The AI content machine that turns ideas into posts that don't sound like slop | Alex Lieberman

频道: How I AI
视频: https://www.youtube.com/watch?v=1_jlukb7gm4
原文语言: en
统计: 共 74 轮 · Alex 42 · Claire 29

[0:00] Alex

Creating content has afforded me so many opportunities, but I am capped on the amount of time that I can spend creating content every day. And so, I've been basically thinking about how can I re-engineer my content process to be AI native or AI assisted. The other problem I've been trying to solve is how do I make it as easy as humanly [music] possible for employees to be creators while they have full-time jobs. And so, that was kind of the genesis of this content machine idea was optimizing for these two problems.

做内容给我带来了太多机会，但我每天能花在做内容上的时间是有上限的。所以我一直在琢磨：怎么把我的内容生产流程重造成 AI 原生、或者至少是 AI 辅助的。我想解决的另一个问题是：怎么才能让员工在有全职工作的前提下，尽可能轻松地成为创作者。所以这台「内容机器」的想法，基本上就是从这两个问题里长出来的。

[0:26] Claire

I do read your posts, they're quite good. So, they don't strike me as slop, but prove it live. How would we go through this process?

我确实在看你发的帖子，写得挺好的。所以它们在我看来不像 slop，但你得现场证明给我看。这套流程具体是怎么走的？

[0:32] Alex

So, the first step of the content machine is called the Oracle. The Oracle just does a scan of the last 7 days of information from all of the channels that I'm in, and it ranks a short list of ideas. And it has like a whole scoring system for what it defines as a good content spike, gives it a positive score. I basically get 15 content spikes.

内容机器的第一步叫 Oracle。Oracle 会把我过去 7 天在所有渠道里的信息扫一遍，然后排出一份点子短名单。它有一整套评分系统，用来判断什么算是一个好的「内容爆点」，是的话就给正分。我一般能拿到 15 个内容爆点。

[0:53] Claire

AI is just out of the box pretty terrible at writing. How you get it to not write slop?

AI 开箱即用的写作水平其实相当糟糕。你是怎么让它不写出 slop 的？

[0:58] Alex

The only time, in my view, that the content machine actually produces slop is more of an indictment of the person not sharing good enough ideas during the interview step than the AI writing bad stuff.

在我看来，内容机器唯一真的产出 slop 的时候，与其说是 AI 写得烂，不如说是在访谈那一步，这个人自己没拿出足够好的想法。

[1:09] Claire

Uh-oh, so what you're saying is we are actually the slop.

糟糕，所以你的意思是——slop 其实是我们自己。

[1:12] Alex

My take is that AI slop is hilariously people just pointing the finger at themselves and saying I'm not intelligent enough.

我的看法是，所谓 AI slop，特别好笑的一点在于：那其实是人在指着自己说，我不够聪明。

[1:23]

[music]

（音乐）

[1:23] Claire

Welcome to How I AI. I'm Claire Vale, product leader and AI obsessive here on a mission to help you build better with these new tools. Today, I have Alex Lieberman at 10x, and he's going to show us how he climbs Cringe Mountain, aka builds a content machine at his company by using AI. And guess what? He doesn't get AI slop out. This is an awesome workflow for anybody trying to market their company and get an edge on distribution. Let's get to it. Quick word from today's sponsor, Fire Crawl. If you're building with AI agents, you've probably hit the same wall. Your agent needs data from the web, but the right pages are difficult to find, buried in JavaScript, or blocked behind logins. Firecrawl is a web data API that lets agents search, scrape, and interact with the web at scale, and get that clean, [music] structured data they can actually use. Over a million developers, including myself, build on it. It's open source, and it's free to start. [music] Stop fighting the web for data, and start powering your AI agents and apps with Firecrawl at firecrawl.dev. Use code howiai to get 10,000 free credits today.

欢迎收看 How I AI。我是 Claire Vo，产品负责人、AI 重度爱好者，来这儿是想帮你用这些新工具做出更好的东西。今天请到的是 10X 的 Alex Lieberman，他会给我们演示他是怎么爬上「尴尬之山」的——也就是怎么用 AI 在自己公司里搭起一台内容机器。而且你猜怎么着？出来的东西不是 AI slop。对任何想给自己公司做营销、想在分发上拿到优势的人来说，这都是一套非常棒的工作流。我们开始吧。先插播一下今天赞助商 Firecrawl 的口播。如果你在用 AI agent 做东西，你八成撞过同一堵墙：你的 agent 需要网页上的数据，但真正有用的那些页面要么难找、要么埋在 JavaScript 里、要么被登录墙挡住。Firecrawl 是一个网页数据 API，能让 agent 大规模地搜索、抓取网页并与之交互，拿到干净、结构化、真正能用的数据。已经有超过一百万开发者在它上面构建，包括我自己。它是开源的，而且免费起步。别再为了拿数据跟网页死磕了，去 firecrawl.dev，用 Firecrawl 给你的 AI agent 和应用供数据。用优惠码 howiai，今天就能领 10,000 个免费额度。

