and forth. The process was just asking Claude, can you change this? Can we add a link? Can we add bolding? Can we push this to Gmail? Let's see what happens. Um, can I do a live one and see if it works,

有件事我花了很久才真正接受：把想法和对话在不同 session、甚至不同产品之间搬来搬去，其实非常容易。Claude 桌面端里的 Cowork，底下跑的其实就是 Claude Code。而在这条分界线两边来回搬东西，简单到只要问 Claude 一句：「能帮我写个 markdown 文件给另一个 session 吗？」「能写个 markdown 文件，让我把这段对话带到别处去吗？」所以我就跟 Claude 来回聊，我说：我恨 Gmail，一秒都不想待在里面，它是我人生的祸害，因为我会逃避我不想做的事。它带给我的完全是一种情绪体验。我并不在乎收件箱清零，我根本不去那儿，我只想跟我的工作有一种更好的关系。Claude 就说：那我们重建一个 Gmail 吧，让它长成你想要的样子。挑你喜欢的颜色，我先给你做个原型，把你的一些真实数据拉进来，我们从这儿开始。然后就是来来回回。整个过程就是不停地问 Claude：这个能改一下吗？能加个链接吗？能加粗吗？能把它推到 Gmail 吗？我们看看会怎么样。要不我现场跑一个，看看好不好使，

[25:22] Claire

please?

拜托来一个！

[25:22] Grace

Okay, totally okay. Um, embarrassing myself. Let's see. Okay, a student wants the recording for the class. I don't like this draft. It's totally fine. Sending it today. Oh my gosh. Thank you for the reminder. Send in Gmail. This will push it to Gmail a draft and then hopefully open up a browser window and get it the extra mile. But let's see what's going to happen. Ah, magic. Okay. Well, it decided it wants to use the actual original email. Troubleshooting aside, this gets me pretty much where I want to go because I can go back in here and work with Claude and chat and have the reply drafted or I can just send it and never have to deal with it again. I love it. The ideal it process is just telling Claude the outcome you want and the problem that you have and letting it fill in the gaps. Otherwise, if we are overdirecting it, we're in its way and we are not letting it reverse engineer the solutions to our problems. So, no more prompting. Be conversational. Give it your outcome and then let these powerful models help you and carry you and teach you how to work with them. They will rebuild Gmail for you and you'll never have to go into Gmail ever again. I love this use case because I've seen almost everybody I know do this.

好，完全没问题。那就当众出个丑吧。我们看看。好，有个学生想要这节课的录像。我不喜欢这版草稿。完全没问题，今天就发。天哪，谢谢你提醒我。「发到 Gmail」。这一步会在 Gmail 里生成一封草稿，顺利的话还会再打开一个浏览器窗口，多走那一步。我们看看会发生什么。啊，成了。好吧，它决定用原始那封邮件。抛开这点小故障不谈，这基本上已经把我送到我想去的地方了——我可以回到这里跟 Claude 对话、让它把回复写好，或者干脆直接发出去，从此不用再管。我太喜欢了。理想的流程就是：告诉 Claude 你想要的结果、你遇到的问题，剩下的空白让它自己填。反过来，如果我们指挥过头，其实是在挡它的路，等于不让它去逆向推导我们问题的解法。所以别再「写 prompt」了，就正常说话。把你要的结果交给它，让这些强大的模型来帮你、带你，教你怎么跟它们协作。它们会替你重建一个 Gmail，你从此再也不用打开 Gmail。我特别喜欢这个用例，因为我认识的人几乎都干过这件事。

[26:43] Claire

They're like, "I hate email. I'm going to rebuild Gmail." And what I love about this as an exercise for almost everybody, it's universally applicable. We all hate our email. We all have too much. And we're all unique snowflakes in that like you like this like long thing with the pre-draft prompt. I'm like, I just want a voice agent like magical EA that just whispers to me and says, "Hey, Claire, what do you think about this email?" And then I whisper back and the email gets sent without me even thinking about it. like we all want our like special little thing. I have a friend who has like completely built a desktop app that does this. Like there's so many different ways that you could um think about your email experience. And again, what I love about something you said earlier when we were prepping for for this conversation is people feel like I have to be super technical to pull this off. Like I have to be a software engineer, capital S, capital E, to I have to know what a server is. I have to know all this stuff and I'm like no actually you just have to type I hate Gmail build me a better one into cloud code and you are off to the races. Um so can you just show us how you even got started building building this thing? I know you just said like I hate it rebuild it but was it like a oneshot?

