I built a Claude Cowork system that does a week of PM work in a day

频道: How I AI
视频: https://www.youtube.com/watch?v=p2qmX6TM0kw
原文语言: en
统计: 共 63 轮 · Daniel 28 · Claire 24

[0:00]

I cannot describe how powerful it is. I'm really able to do in a day now what used to take me a week. But more than that, I'll say it's a lot of deeper work. There's things that I do today that I just wasn't able to do before. Whether it's more in-depth research, whether it's being more databacked, anything you do today, you can do it almost at the top level. One of the things even at a company level that I tell people is look, you got to go through the pain of switching your systems and it's not going to be as efficient. And yes, of course, like the old way will be faster because it's muscle memory, but if you get to the next level, not only will you have a system that works better for you, but everybody will have more skills.

Daniel： 它有多强大，我真的没法形容。以前要花我一周的活，现在我一天就能干完。但不止如此，我想说的是，我做的深度工作也多了很多。有些事我现在在做，以前根本做不到。不管是更深入的调研，还是更有数据支撑，你今天做的任何事，几乎都能做到顶尖水准。

Claire： 哪怕在公司层面，我也常跟大家说：听着，切换系统的阵痛你得熬过去，而且一开始效率不会那么高。当然了，老办法肯定更快，因为那是肌肉记忆；但只要你迈上下一个台阶，你不光会拥有一套更适合自己的系统，每个人也都会拥有更多 skill。

[0:42] Daniel

Today, I say I manage about 70 to 80% of my time in front of the computer with co-work alone.

可以说，现在我在电脑前的时间里，大约有 70% 到 80% 光靠 Cowork 就能打理。

[0:47]

We've seen a couple morning briefs, but I want to see what you think makes yours different. In this daily brief, my coworker runs through my recent slacks, my emails, my personal [music] notes, and it looks for context that it doesn't understand in these things. And it says like, oh, if there's a file I wasn't aware of, there's a milestone, there's a goal, it would ask me about it and say, hey, like, what is this? Is this important? Should I read this? And this is, I think, where the real magic happens.

Claire： 晨间简报我们已经见过好几个了，但我想看看你觉得自己的有什么不一样。

Daniel： 在这份每日简报里，我的 Cowork 会过一遍我最近的 Slack 消息、邮件和个人笔记，从中找出它看不懂的上下文。然后它会说：哦，这儿有个我之前不知道的文件，有个里程碑，有个目标——它就会来问我：嘿，这是什么？重要吗？我该读一下吗？我觉得，真正的魔法就发生在这里。

[1:10] Claire

This is really genius. I'm curious how this scales beyond you. So, you have your system, you're, you know, 10x PM now. Congratulations. How does this affect the team? How do they use systems like this? 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 Daniel Bloom and he's going to show us his clawed co-work system to rule them all. It self-heals, it self-optimizes, and it's way better than the morning briefing you have going on. Let's get to it. Optimizely agent platform for marketers is built on a bigger idea. A whole directory of agents now joined by a newly revealed squad of virtual teammates. Each with a defined role and a personality of their own. This isn't about faster drafts. It's about handling the coordination, too. The research, the approvals, the handoffs, the routing, the back and forth that quietly eats your week and never makes it into anyone's highlight reel. Get the story behind it all. Hear the confessions from marketers who have handed off the busy work. Meet the squad and understand what virtual teammates can do for you at [music] optimizely.com/how ii ai. Daniel, welcome to how I ai. I am so excited to have you on this episode because every week, every month, I think things get better and I'm always impressed with how much more organized everybody but me seems to get with their uh AI and their personal productivity.

这真的太天才了。我很好奇它怎么推广到你之外的人身上。你有了自己的系统，现在也算是 10 倍 PM 了，恭喜。那这对团队有什么影响？他们是怎么用这类系统的？欢迎回到 How I AI。我是 Claire Vo，产品负责人、AI 狂热分子，我的使命是帮你用这些新工具把东西做得更好。今天的嘉宾是 Daniel Blum，他要给我们展示他那套“一统天下”的 Claude Cowork 系统。它能自我修复、自我优化，比你现在跑着的晨间简报强太多了。咱们开始吧。Optimizely 面向营销人员的 agent 平台，建立在一个更宏大的构想之上：一整套 agent 目录，如今又迎来一支新亮相的虚拟队友小队，每位队友都有明确的角色和自己的个性。这不只是为了更快出草稿，更是要把协调工作也接过来——调研、审批、交接、流转分派，还有那些来来回回的拉扯，它们悄无声息地吃掉你的一整周，却从来进不了任何人的高光集锦。来了解这一切背后的故事，听听那些已经把琐碎杂活交出去的营销人的真心话，认识这支小队，看看虚拟队友能为你做什么，尽在 optimizely.com/howiai。Daniel，欢迎来到 How I AI。能请你来上这期节目我特别兴奋，因为我觉得每周、每个月情况都在变好，而且我总是很惊叹：除了我以外，好像每个人在用 AI、搞个人效率这件事上都变得越来越有条理了。

[2:53] Claire

And so tell me like what brought you to building this big system and what problem are you trying to solve?

那跟我说说，是什么促使你搭了这么一套大系统？你想解决的是什么问题？

[3:00] Daniel

Cool. Well, Claire, thank you so much for having me. I'm really excited to be here. As somebody who's been an early adopter of AI for quite a while, I think it took a already like a a good few months ago or even a year ago, we were able to solve some of the big or at least improve some of the big jobs to be done for PMS with uh even with simpler tools like with uh gems. Uh you could improve your specs and you could improve your research. But I think uh as I was progressing with my AI adoption, I really felt that one of the biggest things that hangs over me and and really makes my day-to-day more difficult is all this just overhead of of being a PM of coordinating a bunch of things, remembering endless tasks that arise on Slack and meetings and action items and just trying to maneuver this whole uh this whole chaos that happens every single day. um while at the same time wanting to do in-depth work and focus and and really think I felt that that was something that was really really difficult for me and I was getting nowhere near solving it um until I was able to actually build a system with co-work that I'm actually still a little bit surprised at how powerful it is and how much it's helped transform uh this this problem for me.

好的。Claire，非常感谢你邀请我，能来这儿我真的很兴奋。我算是很早就开始用 AI 的人，用了挺长时间了。我觉得早在好几个月前、甚至一年前，我们就已经能解决——或者至少改善——PM 们的一些核心 jobs to be done，哪怕用的只是 Gems 这类比较简单的工具。你可以把 spec 写得更好，也可以把调研做得更好。但随着我用 AI 越来越深入，我真切地感觉到，压在我头上、让我的日常工作变得更难的最大一件事，就是当 PM 的这一大堆额外负担：要协调一堆事情，要记住 Slack 上、会议里冒出来的没完没了的任务和待办事项，还得在每天都会发生的这一整团混乱里周旋。与此同时，我又想做深度工作、想专注、想真正静下来思考。我觉得这对我来说真的非常非常难，而且我离解决它还差得远——直到我真正用 Cowork 搭起了一套系统。说实话，它有多强大、帮我把这个问题改变了多少，到现在我都还有点惊讶。

[4:14] Claire

So tell me about your personal AI stack as a PM. How did you pick the tools that go into it? And then I want to I want you to show us how how it all kind of works.

那跟我聊聊你作为 PM 的个人 AI 工具栈吧。里面的工具你是怎么挑的？然后我想请你给我们演示一下，这一整套大概是怎么运转的。

[4:26] Daniel

Perfect. Uh so I'll say I think most of my choices were made by my company Mio. Um and the tools that I'm uh I'm are in our stack. Uh but uh thankfully I'm I'm pretty satisfied with those. So for a long time we only had like the Gemini enterprise and we were using that and uh just a few months ago we got uh the claude enterprise and co-work uh and that's when things really changed for me. Uh and besides that I'm basically bound to whatever whatever tools we have and use uh so this is this is what I work based off of.

好。我得说，我的大部分选择其实是我们公司 Melio 替我做的，就是我们技术栈里现有的那些工具。不过好在我对这些工具还挺满意的。很长一段时间里，我们只有 Gemini Enterprise，一直在用它；就在几个月前，我们开通了 Claude Enterprise 和 Cowork，从那时起，一切对我来说才真正变了。除此之外，我基本上就被限定在公司现有、在用的那些工具里，所以我的工作就是以这些工具为基础的。

[4:59] Claire

Okay. So this is good for everybody out there who is stuck with the tools that they're stuck with. You can still make it work. I at work. I think that Claude is obviously like a big part of this stack. Was co-work just the big unlock for you? Like where did you when did you start building this and what was the early tools versus what you use now?