[2:36] Claire

Alex, I am excited about this episode because I tell everybody, in order to make it, certainly as a bootstrapped entrepreneur, you have to do one thing, and it's not Vibe Check, it is climb cringe mountain. You have to post. You have to post on X, you have to post on LinkedIn, you have to talk about yourself, and I will say, as a true engineer's engineer, I will build literally anything to avoid talking about my business.

Alex，我对这期特别期待，因为我逢人就说：想做成事，尤其是做一个自力更生（bootstrapped）的创业者，你必须干一件事——不是 vibe check，而是爬上「尴尬之山」。你必须发帖。你得在 X 上发，得在 LinkedIn 上发，得聊你自己。而我得说，作为一个骨子里的工程师，为了不去聊自己的生意，我几乎什么东西都愿意造出来。

[3:05]

[laughter]

（笑声）

[3:06] Claire

People won't be surprised about this. So, please tell me, how have you started to use AI? Why did you start to use AI to build the content engine at your company?

这一点大家应该不会意外。所以跟我说说，你是怎么开始用 AI 的？你当初为什么想用 AI 来搭公司的内容引擎？

[3:16] Alex

Well, first of all, thanks so much for having me. So, I, first of all, I totally ascribe to climbing uh AI cringe mountain. I I probably uh move on like the totally opposite side of the spectrum, which is like, I am happy to be cringey and just like shoot as many shots on goal to to sell kind of what I'm doing, because my belief is people who believe that like, if you build it, they will come, I just, in a world in which distribution is more important than ever ever before, I don't think that's necessarily the case, or isn't most of the time. So, basically there have been two problems within the context of 10X that I've been trying to solve. And just for folks that don't know what 10X is, applied AI company helping mid-market and enterprise companies go through their AI transformation. And we do everything from strategy to four deployed engineering. So, the first is I have been creating content on the internet for the last 10 years, basically since I started um Morning Brew when I was in college. And I think creating content has afforded me so many opportunities, but I am capped on the amount of time that I can spend creating content every day. Like I would say 25% of my time can be spent creating content. And so I've been basically thinking about how can I maximize the value of that 25% of my time? And what I've thought a lot about is how can I re-engineer my content process to be AI-native or AI-assisted while making sure that I'm not putting out AI slop because like I think that's what creates so much aversion to AI for people, especially around creating content, is that it's just going to spit out kind of just really mid stuff that sounds like everything else that's been said on the internet. The second is, you know, every business I build, whether it's a media company or a non-media company, I think we need to think more about building media companies on top of our businesses in the future. And the reason

首先，非常感谢你邀请我。第一，我完全认同要去爬 AI 这座「尴尬之山」。我大概站在光谱的最另一端——我很乐意显得尴尬，就是尽可能多地出手、去推销我在做的事情。因为我相信，那种「你把东西做出来，人自然就会来」的想法，在一个分发比以往任何时候都更重要的世界里，不一定成立，或者说大多数时候都不成立。所以在 10X 这个背景下，我一直想解决两个问题。给不了解 10X 的朋友说一句：10X 是一家应用型 AI 公司，帮中型市场和大型企业做 AI 转型，从战略一路做到 forward deployed engineering（前线部署工程）。第一个问题是：过去十年我一直在互联网上做内容，基本上是从我大学时创办 Morning Brew 开始的。我觉得做内容给我带来了太多机会，但我每天能花在做内容上的时间是有上限的——我大概最多能拿出 25% 的时间来做内容。所以我一直在想：怎么把这 25% 的时间价值最大化？我想得最多的就是，怎么把我的内容流程重造成 AI 原生或 AI 辅助的，同时又保证自己不是在往外倒 AI slop。因为我觉得，恰恰是这一点让很多人对 AI 产生强烈的抵触，尤其是在做内容这件事上——大家怕它吐出来的东西平庸至极，跟互联网上已经被说过的所有话一个味儿。第二个问题是，我做的每一家公司，不管是媒体公司还是非媒体公司，我都觉得未来我们该更多地考虑「在自己的业务之上再搭一家媒体公司」。原因是

[5:14] Alex

for that is in a post-AI world where technology gets more commoditized than ever before, there are fewer moats in business, and I believe that trusted distribution is one of them. So, overall I am building a media company on top of 10X, and I think that, you know, I'm going to do all the obvious things like bring full-time creators onto staff, writers, we're going to have a newsletter, all these things. But I don't think enough companies think about turning their own employees into creators if they want to do that. And so the other problem I've been trying to solve is how do I make it as easy as humanly possible for employees to be creators while they have full-time jobs? And so that was kind of the genesis of this content machine idea was optimizing for these two problems.