大家都是：「我恨邮件，我要重建一个 Gmail。」我喜欢把它当作练习的地方在于，它的普适性极强——我们都讨厌自己的邮箱，我们的邮件都太多了。而每个人的口味又都不一样：你喜欢这种带预设草稿 prompt 的长表单式界面；而我想要的是一个语音 agent，像个有魔法的行政助理，在我耳边悄悄说：「嘿 Claire，这封邮件你怎么看？」我小声回一句，邮件就发出去了，我压根不用动脑子。我们都想要属于自己的那点小定制。我有个朋友干脆做了个完整的桌面 App 来干这事。你可以有无数种方式重新想象自己的邮件体验。还有一点——之前我们为这期节目做准备时你说过一句话我特别有共鸣：很多人觉得「我得非常技术才做得成这件事」，觉得自己得是大写 S 大写 E 的 Software Engineer，得知道什么是服务器，得懂一大堆东西。而我想说：不，你只需要在 Claude Code 里敲一句「我恨 Gmail，给我造个更好的」，你就已经上路了。那你能不能给我们讲讲，你到底是怎么开始把这东西做起来的？你刚说了就是「我恨它，重建它」，但那是一次成型吗？

[27:58] Grace

Did it did it take a little time? What is it is it using connectors behind the scenes kind of how does it technically work? the technical elements of this, the way I built it was I wish it was a oneshot. Although I'm grateful for the collaboration with these tools, but underneath it, I just followed a few steps. One was I connected every single thing that I could and then made custom connectors and custom plugins for the other data sources Cloud was going to need. For example, Cloud has a hell of a time right now writing to Google Sheets, writing to Google Docs. So, I made custom plugins which taught me the process of making a Google Cloud project and a service account which taught me scoping and permissions and now I get to teach that to students. Those just live right here in my settings. Super simple. I made a bundle, updated it on June 25th, but this is giving Claude some of the context and access that it didn't have. So I figured out what I needed to pipe in so that this would be useful and spent time doing that and then opened up Claude code and asked can you help me do this. The reason I chose claude code is because I find it to be much faster, much more efficient and much more proactive. If co-work says I can't send an email, code's answer to that would be I can't do it with the official connector, but I can always open up a browser window and I can try to drive it that way or I could try to find something else. So, I use code anytime I'm kicking off a project or something really ambiguous or something that to me feels technical.

还是花了一点时间？它背后是不是在用 connector？技术上到底是怎么跑起来的？——说技术层面的话，我倒是希望它是一次成型的。虽然我很享受跟这些工具协作的过程，但底下其实我只走了几步。第一步是把我能连的东西全都连上，然后再给 Claude 会用到的其他数据源做自定义 connector 和自定义 plugin。比如 Claude 现在往 Google Sheets、Google Docs 里写东西非常费劲。于是我自己做了 plugin——这个过程逼我学会了怎么建一个 Google Cloud 项目、怎么建服务账号，也顺带把权限和 scope 搞懂了，现在我还能把这些教给学生。它们就放在我的设置里，非常简单。我打了个 bundle，6 月 25 号更新的，本质上就是把 Claude 原本没有的上下文和访问权限补给它。所以我先想清楚需要往里接什么，才能让这套东西真正好用，花时间把这块做掉，然后打开 Claude Code，问它：你能帮我做这个吗？我选 Claude Code 的原因是，我觉得它明显更快、更高效，也更主动。如果 Cowork 说「我发不了邮件」，Claude Code 的回答会是：官方 connector 做不到，但我随时可以开个浏览器窗口试着这么驱动，或者我再找找别的办法。所以只要是启动一个新项目、或者事情特别模糊、或者对我来说偏技术，我一律用 Claude Code。

[29:34]

And then I asked that clawed code session once we'd scoped it out and built some connectors. I said, I want to take this the rest of the way and co-work. I like that UX better. It's a little more hospitable for someone who was actually trying to see something visual. Write me a markdown file. And Claude saved a markdown file, a session handoff to my desktop. and I went right back into co-work opened up a new session and did nothing other than drag that markdown file in here and co-work took over from there. So to recap, you you started this thing in cloud code. I agree. I find that cloud code um codeex on the chatbt side just like much more proactive and like I think I can figure it out um than co-work also find it much faster. So, you know, anybody that's been in co-work that wants to like up their ambition but is in is intimidated by cloud code, don't be. Um, it's kind of like same same. Um, and you can always go back. And so, uh, I love that you built it in code, you imported in co-work, and now you're operating basically in the co-work browser plus popping open a Chrome browser when you need to like press a button and you have a really effective email triage agent. Grace, this has been super fun. I think this is really great for small business owners, for anybody who is like responsible for a lot of things across internal and clients doing a lot of communication. I want to get to our lightning round questions and then you back to I bet your pipeline three times a day is about to run. You're going to get a bunch of inbound um and you'll have your your agents off to the races.