好。这对所有被现有工具绑住、没得选的人来说是个好消息——你照样能把它用起来。我在工作中……我觉得 Claude 显然是你这套工具栈里很重要的一部分。Cowork 就是你的那个关键突破口吗？比如你是从哪儿——从什么时候开始搭这套东西的？早期用的工具和你现在用的有什么区别？

[5:20] Daniel

Yeah. So I think first of all on your first saying I think it's very correct and in general I'll say like I was bound by constraints of an average PM all the time in terms of having you know not endless tokens and needing to work within budgets and bureaucracy and I think it's very possible and very doable within all those constraints that you know real PMs face and I think it's it's very important to say um and in terms of the tool so co-work was definitely the big unlock for me but I While I love co-work, it's not co-work in itself. Uh, and I think codeex could be great for this as well or uh, Chad GBT work. I think the real importance is for a system to a be able to rewrite its own uh, core files and that means that it keeps on improving and b just have connections and integrations to as much of your ecosystem. when these two rules apply um there is a significant significant opportunity to really really improve the way you work and while I was working with uh Gemini gems earlier and there's obviously some incredible things that can be done with those these two elements make a system really powerful and also really integrated into your daily life but also self-improving which is one of the biggest unlocks here

对。首先，关于你刚才说的第一点，我觉得非常对。总体来说，我一直都受限于一个普通 PM 会碰到的各种约束：token 不是无限的，还得在预算和官僚流程的框架里干活。而我觉得，在真实 PM 面对的所有这些约束之内，这件事完全有可能、完全做得到——我觉得这一点非常重要，一定要说清楚。至于工具，Cowork 对我来说绝对是那个关键突破口，不过——虽然我很爱 Cowork，但关键并不在 Cowork 本身。我觉得 Codex 用来干这个可能也很棒，或者 ChatGPT work 也行。我认为真正重要的是一套系统要做到两点：第一，能改写它自己的核心文件，这意味着它会不断改进；第二，尽可能多地连接、集成你的整个工具生态。只要满足这两条，你就有非常非常大的机会真正大幅改善自己的工作方式。我之前用过 Gemini 的 Gems，用它们显然也能做出一些很了不起的东西，但正是这两个要素，让一套系统真正强大、真正融入你的日常生活，同时还能自我改进——这正是这里最大的突破之一。

[6:36] Claire

okay well let's let's show show it to us I want to I want to see how you actually manage your life as a PM.

好，那就给我们演示一下吧。我想看看你作为 PM，实际上是怎么打理自己的日常的。

[6:43] Daniel

Let me start with this uh notion board, right? So, this is uh nothing incredible. It's just a standard uh PM notion board. It has three sections. One is top of mind, which are the large initiatives uh that I I just think of uh you know on any given uh time. It has my this week, which is basically the priority that I have for this week. And it has inbox, which is things that arise from Slack, email, calendar, and co-work picks up and fills them in here. Now, the cool thing about this board is not that it exists, but the fact that not only did Coowork build this completely on its own, I was working with a very simple and messy uh Google doc before, and I think notion got I think co-work got tired of of it and basically decided to build this for me, but also it actually manages it on my behalf. And I think that's the pretty cool thing that I want to show and I'll walk you through how it does it. I'll say that um the most important thing about this is that it knows me really really well uh because I spent a lot of time uh and effort contextualizing it uh and maintaining this context and also centralizing all of my work in it which took some time to get used to. But the ROI is just absolutely insane there. And I really felt how well it knows me just when prepping for this uh for this episode when I knew that because this is my real environment I couldn't show you know names and projects and sensitive stuff. So basically I just created this skill on cloud called demo and I just told it replace all the terminology and like specifics with generic terms. But because it knows me so well basically in in one prompt I was able to replace my manager's name with your manager and my top initiative. You can see here with headline feature and that's how well it knows me. It was just able to replace all of these and it just makes it for a really really seamless and comfortable collaboration.

我先从这个 Notion 看板说起吧。它本身没什么了不起的，就是一个很标准的 PM 用的 Notion 看板，分三个板块。第一个是 Top of Mind，放的是我随时都会惦记着的那些大型项目。第二个是 This Week，基本上就是我这周的优先事项。第三个是 Inbox，放的是从 Slack、邮件、日历里冒出来的事，Cowork 会把它们捞出来填到这里。这个看板酷的地方不在于它存在，而在于：一方面，它完全是 Cowork 自己搭出来的——我之前用的是一个非常简陋又乱糟糟的 Google Doc，我觉得 Notion……不对，我觉得是 Cowork 受够了那个文档，干脆决定给我搭了这个；另一方面，它还真的会替我管理这个看板。我觉得这是挺酷的一点，我想演示给大家看，也会带你们过一遍它是怎么做到的。我想说，这里面最重要的一点是它非常非常了解我，因为我花了大量时间和精力给它补充上下文、维护这些上下文，还把我所有的工作都集中到它里面，这个我是花了一段时间才适应的。但这方面的 ROI 简直高得离谱。我真正体会到它有多了解我，是在为这期节目做准备的时候：我知道因为这是我真实的工作环境，我不能把人名、项目和那些敏感信息给大家看。所以我就在 Claude 上建了一个叫 demo 的 skill，只跟它说：把所有术语和具体细节都换成通用说法。因为它太了解我了，基本上一个 prompt 下去，它就把我经理的名字换成了“your manager”，把我的头号项目——你们在这儿能看到——换成了“headline feature”。它就是这么了解我。它能把这些全都替换掉，这让我们的协作变得非常非常顺畅、舒服。

[8:37]

Tell me a little bit about how that context is saved. Is it just that you just put everything in notion so when it's searching it kind of knows? Is there something more purposeful around your kind of personal brain? How do you how do you inject context into the system? So I'll say when I built up the system, it was very important for me to build something scalable. So I did a lot of research on like how to build uh how to set up your cloud MD and the and the context files. Um and basically I set it up so there's a context file around every uh every topic that uh that I have and like every area of my work uh and my goals and my uh and you know my colleagues and everything that you can imagine. I spent a lot of time manually contextualizing in the beginning, whether through like feeding it links and and different decks and stuff like that. And also just a lot of, you know, whisper just dictating endless information and blabbering endlessly about everything that I can think of, which is just incredibly powerful on how much context you can give relatively easily and quickly. Um, and then this system basically updates itself on a recurring basis. every uh every few weeks it runs and updates the uh knowledge files based on you know different things that have occurred um and different uh changes and we will see some of that uh just in a minute.

Claire： 跟我讲讲这些上下文是怎么保存的。是不是你就把所有东西都放进 Notion，这样它搜索的时候大概就知道了？还是说围绕你的“个人大脑”，有什么更刻意的设计？你是怎么把上下文注入到这个系统里的？

Daniel： 是这样，我搭这套系统的时候，很看重要做出一个可扩展的东西。所以我做了大量研究，看怎么搭建、怎么设置 CLAUDE.md 和那些上下文文件。基本上我是这么设置的：我手上的每一个主题都有一个对应的上下文文件，比如我工作的每个领域、我的目标、我的同事，你能想到的一切都有。一开始我花了很多时间手动补充上下文，比如给它喂链接、喂各种 deck 之类的。另外就是大量用 Whisper 口述，没完没了地往里灌信息，把我能想到的所有事情絮絮叨叨讲个遍——能相对轻松、快速地提供这么多上下文，这一点真的强大得不可思议。然后这套系统基本上会定期自我更新：每隔几周它就会跑一次，根据期间发生的各种事情、出现的各种变化去更新知识文件，我们待会儿马上就能看到其中一部分。

[9:57] Claire

I love it. Well, let's let's switch over to co-work and show us how kind of the system is architected and then how you maintain it and use it over time.

太喜欢了。那我们切到 Cowork 吧，给我们看看这套系统大致是怎么架构的，以及你是怎么长期维护和使用它的。

[10:05] Daniel

Perfect. So, it all starts with um one recurring task uh which is a composition of several skills called the weekly prep. So the weekly prep is the very basic uh uh task that helps me start my week. I'll say that I won't run this live uh just because uh I want to be mindful of time. But generally speaking uh this task helps me go through all these things that you just saw in the notion and fill them in. So we basically have the top of mind, the this week and the inbox. And what this uh series of uh skills does, it pulls in information from my entire ecosystem um from my notion, my calendar, Slack, um from from uh granola transcripts of meetings and basically recommends what I should be focusing on this week. If there's new top of- mind items that need to be uh added or removed, it would recommend that. It would recommend moving, removing, finishing things from this week. And the same with the inbox. Um and this basically uh on Sunday mornings, this is our beginning of the week. Uh it helps me start with a very very clear and focused list of what I actually want to do. And then finally, it helps me prep for my meetings. So it runs me through the meetings that I'm going to have and asks me how I want to prep for them. Whether it's like a serious preparation that requires its own task in my weekly priority, whether it's a quick reminder or just meetings that don't really need any prep. And then I confirm all of this and I'm set to go. And this is the start. But

好。一切都始于一个定期任务，它由好几个 skill 组合而成，叫 weekly prep。weekly prep 是帮我开启一周的最基础的任务。先说一下，这个我就不现场跑了，主要是想节省点时间。但大体上，这个任务会帮我过一遍你刚才在 Notion 里看到的那些东西，并把它们填好。也就是 Top of Mind、This Week 和 Inbox 这几块。这一串 skill 做的事，是从我的整个工具生态里拉取信息——Notion、日历、Slack，还有 Granola 里的会议转写——然后基本上会建议我这周该聚焦在什么上。如果有新的 Top of Mind 事项需要添加或移除，它会给出建议。This Week 里的事项该挪走、删掉还是标记完成，它也会给建议。Inbox 也一样。基本上每到周日早上——那是我们这边一周的开始——它就帮我带着一份非常非常清晰、聚焦的清单开工，上面列的是我真正想做的事。最后，它还会帮我准备会议：带我过一遍接下来要开的会，问我每个会想怎么准备——是需要认真准备、得在我的每周优先事项里单独列一个任务，还是只需要一个快速提醒，又或者是根本不需要准备的会。然后我把这些全部确认一遍，就可以开干了。这只是开始。不过

[11:37] Claire

and does that push all of it into notion?