在一个 AI 之后的世界里，技术会比以往任何时候都更商品化，生意里的护城河越来越少，而我相信「可信赖的分发」就是其中之一。所以整体上，我是在 10X 之上搭一家媒体公司。那些显而易见的事我肯定都会做——招全职创作者进来、招写手、做一份 newsletter，诸如此类。但我觉得，如果公司想做这件事，认真想过「把自己的员工变成创作者」的公司还太少。所以我想解决的另一个问题就是：怎么才能让员工在有全职工作的前提下，尽可能轻松地成为创作者？这台「内容机器」的想法，基本上就是围着这两个问题优化出来的。

[5:56] Claire

I love this because one I I went through this myself. You know, I am uh at Chief Product Officer on TikTok. I went through this phase where I was like, you know what? Part of my job as an executive when I had those jobs was to be the marketing face of the company and you know, I used to say in SaaS you were you weren't buying the product, you were buying the road map and what I would tell my team is they're not buying the road map, they're buying the team, which is like they believe this team can execute on the vision and so they want to bet with either a subscription or an engagement or whatever on the team. And I talked to a lot of CEOs right now that are really unsettled by the current state of marketing where existing channels have diminishing returns, they're not seeing the old stuff really work and I every single one of them I tell, your number one marketing channel should be your employees. And so I love that you're investing in this. Now my my big concern is um and I have a little bit of a you know, content distribution engine going as well with some AI in it. Is AI is just out of the box pretty terrible at writing compelling copy. And I think we've all seen this. We're all really critical of it, but it seems like you figured this out. So I'd love to hear just a little bit of your process, where you think AI fits, where it doesn't, how you get it to not write slop.

我很喜欢这个话题，一是因为我自己也走过这一遭。你知道，我是 TikTok 上那个「首席产品官」。我也经历过那么一个阶段，当时我想：好吧，我做高管的时候，工作的一部分就是当公司的营销门面。以前在 SaaS 行业我常说，你买的不是产品，你买的是路线图；而我跟团队说的是，他们买的甚至不是路线图，是这支团队——他们相信这支团队能把愿景落地，所以他们愿意用订阅、用合作、用别的什么方式押注在这支团队上。我现在跟很多 CEO 聊，他们对当下营销的状况非常不安：现有渠道回报递减，老一套不太灵了。我对他们每一个人都说同一句话：你的头号营销渠道应该是你的员工。所以我特别喜欢你在这件事上下注。不过我最大的顾虑是——我自己也有一套带点 AI 的内容分发引擎——AI 开箱即用地写有说服力的文案，水平真的相当糟糕。我想这一点大家都见识过，我们也都对它挺挑剔的。但看起来你是把这个问题解决了。所以我特别想听听你的流程：AI 放在哪儿合适、放在哪儿不合适，你又是怎么让它不写出 slop 的。

[7:14] Alex

Yeah, totally. So I have a few thoughts here. We were talking about this before, which is Lulu Meservey has talked about this idea of it's a shame when the best writers in the world use AI because they cap their ceiling and so her view is is AI will raise the floor of bad writers, but then cap the ceiling of great writers. I think I agree with that with nuance. The the first thing I'll I'll say is that writing and creating content is a multi-step process, which we'll go through. And I think just saying that using AI in any part of the content process is going to lower your ceiling is not necessarily true. I think if we were to say for the actual drafting or editing part of the content process, using AI lowers your ceiling, I can buy that and I can understand if you are a gifted writer is in the top 1% of writers, how basically just banking on a model that's been trained on the entire corpus of the internet and so by definition will be average versus you being a above average or an exceptional writer, why it would be worse, totally get that. But there are many parts of a content process and I think some of these parts you can use AI and still continue to actually raise your ceiling, not lower it. The other piece of it is I always think about like alternatives.