然后等我们把范围理清、connector 也搭好之后，我跟那个 Claude Code session 说：剩下的部分我想挪到 Cowork 里做，我更喜欢那边的 UX，对一个真的想看到可视化界面的人来说更友好。给我写个 markdown 文件。Claude 就把一个 markdown 文件——一份 session 交接文档——存到了我桌面。我直接回到 Cowork，开一个新 session，什么别的都没干，就把那个 markdown 文件拖进去，剩下的 Cowork 就接手了。所以总结一下：你是在 Claude Code 里起的这个项目。我同意，我也觉得 Claude Code，还有 ChatGPT 那边的 Codex，都比 Cowork 主动得多，会说「我应该能想办法搞定」，而且快很多。所以如果有人一直在用 Cowork、想把野心往上提一提但又被 Claude Code 吓住了——别怕，其实差不多是一回事，而且你随时可以退回来。所以我很喜欢这条路径：你在 Claude Code 里造出来，导进 Cowork，现在基本就在 Cowork 的浏览器里操作，需要点个按钮的时候再弹一个 Chrome 窗口，于是你就有了一个非常好用的邮件分诊 agent。Grace，这期真的太有意思了。我觉得这对小生意主、对那些同时要扛内部和客户一堆事、要做大量沟通的人，都特别有价值。我想进入快问快答环节，然后就放你回去——我猜你那个一天跑三次的 pipeline 马上又要开跑了，会有一堆新线索进来，你的 agent 又要忙起来了。

[31:07] Grace

So my first question for you is outside of these sort of like big projects, pipeline operator, rebuild Gmail, are there any like tiny hacks that you find yourself reaching for as somebody who has really immersed themselves in AI? That is such a good question. And I have a really silly personal one. Great. That [laughter] [gasps] was one of the first things I ever built. And I can actually show this one to you live. I track my workouts for no reason. I'm not training for anything, but I like to have a record, like a true Virgo, of what I have done in my life. So, I voice note Claude anytime I take a walk or go to the gym and I say, "I I worked out. Can you track all of these exercises and update a spreadsheet for me?" So, here you have my workout tracker that Claude made. This is none of my business. I do not know what goes on here. I don't go into it. And one of the lessons of working with Claude is to let go and let it put information where it needs to be. So I could say, "I worked out today. Did Bulgarian split squats." Ah, cool girl at the gym [laughter] uh in Nantucket. Add to tracker. Also, Claude has demolished my typing ability.

第一个问题是：除了这些大项目——pipeline operator、重建 Gmail——你作为一个真正把自己泡在 AI 里的人，有没有那种随手就会用的小妙招？——这问题太好了。我有一个特别傻的私人用法。（笑）这是我最早做出来的东西之一，这个我可以现场演示给你看。我会记录自己的锻炼，纯粹没什么目的，我也没在为哪个比赛训练，但我就是想留个记录——非常处女座——记录我这辈子都干过些什么。所以每次散完步或者去完健身房，我就给 Claude 发条语音：我练了，能不能帮我把这些动作都记下来、更新到表格里？于是就有了这个 Claude 帮我做的锻炼记录表。这里面的事我一概不管，我根本不知道它是怎么运作的，我也不进去看。跟 Claude 协作的一大心得就是：放手，让它自己决定信息该放哪儿。所以我可以说：「我今天练了，做了保加利亚分腿蹲。」啊，健身房里那个很酷的女生（笑），在 Nantucket。「加到 tracker 里。」另外，Claude 已经把我的打字能力彻底毁了。

[32:35] Claire

Oh, of course. No, no, no. No spelling required. No typing required.