那它会把这些全都推送到 Notion 里吗？

[11:40] Daniel

Yes. And that basically this test that I ran earlier basically created this state that we have uh now the the notion uh that we have right now. But I'll also show you how it updates it.

对。其实咱们现在看到的这个 Notion 的状态，就是我之前跑的那次测试生成的。不过我也会给你演示它是怎么更新的。

[11:53] Claire

Yeah, let's look at that.

好，我们来看看。

[11:54]

Perfect. So the unique thing about this is that I think and this is something I ran into uh after having created the first weekly prep. I think one of the things any any PM uh can resonate with is the fact that there's two worlds, right? There's this beautiful clean organized world that's in our weekly priority or our road map where everything is very clear and beautiful and fits in and then there's reality where as we said this chaos of Slack and meetings and urgent priorities and executive asks and endless things that just flow in and out and what I was feeling in the beginning is this real um this real gap between what I was doing and prepping and what it was happening in real life where I was spending all this time doing all these small tasks. And I think I've built the system to really work alongside me. And part of the fact that it knows me really well is also that it knows how I work. So for example, I'm inbox zero both on Slack on on email. So it knows that if something is was saved on Slack and is no longer, it means it's likely done. Or if something was on inbox in in Gmail and is no longer so, then it probably means I already read it. And that really really helps in how we collaborate smoothly. So basically, let me run to uh my next scheduled uh my next scheduled task. And this is called the morning brief. And this is I think where the real magic happens. So actually we're going to run this live right now. I had to laugh because I think folks AI setup is a real reflection of their internal personality. And you strike me as somebody who is organized and thoughtful. You're an inbox zero person. I am sorry to anybody that tries to text or email me. I am like a chaos reigns person. I cannot put my stuff in a table. I cannot put my stuff in a conbon board. I need just a bot and I need when it's 2 in the morning or it's 6:00 in the morning or it's 2 in the afternoon. I just need to be like,

Daniel： 好。这里比较特别的一点是——这也是我做完第一版 weekly prep 之后才碰到的问题——我觉得任何一个 PM 都会有共鸣：其实存在两个世界，对吧？一个是每周优先事项（weekly priority）或路线图里那个美好、干净、井井有条的世界，一切都清清楚楚、漂漂亮亮、严丝合缝；另一个则是现实，就像我们刚才说的，Slack 和会议搅成的一团混乱，还有各种紧急事项、高管临时提的要求，以及没完没了涌进来又流出去的事情。我一开始的感受就是，我在做的、在准备的，和现实里实际发生的之间，真的有一道鸿沟——现实中我把大把时间都花在了处理这些零碎小事上。而我觉得，我搭的这套系统是真正能跟我并肩干活的。它之所以非常了解我，一部分原因就在于它了解我的工作方式。比如说，我在 Slack 和邮件上都保持 inbox zero（收件箱清零）。所以它知道，如果某条消息原本在 Slack 里存为了稍后处理，现在不在了，那多半说明这件事已经办完了；或者某封邮件原本在 Gmail 收件箱里，现在不在了，那大概说明我已经读过了。这一点对我们顺畅协作真的帮助非常大。好，那我接着跑下一个定时任务，它叫 morning brief（晨间简报）。我觉得真正的魔法就发生在这里。所以我们现在就现场跑一遍。

Claire： 我忍不住笑了，因为我觉得一个人的 AI 配置，真的就是他内在性格的写照。你给我的印象就是那种有条理、考虑周全的人，你是 inbox zero 派的。而我呢，只能跟所有想给我发短信、发邮件的人说声抱歉。我是那种任由混乱当道的人。我没法把我的东西整理进表格，也没法把它们放进 Kanban 看板。我只需要一个 bot，不管是凌晨两点、早上六点还是下午两点，我就只想冲它说一句，

[13:57]

"Dearbot, what's going on? [laughter] What's going on?" And so it's always fascinating to me to watch people who I just pre-AI have like a much more structured approach to their work and then post AI I think it really reflects in how you build build your system because mine looks it probably should look like something like this but it looks nothing like this. [laughter] Well, I think I I'll say first of all that uh you are correct in in me being quite an organized person, but I will say like again in reality a lot is a mess and a lot is like I I've found out how much is like business logic in my brain and I've had this experience with the system where I continuously like do discovery on myself and how I work and I continuously run into this places where the system is not good enough because it doesn't cover this and what it doesn't cover like it has a blind spot and it becomes much less powerful. So over time I've like covered all different angles.

Claire： “亲爱的 bot，现在啥情况？（笑）啥情况？”所以我一直觉得特别有意思：看那些在 AI 之前工作方式就更有章法的人，到了 AI 时代，这一点会真切地体现在他们搭系统的方式上。因为我的系统呢，按理说大概应该长成这样，但实际上完全不是这个样子。（笑）

Daniel： 嗯，首先我得说，你说得没错，我确实是个挺有条理的人。但我也要说，现实中很多事情还是一团乱，而且我发现有大量业务逻辑其实只存在于我的脑子里。在用这套系统的过程中，我有这样一种体验：我不断地对自己、对自己的工作方式做 discovery，也不断撞上系统不够好用的地方——因为它没覆盖到某些东西，而一旦有没覆盖到的地方，它就有了盲区，威力也就大打折扣。所以随着时间推移，我把各个角度都补齐了。

[14:57] Daniel

Um and I think it's doable. I think it's doable. And up until recently I would say in a way I was also operating chaotically. You know some of the tasks were saved on Slack. Some of them were like in my inbox. Others were just like tabs that are open in Chrome. And I've just had Claude like learn to work with all of these um and say, "All right, I'll take unfortunately it can't take an open tab as as a task, but uh everything else I just learned to look at uh the same way I do."

我觉得这是做得到的，真的做得到。而且我得说，直到不久前，我在某种程度上也是乱着来的。你知道，有些任务存在 Slack 里，有些在我的收件箱里，还有些干脆就是 Chrome 里开着的标签页。我就让 Claude 学会处理所有这些，让它能说：“好，这些我来接。”可惜它还没法把一个开着的标签页当成任务，但除此之外的一切，它都学会了用和我一样的方式去看待。

[15:24] Claire

Well, let's look at this morning brief because we've seen a couple morning briefs, but I want to see what you think makes yours different.

那我们来看看这个 morning brief 吧，因为我们已经看过好几个晨间简报了，但我想知道你觉得你的有什么不一样。

[15:29] Daniel

Okay. So, I think that there's several things, but I I'll walk through this uh very quickly and and show uh the different parts and I'll highlight the interesting things. So the first part is that uh this brief basically walks me through my meetings. I think one of the biggest pains of PMs is we have ton of meetings. Uh we need action items, we need to act on them. Almost every meeting creates new work, but we have to run to the next meeting and often times it's backtoback until the end of the day and then you know so this helps me kind of catch up on on yesterday's meetings. Um and it shows all the different things and basically you can see this takes the granola transcripts uh and it gives me a oneliner about each meeting and if there's an action item. So here I can see uh from all of these meetings looking through them uh there is one action item uh to share the walkthrough and recording of a of and a write up in the feature channel uh with go to market. Um and that's it. So it's going to offer me to uh expand or draft any summaries. I'm going to say uh no, let's continue for now. And we're going to walk walk on over to the next phase which I think is pretty cool. One of the things that I think is super powerful with co-work and and with these harnesses is that you can set them up for improving themselves. And one of the things I've done is I've over time uh I'm building up towards uh Claude being more agentic and being more able to act on itself. Um and I'll show you this and I think it's pretty visible in this flow. This part is pretty cool where basically one of the difficulties in life is keeping the context clear,

好。我觉得有好几点不同，不过我先很快地过一遍，展示一下各个部分，再把有意思的地方点出来。第一部分是，这份简报基本上会带我把会议过一遍。我觉得 PM 最大的痛点之一就是会特别多。我们需要行动项，也得去落实。几乎每个会都会产生新的工作，可我们又得赶去下一个会，很多时候一个接一个连轴转到下班，然后你知道的……所以它能帮我把昨天那些会的内容补上。它会把各种事项都列出来。你可以看到，它基本上是读取 Granola 里的会议转写，给每个会一句话概括，并标出有没有行动项。所以在这里我能看到，把这些会都过了一遍之后，有一个行动项：在功能频道里跟 go-to-market 团队分享演示讲解和录屏，外加一份书面说明。嗯，就这些。接下来它会问我要不要展开细说，或者起草某份总结。我就说：不用，先继续吧。然后我们进入下一个阶段，我觉得这部分挺酷的。我觉得 Cowork 以及这类 harness 特别强大的一点在于，你可以把它们设置成能自我改进。我做的其中一件事就是，随着时间推移，一步步让 Claude 变得更 agentic、更能自主行动。我演示给你看，我觉得在这个流程里体现得很明显。这部分挺酷的：基本上，现实中的一大难点就是保持上下文清晰，

[17:00]

right? So you have the system, you contextualize it, it knows things, but then how do you make sure that it keeps track. So this is one of the ways I've thought of. So basically what it does is imagine in this daily brief my coworks through my recent slacks, my emails, my granola notes and it looks for context that it doesn't understand in these things and it says like oh if there's a file I wasn't aware of, there's a milestone, there's a goal, it would ask me about it and say hey like what is this? Is this important? Should I read this? Uh so here it flagged something. Let's see what it is. A term I didn't know the settlement cap. All right. This is some payments thing and it says like I read the thread and understand it's a limit working through. Want me to save? Yeah. So I'll say save. Yes. Save it to the context. So basically this is how Claude helps me keep it contextualized and not lose track of things. This is really genius and I want people to just take a pause and reflect on what this is because it looks like for your context you are making sure that Claude understands internal definitions, terms, goals, etc. And so many of us work inside companies where the things that we say are not in the training data.

Daniel： 对吧？你有了系统，给它补足了上下文，它也知道了很多事，但接下来你怎么确保它能持续跟上最新情况？这就是我想到的办法之一。它做的事情基本上是这样：想象一下，在这份每日简报里，我的 Cowork 会把我最近的 Slack 消息、邮件、Granola 笔记过一遍，从中找出它不理解的上下文。比如说，如果出现一份它之前不知道的文件、一个里程碑或一个目标，它就会来问我：嘿，这是什么？这个重要吗？我该读一下吗？你看，这里它就标出了一个东西，我们看看是什么。一个我不认识的术语：settlement cap（结算上限）。好，这是个支付方面的东西。它说：我读了那个讨论串，理解这是一个正在处理中的额度上限，要我保存吗？嗯，那我就说保存。对，把它存进上下文。所以基本上，Claude 就是这样帮我让上下文保持完整、不漏掉任何事的。

Claire： 这真的太妙了，我希望大家停下来好好想想这到底是什么：看起来你是在为自己的上下文把关，确保 Claude 能理解公司内部的定义、术语、目标等等。我们很多人都在公司里工作，而我们平时说的那些东西，并不在训练数据里。

[18:10] Claire

They well not yet at least they're not in the training data. And so actually proactively prompting Claude to say like I don't understand what you're talking about here. Can we define it together and then I'll save it to my context and I'll know what you're talking about moving forward is really really sharp and something we haven't seen on the podcast yet. I love this.

它们……嗯，至少现在还不在训练数据里。所以主动 prompt Claude，让它说出“你这里说的我不太明白，我们能一起把它定义清楚吗？然后我会把它存进我的上下文，以后就知道你在说什么了”——这一招真的非常高明，也是我们这档播客里还没见过的。我太喜欢了。

[18:28] Daniel

Yeah. So, it's it's super hard to just keep track and and the unlock is that you're able to say, "Oh, I can just ask Claude to be my partner in doing this and automatically, you know, follow up and and insist on on getting this knowledge is a really really big unlock."

是的。光靠自己去跟踪所有事情真的超级难，而关键的突破在于，你可以说：“哦，我可以直接让 Claude 当我的搭档来做这件事，自动去跟进，并且坚持把这些知识弄到手。”这真的是一个非常非常大的突破。

[18:45] Claire

Amazing. So you run it weekly, you run it daily. Do you actually spend time in the notion or are you still like navigating all this with Claude?

太棒了。所以你有每周跑的，也有每天跑的。那你实际上会花时间待在 Notion 里吗？还是说你仍然主要通过 Claude 来处理这一切？

[18:58] Daniel

So the main idea is that notion is almost a read only for me. I I look at the notion to get my focus uh to remind me of my top of mind to remind me of uh my priorities and focus this week. Um, you know, every so often I'll I'll mark something as done if I just finished it. But generally my goal is to have uh notion as as something I only look at and Claude manages on my behalf. And in general, I I'll say this is something that has happened to me naturally, but I think it's one of the core pillars of building a strong system. Today, I say I manage about 70 to 80% of my time in front of the computer with co-work alone. Almost everything I do here is through co-work um and I try to do more and more of it because anything that I don't do in co-work then the context is not as well caught as if I do it through co-work whether it's you know other applications whether it's Jira Slack uh queries pro anything generally so

核心思路是，Notion 对我来说几乎是只读的。我看 Notion 是为了找准重心——提醒自己 top of mind 里最挂心的事，提醒自己这周的优先事项和重点。当然，偶尔我刚做完一件事，也会顺手把它标成已完成。但总的来说，我的目标是让 Notion 成为一个我只负责看、由 Claude 替我打理的东西。而且总体上我得说，这是在我身上自然而然发生的，但我认为这是搭建一套强大系统的核心支柱之一。如今，我在电脑前的时间里，大约有 70% 到 80% 是单靠 Cowork 来处理的。我在电脑上做的几乎所有事情都是通过 Cowork 完成的，而且我在努力让这个比例越来越高——因为凡是我不在 Cowork 里做的事，上下文就捕捉得不如通过 Cowork 做的时候那么完整。不管是其他应用，还是 Jira、Slack、各种查询、pro……基本上任何事都是这样，所以

[19:57] Claire

are you using co-work with Chrome at all like are you using browser use or are you purely just using like connectors and MCPS for all of this.

那你会把 Cowork 和 Chrome 搭配着用吗？比如用 browser use（浏览器操控）？还是说这一切纯粹只靠 connector 和 MCP？

[20:09] Daniel

Yeah. So I I would say in the beginning I was using like Chrome a lot. In my personal co-work that I have I use Chrome a lot. um as as time proceeds and and there's more and more MCPS and connectors then obviously those are comfortable but like the Chrome connector is a critical critical piece of uh you know of the puzzle for when you don't have these connectors and integr integrations and you need to use the browser

嗯，我会说一开始我用 Chrome 用得很多。在我自己私人用的那个 Cowork 里，我也大量用 Chrome。随着时间推移，MCP 和 connector 越来越多，用它们显然更顺手，但 Chrome connector 依然是整块拼图里非常非常关键的一块，专门应对那些没有现成 connector 和集成、必须动用浏览器的情况

[20:35] Claire

and do you feel like this is really saving you like what's the benefit here? Is it like saving you time? Do you feel like you just know everything? Are you getting better work done? Like why? because, you know, again, these systems take time to maintain and I think we're all sort of at this point with our AI systems where we're like, okay, they're all super cool, but like why are we why are we doing this? And so, I'm just curious like what is the benefit you're getting as a PM to to running this system?

那你觉得这套系统真的帮你省下了什么吗？它的好处到底在哪？是帮你省时间？是让你感觉什么都了然于胸？还是让你把工作做得更好了？到底图什么？因为你知道，这些系统是需要花时间维护的，而且我觉得我们现在对自己的 AI 系统都差不多到了这么个阶段：好吧，它们都超酷，但我们到底为什么要搞这些？所以我就很好奇，作为一个 PM，你跑这套系统到底得到了什么好处？

[21:01] Daniel

Yeah. So, it's it's a great question and I think honestly oftenimes we make systems that don't give us great benefit. I just uh I built a personal uh agent that its main work is sending memes about my wife in our shared panel on Telegram, which hey, to me is huge value. I think it's helping me uh win arguments. But uh jokes aside,

好，这是个很好的问题。说实话，我觉得我们常常会搭一些其实没带来多大好处的系统。我刚刚就做了一个私人 agent，它的主要工作是在我们俩共用的 Telegram 频道里发关于我老婆的梗图——嘿，这对我来说价值巨大，我觉得它在帮我吵架吵赢。不过玩笑归玩笑，

[21:23] Daniel

I I think

我觉得……

[21:24] Claire

because you need to watch our episode with Jason, the CEO of Meme Lord, who actually has a meme API, so you can plug that into your agent. That was a a very weird episode that we did.

因为你真得去看看我们跟 Jason 录的那期——他是 Meme Lord 的 CEO，他们还真有个 meme API，你可以直接把它接进你的 agent。那是我们做过的特别怪的一期节目。

[21:37] Daniel

So, I I haven't watched it yet, but I will say it's literally like the top to-do task in my in like the Claude session that I manage my agents. And what what's next? The meme API is is definitely the next one.

我还没看，但说真的，这事现在就挂在我待办清单的第一条——就在我用来管理各个 agent 的那个 Claude 会话里。下一步做什么？meme API 绝对就是下一个。

[21:53] Daniel

Yeah. Um, but I'll say I think it's a really really good question and I will say also that probably in the beginning um it's not going to like the ROI is not going to look great especially when you set up something that's as robust and as significant it's going to take some time uh because while it doesn't know you and your personal we can call it business logic and like the way you operate it's going to be very frictionful and everything it does or says or thinks is going to be a little bit and you're going to lose trust and you're going to be very skeptical and then you're going to double check everything. But the value tremendously compounds. It's just incredible the more you kind of power through these first few weeks of a contextualizing ruthlessly and continuously sharing everything you can and b centralizing whatever work you can in co-work or in your harness. Even if it doesn't feel so natural over time, it is just in I cannot describe how powerful it is. And I'll say personally, I think first of all, I'm really able to do in a day now what used to take me a week and and I stand behind that. It's and you know, it's something that sounds like a like a hypy LinkedIn post headliner,

对。不过我想说，这真的是个非常好的问题。我还想说，一开始 ROI 大概不会太好看，尤其是当你搭的东西这么完整、这么重量级的时候，这是需要时间的。因为在它还不了解你、不了解你个人的——姑且叫它“业务逻辑”吧——以及你做事方式的时候，用起来会处处是摩擦，它做的、说的、想的每件事都会有点……然后你就会失去信任，会变得非常怀疑，什么都要再核对一遍。但它的价值是会大幅复利增长的。你越是咬牙熬过最初这几周——一是毫不手软地补充上下文，持续把你能分享的一切都分享给它；二是把你能集中的工作都集中到 Cowork 或你自己的 harness 里——效果就越惊人。哪怕感觉没那么自然，时间一长，它真的是……它有多强大，我真的没法形容。就我个人来说，首先，以前要花我一周的活，现在我一天就能干完，这话我敢负责。我知道，这听起来很像 LinkedIn 上那种炒作帖的标题，

[23:09] Daniel

but it's it's it's true, you know. It's it's insane. And I I look at um I look at my the way I work through these like this week and often times I can get through a this week in a good day. Uh which is just mind-blowing. But more than that, I'll say it's a lot of deeper work. Um there's things that I do today that I just wasn't able to do before. Whether it's more in-depth research, whether it's better uh being more databacked when checking things, working less off of hunches and more off of like facts and researches, whether it's competitive researchers, like anything you do today, you can do it almost at the top level. Um as opposed to back in the day, moving moving fast meant you need to, you know, operate on a lot of hunches. um not work off of data, um not be as robust in terms of your documentation. And today it really really changes.