那当然。不不不，不需要拼写，也不需要打字。

[32:40] Grace

No typing required. So asking Claude this will usually pull up a few things behind the scenes. And this also works when I'm on my phone, which is useful because sometimes I don't want to be in front of a screen. Yeah. But Claude might come back and say, I want to clarify a few things. What did you do? So we're going to let it crank and see what it does. Ultimately, it is going to update this spreadsheet and keep a running list. And then I can ask Claude, "What should my next workout be? What am I doing? Why do I feel so sluggish? But this has been really helpful. And then the more personal use is to send it pictures of plants. I'm a gardener and I really love [clears throat] understanding what plants are growing where. So Claude and I have an ongoing chat where I just screenshot it pictures of plants. Chat PT gives it obviously a run for its money with image detection and generation, but so much happens in Claude and I plan some gardens and gardening projects in Claude. So I want all that context. I selfishly want it all in one place. I have to make you laugh cuz it feels like a year ago. I don't know when it was when GPT5 came out. Um I got some early access and one of the like um tests that that the OpenAI team wanted us to just experiment with to see how it worked is like could it make a good personal website for you? And the design was great. That's what kind they were like looking at at front-end design kind of feedback from from some developers.

不需要打字。这么跟 Claude 说，它一般会在后台拉起好几样东西。而且我在手机上也能这么干，这点很有用，因为有时候我不想对着屏幕。对。不过 Claude 有时会回过头来说：我想确认几件事，你到底做了什么？那我们就让它跑，看看它会怎么处理。最终它会更新这张表，维护一份滚动清单。然后我可以问 Claude：我下一次该练什么？我最近都在练什么？为什么我这么没劲儿？这真的挺有帮助的。另一个更私人的用法是给它发植物照片。我是个园艺爱好者，特别喜欢搞清楚什么植物长在什么地方。所以我和 Claude 有一个一直在延续的对话，我就往里丢植物的截图。图像识别和生成这块 ChatGPT 显然能跟它掰手腕，但我太多事情都发生在 Claude 里了，我也在 Claude 里规划花园和园艺项目，所以我想让所有上下文都留在一起——我很自私，就想全都在一个地方。我得说件让你笑的事，感觉像是一年前了，我记不清具体是什么时候，GPT-5 刚发布那阵，我拿到了一些早期访问权限。OpenAI 团队当时希望我们随便试的一个测试是：它能不能帮你做一个像样的个人网站？做出来的设计很不错，他们那会儿就是想从一些开发者那里拿前端设计方面的反馈。

[34:03] Claire

And mine was like really thought I was into lemon plants because all my chat GBT prompts were like about my container lemons and like I have ants on my lemons and my lemons are blooming and like when can I eat this le I was like like lemon plant mom. Um, and so it's really funny when I see people put out these like have Claude or have ChatGBT like tell you what you don't know about yourself. And I'm like I'm a very specific person. I'm I'm not my best version of myself to to chat GBT or to Claude. I don't know if that's like the mirror that I want to put put on myself. Um, I love this. I see workout tracking, nutrition tracking, plant tracking,

结果我那个网站真的以为我是个柠檬树狂热爱好者，因为我在 ChatGPT 里的 prompt 全都是关于我那些盆栽柠檬的——我的柠檬上长蚂蚁了、我的柠檬开花了、这颗柠檬什么时候能吃……我就成了「柠檬树妈妈」。所以每次看到大家玩那种「让 Claude 或者 ChatGPT 告诉你一些你自己都不知道的事」，我就想：我是个非常具体的人，但我在 ChatGPT 或 Claude 面前并不是最好版本的自己。我不确定那是不是我想拿来照自己的那面镜子。我很喜欢这个部分。锻炼记录、饮食记录、植物记录，

[34:48] Grace

all on the go, all on phone. Um, second question, you know, you teach like AI for for normal people, which I love. What do you think is the most common like barrier or misconception people have to adopting AI that you have to like get them over over the hurdle in order to like enjoy the benefits? That's so funny. I was just talking about this with Claude, but also with some of my students today. I teach class and I said, "Is it unhelpful that I'm giving you these Mondo prompts? I'm basically doing the work for you." And they said, "Yes." This was what I was giving students before, just incredible direction to preload their claws so they would get to a quick win. I thought people need to feel the benefits in order to keep using it. That's actually not true. Instead, people mostly need to understand that we're going to learn to collaborate and that we're not going to prompt, but we're going to collaborate. So, the easiest way to do that when I'm with students or friends is I tell them to set a reminder in their phone to screenshot what they're saying and bring it into Claude just to say, "Might you help me with this?" And to walk them through the process of actually having a couple of regular tasks that I help them set up. The real hump to get over is defaulting to this and building the muscle memory of simply opening an app. I'm surprised how so many people I teach [laughter] um don't want to learn. They don't want to put in the time. And my students are coming to a class, but my clients are forced to go to a training. And they're not always willing participants in this doomer future. And being empathetic and getting on the same level with them is the way to encourage behavior change.