但这是真的，简直疯狂。我回头看自己这阵子的工作方式，比如这周，很多时候状态好的一天就能把一周的活干完，这真的太震撼了。但不止如此，我想说的是，我做的深度工作也多了很多。有些事我现在在做，以前根本做不到。不管是更深入的调研；还是在核查东西时更有数据支撑，少凭直觉、多靠事实和研究；又或者是竞品调研——你今天做的任何事，几乎都能做到顶尖水准。而以前，想跑得快就意味着你得大量凭直觉行事，不基于数据，文档也做不到那么扎实。今天这一切真的彻底变了。

[24:07] Claire

I I love it. I love it. You you've sold me. Um I don't know if I'm going to get as organized as you, but I and I think that's absolutely right. I one of the things even at a company level that I tell people is look, you got to go through the pain of switching your systems and it's not going to be as efficient. And yes, of course, like the old way will be faster because it's muscle memory, but if you get to the next level, not only will you have a system that works better for you, but everybody will have more skills. And I think that's almost the more important part where this is a personal productivity system that works for you. That's great. Um, but I'd say probably the the bigger benefit is it forced you to learn how to use co-work and use these AI tools in a different way. And as a product builder, you have to know that stuff. You have to think about, well, what if my product's being used only through Claude? What what do I want that to look like? And I think it's just if you are in a product, it is just one of the most important things that that you could you could do. This episode is brought to you by Jira by Atlassian. The teamwork graph in Jira delivers 44% more accurate agent results with 48% less token usage. That's a huge difference when working with AI coding agents like Claude, Cursor, Codeex, or C-Pilot. The hardest part of shipping with AI isn't the [music] code, it's the context.

太喜欢了，真的太喜欢了。你把我说服了。我不知道自己能不能像你这么有条理，但我觉得你说得完全对。哪怕在公司层面，我也常跟大家说：听着，切换系统的阵痛你得熬过去，而且一开始效率不会那么高。没错，老办法当然更快，因为那是肌肉记忆；但如果你能迈上下一个台阶，你不仅会拥有一套更适合自己的系统，每个人也都会掌握更多技能。我觉得这几乎是更重要的部分。这是一套为你服务的个人效率系统，这很棒；但我想说，更大的收获可能是它逼着你去学怎么用 Cowork，用一种不一样的方式去用这些 AI 工具。作为做产品的人，这些东西你必须懂。你得去想：要是我的产品只通过 Claude 来使用呢？我希望那会是什么样子？我觉得只要你是做产品的，这就是你能做的最重要的事情之一。本期节目由 Atlassian 旗下的 Jira 赞助播出。Jira 的 Teamwork Graph 能让 agent 产出结果的准确率提高 44%，同时 token 用量减少 48%。当你和 Claude、Cursor、Codex 或 Copilot 这类 AI 编程 agent 一起干活时，这差别可太大了。用 AI 交付产品，最难的不是代码，而是上下文。

[25:30] Claire

What's the right ticket? What did the spec say? What got decided in Slack? The teamwork graph pulls all of that from across your entire stack from Jira and Confluence to GitHub and feeds it directly to your agents before they write a single line. [music] You assign the work, the agent gets everything it needs, and a PR surfaces when it's ready. No digging, no contact switching, no broken flow. Same team, smarter agents. Try and free at jira.dev. That's jir.dev. I'm curious how this scales beyond you. So you have your system, you're, you know, 10x PM now. Congratulations. How does this affect the team? How do they use systems like this? How do you share kind of how do all how does that work?

该对应哪张工单？spec 里是怎么写的？Slack 里最后定了什么？Teamwork Graph 会从你的整个工具栈——从 Jira、Confluence 到 GitHub——把这些信息全都拉过来，在 agent 写下第一行代码之前就直接喂给它们。你分派任务，agent 拿到所需的一切，准备就绪时就会冒出一个 PR。不用四处翻找，不用切换上下文，心流不被打断。团队还是那个团队，agent 更聪明了。去 jira.dev 免费试用，网址是 jira.dev。我很好奇它怎么推广到你之外的人身上。你有了自己的系统，现在也算是 10 倍 PM 了，恭喜。那这对团队有什么影响？他们是怎么用这类系统的？你们怎么分享——整个是怎么运作的？

[26:19]

A lot of what you brought up is is super relevant in that sense. Um and I think one of the things that have made me you know in a sense uh wait for uh for co-work for example when a few months ago there was this endless hype about cursor and about cloud code and you know I was kind of sitting on the sidelines um and and waiting like should I adopt these like tools that are not meant for me and are super super technical can be scary um and I'm glad I waited for for cowork because it's a lot about like you know just UX and and experience and simplicity and a lot of the tools that we've built for for the team and whether it's the gem that helps the team write uh specs product specs that we did almost a year ago or uh what we call the workstation that we have that I'll show a little bit later at the end of the day if you make something accessible and simple and the value is clear um then it then it gets adopted much better. All right, before we get to the workstation flow, which is going to be how you get this across the company, across the team, you also have some self-improving loops that I think are really interesting. So, how do you improve this over time? How does it get better? What have you figured out there?

Daniel： 你刚才提到的很多点，在这方面都特别相关。我觉得，某种程度上让我一直在等 Cowork 的原因之一是——比如几个月前，大家对 Cursor、对 Claude Code 的炒作没完没了，而我基本是在场边观望、在等：我该不该去用这些本来就不是为我设计的工具？它们技术性超强，可能挺吓人的。我很庆幸自己等到了 Cowork，因为它很大程度上关乎 UX、体验和简洁。我们给团队做的很多工具也是一样，不管是差不多一年前做的、帮团队写产品 spec 的那个 Gem，还是我们现在叫 Workstation 的那个东西（我待会儿会演示）——说到底，只要你把东西做得容易上手、足够简单，价值又一目了然，大家就会更愿意用它。

Claire： 好，在讲 Workstation 这个流程——也就是你怎么把这套东西推到全公司、全团队——之前，你还有一些自我改进的 loop，我觉得特别有意思。那你是怎么随着时间去改进它的？它是怎么变得越来越好的？你在这方面摸索出了什么？

[27:31] Daniel

Yeah, so I think this is one of the coolest unlocks that have helped me continuously improve the setup while not dedicating too much time to it. And basically, it's almost not self-improvement, but it's Claude helping me improve it. So, I'll walk you through something called the self-improvement uh loop. It's a scheduled test task that runs weekly and it has several skills attached to it and and I'll walk through them. So, again, it it works in parts. Uh a trick I found really helpful when when I have a bunch of information that Claude feeds me. So, I'm going to walk through this uh in retroactive. Uh so, first of all, the first part is how my drafts landed. So, what I found over time is that when I write with Claude, it'll give me a draft and then I'll change it and I'll just send it. But the thing there is that Claude doesn't learn what the gap is between uh the draft that it wrote and what I actually sent. So, I have Claude uh when running this task looking for drafts that it it shared with me and I didn't respond and then looking to see if I sent them in a different way and learning from those draft and saying, "Oh, so we actually ended up writing this and doing the learning on it on its own." So in this specific uh in this specific week I see that there was nothing learned and insignificant but this is very helpful in continuing improving and you know uh polishing claude on on how I write. Now this is a cool one. The second one and this is skills that are worth building. So basically Claude on an ongoing basis looks for things that I do uh on a recurring uh basis and suggests turning them into skills. Many of my skills actually ended up being uh from this flow. Uh, so you could see it. Uh, I've been prototyping a bunch lately. So,