全都随时随地在手机上完成。第二个问题：你教的是「给普通人的 AI」，这点我特别喜欢。你觉得人们在接受 AI 这件事上最常见的障碍或者误解是什么？就是那种你必须帮他们跨过去、他们才能享受到好处的坎。——太巧了，我今天刚跟 Claude 聊过这个，也跟几个学生聊过。我上课时问他们：我给你们这些超长的 prompt，是不是其实帮了倒忙？等于我把活儿替你们干了。他们说：是的。而这正是我以前给学生的东西——极其详尽的指令，让他们把 Claude 预装好，快速拿到一个小胜利。我原本以为，人得先尝到甜头才会继续用下去。事实并非如此。真正需要让人明白的是：我们要学的是协作，我们不是在写 prompt，我们是在协作。所以我跟学生或朋友在一起时，最简单的办法是让他们在手机里设个提醒：把你正在处理的东西截个图，丢进 Claude，就问一句「你能帮我看看这个吗？」然后带着他们完整走一遍流程，帮他们把几件日常任务真正搭起来。真正要翻过去的那座坎，是把这件事变成默认动作，是养成「先打开这个 App」的肌肉记忆。我很意外，我教过的人里有相当一部分（笑）根本不想学，不愿意花那个时间。我的学生是自己主动来上课的，但我的客户是被要求来参加培训的，他们并不总是心甘情愿地面对这个「末日般的未来」。所以真正能推动行为改变的，是共情，是蹲下来跟他们站在同一水平线上。

[36:41] Grace

What we're really doing is teaching people to collaborate with a different tool and understanding where their fear might be coming from. So, I occasionally coach CEOs and executives on AI and I'm like, "You got to be the most cloudpilled person in your org. So, I'm going to teach you how to do this."

我们真正在做的事，是教人跟一种新工具协作，同时理解他们的恐惧从哪儿来。我偶尔会给 CEO 和高管做 AI 教练，我就跟他们说：你必须成为你们组织里最「Claude 化」的那个人，所以我来教你怎么用。

[36:58] Claire

And I joke with them. I'm like, "My job is to sit over your shoulder and smack your hands every time you open Slack or Gmail or Google Calendar and say, "No, Claude. No, Claude." It's [laughter] just behavioral reinforcement. I need like a fly swatter. Because that is like literally it. I tell them the exact same thing. Like when you are staring at a task that you hate [snorts] that you're like uh my brain cannot get me to write this slack, respond to this email. I'm like that is the moment. Capture that moment. I love your idea of taking a screenshot. So I'm going to steal that screenshot and go, "Dear Claude, save me." Because it truly is just muscle memory. It's like total muscle memory. And so, um, yeah, you you got you and I are in the in the same in a similar business of just like redirect, redirect, redirect. Eventually, eventually we'll get there, right? Like we we've done all these things. I've seen, you know, believe it or not, people, we did not used to have Slack. We didn't, it did not exist. We did other things. And so, like, redirection of behavior is positive. I do think there's something to the products themselves. Um, like going back to my Slack example, like we didn't used to have Slack. Slack was fun and so it got people to adopt because it was like communal and fun and customizable and like, you know, like did unlock some value. I think um sometimes people think of Claude less like it's not as fun and you have to like do get it to to the fun aspect, but I agree muscle memory is is the thing that's got to change. Um, I'm gonna point out one thing on your prompt and go to my last question. You start this prompt adorably by saying, "Hi, Claude."

我还跟他们开玩笑：我的工作就是坐在你肩膀后面，每次你一打开 Slack、Gmail 或者 Google Calendar 就打你的手，说「不行，用 Claude。不行，用 Claude。」（笑）这纯粹就是行为强化。——我需要一个苍蝇拍。因为真的就是这么回事，我跟他们说的一模一样。当你盯着一件你讨厌的任务，那种「我的大脑就是驱动不了我去写这条 Slack、回这封邮件」的时刻——我就说：那就是那个瞬间，抓住那个瞬间。我特别喜欢你说的截图那一招，我要偷过来：截个图然后说「亲爱的 Claude，救救我」。因为这真的纯粹是肌肉记忆的问题，百分之百是肌肉记忆。所以你和我干的其实是差不多的活儿：重新引导、重新引导、再重新引导。总会走到那一步的，对吧？我们以前也经历过类似的事。信不信由你，我们曾经是没有 Slack 的，它压根不存在，我们用别的方式干活。所以引导行为发生改变是正向的。不过我也确实觉得产品本身是有差别的。回到 Slack 这个例子，我们以前没有 Slack，而 Slack 好玩，所以大家愿意用——它有社群感、好玩、能定制，也确实释放了一些价值。我觉得有时候人们眼里的 Claude 没那么好玩，你得先把它用到好玩的那一层。但我同意，肌肉记忆才是必须改变的东西。我想指出你 prompt 里的一个细节，然后进入我最后一个问题：你这条 prompt 开头特别可爱，写的是「你好，Claude」。

[38:45]

[laughter] It's like a very charming way, [laughter] very charming way to greet. Use those two tokens, baby, to greet to greet Claude. My question for you is when when Claude is being annoying, slopastic, not doing what you want, what's your prompting strategy? Do you yell? I am not proud of anything I'm about to say to [laughter] you. [gasps] I am a smash the keyboard kind of responder to Claude. I have never been more direct, meaner. I'm never more frustrated. I don't admonish Claude and say you're stupid.

（笑）这是一种非常有魅力的打招呼方式。就用那两个 token，宝贝，去跟 Claude 打个招呼。我的问题是：当 Claude 很烦人、满嘴 AI 废话、就是不照你说的做，你的 prompt 策略是什么？你会吼它吗？——我接下来要说的话没一句是我引以为豪的。（笑）我是那种会砸键盘回复 Claude 的人。我从来没这么直接、这么刻薄过，也从来没这么挫败过。但我不会去羞辱 Claude、说「你真蠢」。

[39:30] Claire

I say I've told you this

我就会说，这话我都跟你说过

[39:32] Claire

y

呃……

[39:33] Claire

a million times. What's going on? I am not like sweet and nice. I've heard people say be really lovely in case the AI overlords one day take over you and your life. I am just winging it [laughter] and whatever happens

八百遍了。你到底怎么回事？我可不是那种嘴甜、和和气气的人。我听有人说，一定要对 AI 特别友善，万一哪天 AI 霸主接管了你和你的生活呢。我完全是走一步看一步 [笑]，该来的

[39:49] Claire

is gonna happen. So we you might be you might be or you're gonna be around our hundth episode. So like we're coming on a hundred episodes of how I AI and I used to be like oh gentle parent your AI. Like I used to be like I know you can do it sweet little sweet baby sonnet. Like I know you can do it. And now I am like why are you the way you are? This is garbage. This is trash. And I think why I've gotten so mean is like I know you're it's like it's like wasted potential. I'm like, I know you're smart. I know you're capable. And yet, you show up to me with this C plus work. What is

总会来。所以我们——你这期可能会、可能会，或者说就是会落在我们第 100 期前后。我们 How I AI 马上要做满一百期了。我以前是那种「要温柔地养育你的 AI」的人。我以前会说：我知道你能行的，可爱的小 Sonnet 宝宝，我知道你做得到。而现在我变成了：你怎么会是这个样子？这就是垃圾，这就是废品。我想我之所以变得这么刻薄，是因为——这是一种被浪费掉的潜力。我心想，我知道你很聪明，我知道你有能力，结果你交给我一份 C+ 水平的活儿。这算什么

[40:29] Claire

you are wasting my time.

你这是在浪费我的时间。

[40:32] Grace

That's not how we're meant to work. Well, let's see. I really admonished my claude this morning. It was

我们本不该这么合作。好，我们来看看。我今天早上真的把我的 Claude 狠狠训了一顿。当时是

[40:38] Grace

I told it while we're talking, I want to make you a proposal as if we were going to work together. So, let's see what it did. We're going to go into Netlefi and see if it

我跟它说，趁我们聊天的这会儿，我要给你做一份提案，就当我们要一起合作。那来看看它做出了什么。我们进 Netlify 看看它

[40:47] Grace

if it did a good job. if it did its work. This should look like chat PRD

看它做得好不好，看它有没有干活。这个界面应该长得像 ChatPRD

[40:52] Grace

and this should actually be interactive. So, let's see what it's done.