对，我觉得这是最酷的突破之一：它帮我持续改进这套配置，又不用我投入太多时间。基本上，与其说是自我改进，不如说是 Claude 在帮我改进它。我带你看一个叫“自我改进 loop”的东西。它是一个每周运行一次的定时任务，挂了好几个 skill，我逐个讲一下。还是那样，它是分几个部分来的——这是我在 Claude 一下子喂给我一大堆信息时，发现特别好用的一个技巧。我就拿已经跑出来的结果，回头给你过一遍。首先，第一部分是“我的草稿最后是怎么落地的”。我慢慢发现，我用 Claude 写东西时，它会给我一版草稿，我改一改就直接发出去了。但问题在于，Claude 学不到它写的草稿和我实际发出去的内容之间差在哪儿。所以我让 Claude 在跑这个任务时，去找那些它发给我、但我没有回应的草稿，再看看我是不是换了种写法发出去了，然后从这些草稿里学习，自己总结出“哦，原来我们最后是这么写的”，自己完成这部分学习。像这一周，我看到没学到什么，也没什么值得一提的，但这对持续改进、把 Claude 打磨得更贴合我的写法非常有帮助。接下来这个很酷。第二部分是“值得做的 skill”。基本上，Claude 会持续留意我反复在做的事情，然后建议把它们做成 skill。我的很多 skill 其实都是从这个流程里来的。你可以看到，我最近在做一大堆原型，所以，

[29:09] Daniel

it's, uh, helping me. Uh, it's suggesting I create a claw design handoff, uh, skill between co-working where I do the research and design where I actually build the prototype. Um, and a bunch of other interesting skills uh, that I've built. I have a skill like this, by the way, that's handoff between co-work uh, which I use kind of as the exe executing arm of co-work. This is the second part. The third part is fixes to my existing setup. So basically all of my skills and recurring tasks have uh lines in their in their files to collect feedback and friction for my interaction with them. So over time they gather this I don't need to do this proactively. Whenever I ask for a fix or something they gather and track this information and every week it surfaces the top frictions from my interaction with Claude and suggests to improve them. Um so these are also super helpful as you can see. And then finally, this is a a really cool one. This is actually the uh the improve scale. So basically, uh we're all dealing with an endless flood. I think this resonates with anybody.

它在帮我——它建议我做一个 Claude Design 交接 skill，衔接 Cowork 和 Design：调研我在 Cowork 里做，原型则是在 Design 里真正搭出来。还有其他一堆有意思的 skill，也都被我做出来了。顺便说一句，我已经有一个类似的交接 skill，是从 Cowork 往外交接的，我基本上把它当成 Cowork 的执行臂来用。这是第二部分。第三部分是“对现有配置的修补”。基本上，我所有的 skill 和定期任务，文件里都写了几行，用来收集我跟它们交互时的反馈和摩擦点。这样随着时间推移它们就会自己攒下来，不需要我主动去做。每当我要求修点什么，它们都会收集并记录这些信息，每周再把我跟 Claude 交互中最突出的摩擦点列出来，并建议怎么改进。你也看到了，这些同样非常有用。最后一个，这个真的很酷，其实就是那个 improve skill。基本上，我们都在应对一场无穷无尽的洪流，我想这一点谁都有共鸣。

[30:15] Daniel

Endless flood of tips, uh hyping up from X and LinkedIn and and and newsletters and blogs about setups and how to improve your AI usage and you must have this and you must build that. And I found this incredibly overwhelming. on one hand I don't want to drown in it at the same time I want to be able to actually implement what's relevant so I've built a skill called improve which is basically like a very critical um auditor for these uh and I share them uh usually through my phone through a slack channel with claude and I ask them like is this real is this powerful how how does this adapt to what we have now is this needed and then it helps me review it and together we basically decide whether we want to build it or not uh so here I ran it against uh Claire, your uh how I AI episode on designing loops. Um and it basically gave me like a a criticism of the episode. It said it's like a good vid a video and it says like listen loops at the end of the day are essentially like scheduled test. So you're running loops in many ways but I don't have goal loops that are uh the way you suggested. So this is something I can adapt. um uh there's some things I can adopt and basically this is helps me you know keep track of all the things that I run into and all these like tips and tricks and at the same time not drown in them and also not um you know give into the hype too much. So these are all the different things I do to you know claude helps me continuously improve it over time with as little you know dedicated effort as possible.

无穷无尽的各种技巧和炒作，来自 X、LinkedIn、各种 newsletter 和博客，讲怎么配置、怎么提升你的 AI 用法，你必须得有这个、你必须得搭那个。我觉得这些让人应接不暇。一方面我不想被淹没，另一方面我又希望能真正把相关的东西落地。所以我做了一个叫 improve 的 skill，基本上就是一个对这些内容非常挑剔的审核员。我通常是用手机，通过一个和 Claude 连通的 Slack 频道把这些东西发过去，然后问它：这是真的吗？这东西厉害吗？放到我们现有的这套里要怎么适配？有必要吗？然后它帮我评估，我们一起决定要不要做。比如这里，Claire，我拿它跑了一下你们 How I AI 讲设计 loop 的那一期。它基本上给了我一份对这期节目的批判性点评。它说这是个不错的视频，还说：你听着，loop 说到底本质上就是定时任务，所以你其实在很多方面已经在跑 loop 了，但我还没有你讲的那种 goal loop。所以这是我可以借鉴改造的，有些东西我可以直接采用。基本上，它帮我把碰到的各种东西、各种技巧窍门都记下来，同时又不被它们淹没，也不会太被炒作带着走。所以这些就是我做的各种事情，让 Claude 帮我随着时间持续改进这套系统，同时尽可能少地投入专门的精力。

[31:43] Claire

Love the we could have done the entire episode on the last three minutes. so good. You know, I the first one where it's like take everything that you wrote for me that I hated and did it did it separately or dropped or rewrote and then just learn from that. That's genius. We saw Alex Lieberman had like a write like me self-learning loop, but this one is really interesting because it takes the indirect signals of rewriting and built it. I also in the loops episode have a skill writing loop. So, I think that's a good one. look at our work over the last couple week or last week and and write loops. And then what was it? Let me look at the third one real quick.

太爱了，光是最后这三分钟就够我们做一整期了，太精彩了。你知道吗，第一个——就是把你替我写的、但我嫌弃的东西，不管我是另外自己写了、直接弃用了还是重写了，全都拿来学习——这简直是天才。我们之前看过 Alex Lieberman 有一个“像我一样写作”的自学习 loop，但你这个特别有意思，因为它是从“重写”这种间接信号里学习、搭起来的。我在讲 loop 的那期里也有一个写 skill 的 loop，所以我觉得这个很好：回顾我们过去一两周、或者上周的工作，然后写 loop。然后是什么来着？我快速看一眼第三个。

[32:22] Daniel

So the third one is just taking interactions with Claude over the week. It's similar to building new skills but just fixing existing.

第三个就是把这一周里和 Claude 的交互拿来用。跟做新 skill 类似，只不过是修补现有的。

[32:30] Claire

Yeah. And then um and then the third one which is so genius and I wish more people did this is build in feedback or telemetry into your skills and then review that over time and improvement. I've seen a couple folks like I think intercom does this a little bit with some of their skills that they share across the org, but I haven't seen anybody for themselves say anytime that you gather feedback from me, positive, negative, on a skill or you hear feedback from me that's indirect, put it somewhere that we can review on a weekly basis. Super super super smart. And then of course ingest all this great content, especially from How AI and improve your systems. I love it. What I also love that you've told us you do is you take this mega system that you've built across notion across claude this improving loops and you figured out a way to scale this to the org in is it called workstation workstation okay let's let's see that

对。然后第三个真的太天才了，我希望更多人这么做：在你的 skill 里内置反馈或遥测机制，然后随着时间去复盘、改进。我见过几个人这么做，好像 Intercom 在他们全公司共享的一些 skill 里就有点这么做，但我还没见过谁为自己这样设定：只要你从我这儿收集到关于某个 skill 的反馈，不管正面还是负面，或者听到我间接表达的反馈，就把它放到一个我们每周可以复盘的地方。真的超级超级聪明。当然还有，把这些优质内容——尤其是 How I AI 的——吸收进来，改进你的系统。太喜欢了。我还特别喜欢你告诉我们的另一件事：你把这套横跨 Notion、横跨 Claude、带着这些改进 loop 的超级系统，想出了办法推广到整个公司，叫……是叫 Workstation 吗？Workstation。好，我们来看看。

[33:33] Daniel

for sure so basically um I'll say that one of the one of the main things that were important for us and we were debating whether to do this with cloud code which was much more powerful or with uh co-work and we decided on doing it with co-work uh we have this what we call the workstation which is essentially this uh I would say operating system it started out for PMs andio um but it actually expanded to all Amelio workers and basically uh what it does is you know it it it creates an ecosystem u an operating system for you but I think the cool thing here is this on boarding and it's important to say I built this with some very very talented and uh smart uh colleagues of mine. Um we really insisted in the beginning on on a good UX and after having you know a few PMs try to do it with cloud uh code through the terminal and taking 3 days you know to install everything and run all the dependencies we understood that we need to do something that's simple.