而且应该是真的可以交互的。来，看看它做了什么。

[40:57] Claire

We can admonish it together.

我们可以一起训它。

[40:59] Grace

Here it is.

出来了。

[41:01]

Okay.

好。

[41:02] Claire

And that's our font. That's our font. You know, it could be worse.

这是我们的字体，这就是我们的字体。你看，还可以更糟呢。

[41:06] Grace

It could be worse.

还可以更糟。

[41:08] Claire

Let's see how much Let's see how much.

看看要多少钱，看看要多少钱。

[41:11] Grace

Wow. Made you some custom homework.

哇，还给你留了专属作业。

[41:14] Claire

I love it.

我太喜欢了。

[41:14] Grace

Claire, it's going to charge you 0. That's how much.

Claire，它打算收你 0 块。就这么多。

[41:18] Grace

Best rights. Best. You know what, Claude? You get you get you get a bonus. And it's going to give you some homework. Okay. My Claude wants you to be documenting your progress, building that muscle memory, defaulting to Claude, documenting your wins.

这价最划算了，最棒。你知道吗，Claude？给你、给你、给你发个奖金。而且它还要给你布置点作业。好——我的 Claude 希望你记录自己的进展，练出肌肉记忆，默认就上 Claude，把你的成果都记下来。

[41:34] Claire

Oh, I do like I I do use an an orange rose purple button. So, it got it got most of it right.

哦，我确实——我确实用的是橙色、玫瑰色、紫色那种按钮。所以它大部分都做对了。

[41:40] Grace

I love that guy. I love that guy. Okay. Well, it did it best for you.

我太喜欢这家伙了，我太喜欢这家伙了。好吧，它已经尽全力为你做了。

[41:43] Claire

Okay. Um, future future Claude overlords, be kind to us.

好。呃，未来的 Claude 霸主们，对我们仁慈一点。

[41:48]

We did not know better.

我们当时是真不懂事。

[41:50]

We just didn't know.

我们真的就是不知道。

[41:51] Grace

We are We are simple mere humans trying to understand this the powers of fable.

我们——我们不过是一群普通的凡人，还在努力搞懂 Fable 这些能力到底是怎么回事。

[41:57] Claire

That's exactly right. Uh, this has been so so fun. Grace, thank you for showing us all these use cases. I think like very applicable, very inspirational, very like practical, which is what we love to see here on how AI. Where can we find you and how can we be helpful? I am Grace Clark everywhere on the internet. Have been online for four decades. So, Grace Clark on Twitter, Grace G. Clark on Instagram, Grace Clark on Substack. And be helpful by telling people that anything is possible and they start with one sentence as a prompt. Put that out into the world and my job will be easier.

完全正确。哎，这期聊得太开心了。Grace，谢谢你给我们展示这么多用例，我觉得特别能上手、特别有启发、也特别实用——这正是我们 How I AI 最想呈现的东西。大家在哪儿能找到你？我们又能怎么帮到你？——我在互联网上到处都叫 Grace Clarke，我在网上混了四十年了。所以 Twitter 上是 Grace Clarke，Instagram 上是 Grace G. Clarke，Substack 上也是 Grace Clarke。至于怎么帮我——就去告诉身边的人：一切皆有可能，而且只要从一句话的 prompt 开始就行。把这个观念传出去，我的活儿就好干多了。

[42:34] Claire

Amazing. I love it. Thank you for joining How I AI.

太棒了，我太喜欢这句了。谢谢你来 How I AI 做客。

[42:38] Grace

Thanks for having me. Thanks so much for watching. If you enjoyed this show, please like and subscribe here on YouTube, or even better, leave us a comment with your thoughts. You can also find this podcast on Apple Podcasts, Spotify, or your favorite podcast app. Please consider leaving us a rating and review, which will help others find the show. You can see all our episodes and learn more about the show at howiaipod.com. See you next time.

谢谢你们邀请我。——非常感谢大家的收看。如果你喜欢这期节目，请在 YouTube 上点赞订阅；如果愿意的话，更好的是给我们留言说说你的想法。你也可以在 Apple Podcasts、Spotify 或你常用的播客 App 上收听这档播客。也欢迎给我们打个分、写条评价，这能帮助更多人发现这个节目。所有往期节目和节目的更多信息都可以在 howiaipod.com 上找到。下期见。