当然。基本上，我想说，对我们来说很重要的一点是——我们当时在纠结，是用强大得多的 Claude Code 来做，还是用 Cowork 来做，最后我们决定用 Cowork。我们有一个叫 Workstation 的东西，本质上是一套——我会说是操作系统。它最早是给 Melio 的 PM 用的，后来其实扩展到了 Melio 的所有员工。基本上，它做的就是为你打造一个生态系统、一套操作系统。但我觉得这里最酷的是它的 onboarding。而且得说明一下，这是我和几位非常非常有才华、非常聪明的同事一起做的。我们一开始就特别坚持要有好的 UX。之前让几个 PM 试着通过终端用 Claude Code 来搞，光是装好所有东西、跑通所有依赖就花了 3 天，我们就明白了：必须做一个简单的东西。

[34:31] Daniel

So basically one of the key elements of this workstation is that it basically does everything for you. So it's a UX built into the flow into the chat. Um, so basically what it does is walks you through everything you need to do. So besides just, you know, clicking on a few confirms and maybe uh addressing a few questions, it basically gets you up to speed on everything. It gets all your tools connected, confirms your roles, maps your colleagues, your management, where you sit, um, looks at your calendar and what that looks like and your Slack, helps you create your own voice so that you basically from day zero, you're writing in a way that sounds like you and just doesn't sound like a stranger. Um, captures your goals and basically helps you um, get moving really, really fast. I I couldn't do the full flow here because I I have it set up already. So uh but you could see it's very very simple. It takes you hand in hand uh and guides you to this like very very robust uh system that helps you you know work in co-work fully and become native.

所以这个 Workstation 的一个关键点是，它基本上把所有事都替你做了。它的 UX 就内置在流程里、内置在对话里。基本上，它会一步步带你完成所有需要做的事。除了点几下确认、可能再回答几个问题之外，它基本上能让你在各方面都迅速上手。它会帮你把所有工具连上，确认你的角色，梳理你的同事、你的上级、你在组织里的位置，看看你的日历是什么样、看看你的 Slack，还会帮你打造你自己的写作口吻，这样基本从第一天起，你写出来的东西就像你本人，而不像个陌生人。它还会记下你的目标，基本上就是帮你非常非常快地跑起来。我这里没法演示完整流程，因为我已经配置好了。不过你能看出来，它非常非常简单，手把手带着你，引导你用上这样一套非常非常扎实的系统，帮你完全在 Cowork 里工作，真正成为原生用户。

[35:34] Claire

I love this. I have seen so many people try to train this into their team like sit and say okay step one step two step three step four etc. just making a skill to say, "Okay, do you have all your connectors set up?" Great. Have we pulled your voice? Great. Have you told me about you? Great. Like, tell me, you know, let me set all this up and then from day zero, someone on your team, look, they're going to customize it. Maybe some people won't though. And so, if everybody's starting from this, it takes what, 15 minutes? Something like that.

我太喜欢这个了。我见过太多人想把这套东西硬教给团队，坐下来一条条讲：好，第一步、第二步、第三步、第四步，等等。其实只要做一个 skill，让它来问：“好，你的 connector 都连好了吗？”好。“我们把你的文风提取出来了吗？”好。“你跟我介绍过你自己了吗？”好。就是“跟我讲讲你的情况，我来帮你把这些全配好”，然后从第一天起，你团队里的人——你看，他们肯定会自己去定制。不过也可能有些人不会。所以如果每个人都从这个起点出发，这大概要花多久，15 分钟？差不多这样？

[36:11] Daniel

Yeah.

对。

[36:13]

I love it. And then do you just do this as like a shared skill? This is a plugin so everybody has it in our library and anybody can access it and you know now it's already like it's become quite a big thing like my colleagues have taken it to to the next level and it's really really impressive.

Claire： 太棒了。那你是把它做成一个共享的 skill 吗？

Daniel： 这是一个 plugin，所以我们库里每个人都有，谁都能用。而且你知道，现在它已经发展成挺大一件事了，我的同事们把它又推上了一个新台阶，真的非常非常厉害。

[36:28]

Uh but yeah I I couldn't agree with you more. I think for me this this landed when about a year ago I built uh uh our spec writing gem call it call it uh spectacular and I I was using it for my own and I built it to be really you know back back in the day a year ago you know getting you know holding up the template was something really hard to achieve and making it not hallucinated in a long spec was like something that's hard and after I built something that was really powerful and good for for the way I was working and I was exciting to I was excited to distribute it. I had like uh live demos with several PMs that I watched them operated from zero and I saw how different like I thought it was already this complete uh fully like baked uh tool and I saw them using it and struggling because it was completely adjusted to how I work and that's when it started when into this gem I built this UX that guides you hand in hand that tells you like all right I'm your I'm your spec writing assistant here's how you can work with me here's what I recommend you to do here's how you can start and basically guiding them hand in hand making it simple and accessible. Um so yeah good UX and simplicity is critical even for you know internal teams and PMs and technical people. I I love this so much. This is I mean this is something if you haven't taken it away as somebody watching this podcast go every company should go do this and build like your sort of like bootstrap my personal AI skill or plugin inside quad or inside codeex. Everybody starts um from the same page and you can just get your most like AI pill on the team to build it for you and share and then everybody can have shared ownership over it over time. Daniel, this has been awesome. Just to recap for folks, you have notion is sort of like a readonly database source of truth for all your context and priorities for a week. You run two anchoring automations in co-work which is your weekly setup which sets up

Daniel： 呃，不过是的，我跟你的看法简直不能更一致了。对我来说，真正让我意识到这一点，是大概一年前我做了我们的 spec 写作 Gem，给它起名叫……叫 Spectacular。一开始是我自己在用，我把它做得非常——你知道，放在一年前那会儿，让它守住模板是件很难做到的事，要让它在长篇 spec 里不出现幻觉也很难。我做出了一个很强大、很贴合我自己工作方式的东西之后，特别兴奋地想把它推广出去。我跟好几位 PM 做了现场演示，看着他们从零开始上手操作，结果发现差别好大——我原以为它已经是个完整的、完全成熟的工具了，结果看他们用的时候很吃力，因为它完全是按我的工作方式调出来的。就是从那时候起，我在这个 Gem 里做了一套手把手引导你的 UX，它会告诉你：好，我是你的 spec 写作助手，这是和我协作的方式，这是我建议你做的，你可以从这里开始——基本上就是手把手地带着他们，让它简单、好上手。嗯，所以说，好的 UX 和简洁真的至关重要，哪怕对象是内部团队、是 PM 和技术人员。

Claire： 我真是太喜欢这个了。我是说，如果你作为这档播客的观众还没 get 到这一点——每家公司都应该去做这件事，搭一个类似“帮我初始化个人 AI”这样的 skill 或 plugin，放在 Claude 或者 Codex 里。每个人都从同一个起点出发，你只需要让团队里最“AI 上头”的那个人帮大家做出来、分享出去，之后大家就可以随着时间推移共同拥有、共同维护它。Daniel，这期太精彩了。给大家简单回顾一下：你把 Notion 当成一个只读数据库，是你一周所有上下文和优先级的 source of truth。你在 Cowork 里跑两个打底的核心自动化：一个是每周准备（weekly setup），它会把

[38:26] Claire

this whole notion thing preps you for meetings priorities and then your um daily briefing that um kind of cycles through the next day takes care of stuff and then you've been able to scale this in the org through kind of a a workstation plugin that anybody can use and kind of set up this thing for for themselves and now you're living your best life. And again, um you know, bosses don't listen to this, but the man says that he can do in one day what he used to do in a week. So, you're getting real just like real productivity gains from this. Um and you have learned maybe to be more organized and Claude uh plus notion have made you more more organized as a PM. So, everybody's everybody's benefiting here. Um, this has been awesome. I want to do like a couple of quick lightning round questions and then we will get you back to all of your highly productive work. You know, I want to go back to you saying that you operate almost 70 80% through Claude and through co-work. What do you think is missing? what what what is that remaining 20 30% and do you think it's going to be 100% at some point or kind of what's that gap and what's missing in in terms of you being able to use AI or use claude for everything?

整个 Notion 这一套都搭好，帮你准备好会议、理清优先级；然后是你的每日简报（daily briefing），它会把第二天的事过一遍，把该处理的处理掉。接着，你还通过一个 Workstation plugin 把这套东西推广到了整个组织，谁都能用，都能给自己搭起这一套。现在你的日子过得美滋滋。再说一遍，嗯，老板们别听这段啊，这位老兄说，他以前一周干的活，现在一天就能干完。所以你是真真切切地从中拿到了生产力提升。嗯，而且你可能也学会了更有条理，Claude 加上 Notion 让你作为一个 PM 变得更有条理了。所以大家都是赢家。嗯，这期太棒了。我想来几道快问快答，然后就放你回去继续高效干活。我想回到你刚才说的，你差不多七八成的工作都是通过 Claude、通过 Cowork 完成的。你觉得还缺什么？剩下的那两三成是什么？你觉得它有一天会到 100% 吗？或者说，这中间的差距在哪儿，要让你事事都能用 AI、用 Claude 来搞定，还缺些什么？

[39:52] Daniel

So it's a really really good question. I think the missing piece and I'm my my personal sense is that it's very close is a the ability to operate in a in the cloud essentially in a way that's uh unrelated to when my computer is open and running. I think that's critical and I think that was also created with the uh with a cloud tag which we don't yet have in our company but I've heard good things also that it's expensive. I don't know. But I think the ability to run uh Claude away from your computer or have it run on its own, something like OpenClaw or like a Hermes, uh is really really critical to do more of these things and to achieve more. I've found some ways to work around this. Like I have a ch a Slack channel that claude poll polls and when I'm away from the computer, I can send him some things uh and then he could read them later. But it's it's still something that has to happen online. Um, and this is I think the the real missing piece. Once I have something that's able to operate uh in the cloud when I'm away from my computer when it's off and you know uh coordinate like act on things uh that that will be huge. And also closely related is just the ability to actually like you know do do things. So I've been building up the system in order to enable to to basically prepare for it. Um, I've been having Claude not only capture my tasks and when they're done, but basically it looks at my tasks, it assigns like what it looks what it would imagine is a closed scenario. So, let's say something somebody asked a question and I saved it on Slack. So, it would say, okay, if there's a reply and it's unsaved, it means it's done. And then it looks at how I resolve these things. And that's building up towards the the time where I hope that for the simplest task that may take up time and be annoying, you know,

这个问题问得非常非常好。我觉得缺的那块——我个人感觉离它已经很近了——就是能在云端运行的能力，说白了就是跟我电脑开没开、在不在运行都没关系。我觉得这点至关重要，而且我觉得 Claude Tag 某种程度上也是冲着这个做的，不过我们公司还没开通；我听到过一些好评，也听说它挺贵的，我也说不准。但我觉得，能让 Claude 脱离你的电脑运行，或者让它自己独立跑起来，像 OpenClaw 或者 Hermes 那样，对于做更多这类事情、实现更多目标来说真的非常非常关键。我找到了一些变通的办法。比如我有一个 Slack 频道，Claude 会去轮询它，我不在电脑前的时候，可以给他发点东西，然后他之后再去读。但说到底，这事还是得在线的时候才能进行。嗯，我觉得这才是真正缺的那一块。一旦我有了一个能在云端运转的东西——在我离开电脑、电脑关着的时候也能跑，还能协调、能对事情采取行动——那将会是个巨大的突破。和这个密切相关的，就是真正去“动手做事”的能力。所以我一直在搭建这个系统，基本上就是为那一天提前做准备。嗯，我现在让 Claude 不仅记录我的任务、记录它们什么时候完成，基本上它还会查看我的任务，给每个任务设定一个它设想中的“结案场景”。比如说，有人问了个问题，我在 Slack 上把它收藏了。那它就会判断：好，如果这条有了回复、而且被取消收藏了，就说明这事办完了。然后它还会观察我是怎么解决这些事情的。这些都是在为那一天做铺垫——我希望到时候，对于那些最简单、但可能很占时间又很烦人的任务，你懂的，

[41:39] Daniel

when people somebody asks you for something or somebody asks about a file that you need or or know or to or to schedule a meeting, then it would just know to do that on its own and it wouldn't need me and and that's what I'm building up towards and that will be really powerful.

比如有人找你要个东西，或者有人问起一个你需要的、或者你知道的文件，又或者要约个会，它就会知道自己去把这些事办了，完全不需要我。这就是我正在一步步搭建的方向，到时候会非常强大。

[41:53] Claire

That's that's the dream. But we're not at the dream yet. And so I'm curious, you know, we our audience is a lot of PMs and I'm just curious now that you have this extra capacity and time, what are the the product tasks that you're doubling down on that you're doing even more, you're spending even more time on, where do you think the effort should go?

那就是理想状态了。但我们还没到那一步。所以我很好奇——我们的听众有很多是 PM——我就是好奇，现在你多出了这些产能和时间，你在哪些产品工作上加倍投入、做得更多、花的时间更多？你觉得精力应该往哪儿使？

[42:15] Daniel

I'm not going to say anything surprising, but just purely in-depth work, that's one. Um so talking to customers, talking to users and a lot of research. I think those two are uh really really important and given and also just the standard things that NEPM aims to get more time for. But I think that in today's world something that really really is important and interesting um and that is new is a um improving the system. So I think the beauty about this is that if a system is built correctly and on the right pillars, it can grow and improve over time. And I've improved my system basically week over week. Uh and this is done through, you know, dedicating time to research and learn about AI, catching up to speed on how things are progressing and implementing them. Um, I have a skill that's called improve uh which I send uh posts and tweets and articles to with like recommendations and tips about uh AI setup and together we review it and see if it fits for me or if it's just like LinkedIn hype and how it can collaborate with what my system is today and we continuously improve the system together with you know tips from uh from a bunch of resources. This is a skill that I think is critical today in order to be able to keep up and and become better. Um, yeah,

我说的不会有什么让人意外的，首先就是纯粹的深度工作，这是一个。嗯，也就是和客户聊、和用户聊，再加上大量的调研。我觉得这两块真的非常非常重要，另外就是任何一个 PM 都想多挤出时间去做的那些常规事情。但我觉得在当今这个时代，有一件真正非常重要、也很有意思、而且是新出现的事，那就是——改进系统。我觉得这件事妙就妙在：如果一个系统搭得对、立在正确的支柱上，它就能随着时间不断成长、不断变好。我的系统基本上是一周一周地在改进。呃，这靠的是专门抽时间去研究和学习 AI，跟上各种进展，然后把它们落地。嗯，我有一个叫 improve 的 skill，我会把各种帖子、推文和文章丢给它，里面都是关于 AI 配置的建议和技巧，然后我们一起审一遍，看它到底适不适合我，还是只是 LinkedIn 上的炒作，以及它怎么跟我现在的系统配合起来。就这样，我们借助各种来源的技巧，一起持续改进这个系统。我觉得在今天，这是一个至关重要的 skill，能让你跟上节奏、不断变得更好。嗯，对，

[43:34]

I I I love it. And then finally, my favorite question I ask everybody, you're organized, so maybe maybe you're nicer than me. When Claude is not doing what you want, when AI is not giving you the response that you want, what do you do? Do you yell? Are you nice? Are you polite? So, I think after having watched quite a few of your episodes, I think my response is pretty similar to everybody. Um, if I'm using whisper, I may yell. I don't get road rage, but I get clawed rage. So, I think like I can I don't I don't know if I've cursed, but I I've definitely like raised my voice. And I'm also frustrated at the fact that that doesn't go through whisper well enough. like I'm like, you know, exclaiming passionately and then I just see the transcript and it's like, why isn't this working? So, whisper folks, if you're out there, this is a feature request. Make make my exclamations be more felt.

Claire： 我、我、我太喜欢了。最后一个，也是我每期都会问、最爱问的问题：你这么有条理，说不定脾气比我好。当 Claude 不按你的意思来、AI 给不出你想要的回答时，你会怎么办？你会吼它吗？还是态度很好？很有礼貌？

Daniel： 嗯，我看了你们挺多期节目，我觉得我的回答跟大家都差不多。嗯，如果我在用 Whisper，我可能会吼。我不会路怒，但我会 Claude 怒。就是说，我不确定自己有没有骂过脏话，但我肯定提高过嗓门。而且还有一点让我很抓狂：这种情绪 Whisper 根本转不好。我这边正慷慨激昂地喊着，结果一看转写出来的文字，就是：“这怎么不管用？”所以，Whisper 的各位，如果你们在听，这是个功能需求：让我的呐喊能更被感受到。

[44:32] Claire

I love it. Yeah, I have this, Speaking of Claude Rage, I have this hypothesis. I have this aura on. I have this hypothesis that you can tell what model I'm working with with like how agitated and stressed I am. So maybe maybe that will be a future how I AI benchmark. I'm I'm baking annoy bench which is like how furious

太喜欢了。对，我有个——说到 Claude 怒，我有个假设。我手上正戴着 Oura 戒指呢。我的假设是：光看我有多焦躁、压力有多大，就能看出我在用哪个模型。所以说不定哪天这能变成 How I AI 的一个 benchmark。我、我正在酝酿一个 AnnoyBench，测的就是有多让人火大——

[44:54]

that would be amazing an agent or an AI makes me. This has been so much fun. Thank you for showing us your full system. I think it's totally accessible. Um I think a lot of people are going to put this in their own improved skill and I love how you're figuring out ways to share this with the rest of your team. Where can we find you and how can we be helpful to you? Awesome. So, this has been super fun, Claire. Thank you so much. Um, people can find me on LinkedIn, Daniel Bloom. Um, and also on my website, I'm Daniel Bloom.com. Uh, I can share some of these resources. Um, this this a big part of this uh system is available uh for download. So, check out the website and yeah.

Daniel： 那可太棒了。一个 agent 或者说一个 AI 能把我气到什么程度。

Claire： 这期真的太好玩了。谢谢你给我们完整展示了你的整套系统，我觉得它完全不难上手。嗯，我觉得很多人会把这期内容放进他们自己的 improve skill 里，而且我很喜欢你一直在想办法把它分享给团队里的其他人。大家可以去哪儿找到你？我们能帮上你什么忙？

Daniel： 太好了。这期真的超级好玩，Claire，非常感谢。嗯，大家可以在 LinkedIn 上找到我，Daniel Blum。嗯，还有我的网站 imdanielbloom.com。呃，我可以分享其中一些资源。嗯，这套系统的很大一部分都可以下载。所以去网站看看吧，嗯。

[45:35] Claire

Awesome. Well, thank you for joining How AI.

太好了。那就感谢你来 How I AI 做客。

[45:38] Daniel

Thank you, Claire.

谢谢你，Claire。

[45:40] Claire

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 howiipod.com. See you next time.

非常感谢收看。如果你喜欢这档节目，请在 YouTube 上点赞、订阅，或者更好的是，给我们留言分享你的想法。你也可以在 Apple Podcasts、Spotify 或你常用的播客 App 上找到本节目。欢迎给我们打分、写评论，这能帮更多人发现这档节目。所有往期节目和更多节目信息，都可以在 howiaipod.com 查看。我们下期见。