对，完全同意。我这里有几点想法。我们刚才聊过这个：Lulu Meservey 讲过一个观点——世界上最好的写作者去用 AI 是件可惜的事，因为那会给他们的天花板封顶。她的看法是，AI 会抬高差写手的地板，但会压低顶尖写手的天花板。我基本同意，但有些细微差别。我要说的第一点是：写作和做内容是一个多步骤的过程，我们待会儿会一步步过。我觉得，简单地说「在内容流程的任何环节用 AI 都会拉低你的上限」，这不一定对。如果我们说的是实际起草或编辑这一环，用 AI 会拉低你的上限——这个我认。我也完全能理解：如果你是个天赋型写手、排在写作者的前 1%，那你去押注一个在整个互联网语料上训练出来的模型——按定义它就是平均水平——而你本人是高于平均、甚至是卓越的写手，结果为什么会更差，这我完全懂。但内容流程有很多个环节，我觉得其中一些环节你用 AI，反而还能继续把上限往上抬，而不是往下压。另外一点是，我总会去想「替代方案是什么」。

[8:35] Alex

So, the alternative for, you know, our team at 10X and and we can talk a little bit in a uh in a bit about how we're incentivizing people to create, but the alternative before this content machine was people just not creating content. And so it's like, if I've two options, get people to create content around their expertise and the things that they're doing in work or not create, of course I'm going to take the first every single day of the week. And so, what I can do is kind of take you through how I thought about my content process and then how I transformed it. And even though we're talking about it in the context of using AI for content, our our view at 10X and my view is like, you kind of use this kind of process mapping and then rebuilding for any process in your work. I This is just kind of the world that I know. Okay. So, this is when I started this process, I basically just drew out exactly how the content process works today. So, for anyone creating content, you basically start by finding inspiration. And historically speaking, the hardest thing for not even just people who are trying to create content for the first time, but like anyone I would actually say especially people who are creators is finding new ideas that feel novel and interesting and just don't feel like tired regurgitation of what's been said is really hard. So first step is finding an idea. Second step or like one B is if it is an idea related to a lived experience where you feel for fully informed about the idea, then you can skip the research step. But for a lot of ideas you'll have to research. Then the next step is you're going to get all of your thoughts out on paper where it's kind of like the the the mental vomiting of your thoughts around the idea. Next step is you decide what is kind of the anchor format that you want to create a piece of content around whether it's long form website, LinkedIn, newsletter, etc. Then you

对我们 10X 团队来说——我们待会儿可以稍微聊聊我们是怎么激励大家去创作的——在这台内容机器出现之前，替代方案就是大家干脆不做内容。所以说白了，如果我只有两个选项：让大家围绕自己的专长和手上正在做的事去创作，或者干脆不创作，那我当然天天都选第一个。所以我可以带你过一遍，我当初是怎么想我的内容流程的，以及我后来怎么把它改造了。虽然我们是在「用 AI 做内容」这个语境里聊，但我们 10X 的观点、也是我的观点是：这种「先把流程画出来、再重建」的做法，可以用在你工作中的任何流程上，内容只是我最熟的那个领域而已。好，所以我刚开始做这件事的时候，就是把今天的内容流程原原本本地画了出来。对任何做内容的人来说，第一步都是找灵感。从历史上看，最难的一件事——不只是对第一次做内容的人来说，其实我觉得对任何人、尤其是对创作者来说——找到那种新鲜、有意思、不像是把别人说过的话嚼第二遍的新点子，真的非常难。所以第一步是找点子。第二步，或者说 1B：如果这个点子来自你的亲身经历、你对它了如指掌，那研究这一步可以跳过；但很多点子你还是得去做研究。下一步是把你脑子里的想法全倒到纸上，就是围绕这个点子来一次「思维呕吐」。再下一步，你要决定用什么作为承载这条内容的主形式——长文网页、LinkedIn、newsletter 等等。然后你

[10:29] Alex

write the piece, then you review a draft, you revise it so you go through the editing process, you publish it, and then um as kind of Gary V talks about in this idea of the content pyramid, you take your anchor piece of content, you turn it into micro content, and you distribute it across all social platforms or platforms where people consume. So this is the way and whether it's video or writing, this is how the process of creating content goes. And so I started with this to basically be like this is what things look like today. Where does it make sense for AI to live in this process, and then should AI be the driver or should AI be the co-pilot?

把这篇写出来，接着审一遍初稿、改一遍，也就是走一遍编辑流程，然后发布。再然后，就像 Gary Vee 讲的那个「内容金字塔」的思路，你把这条主内容拆成微内容，分发到所有社交平台，或者说所有有人在消费内容的平台。所以不管是视频还是文字，做内容的流程就是这么走的。我就是从这儿开始的，先把「今天是这个样子」摆清楚，然后问：AI 放在这个流程的哪个位置才合理？以及在那个位置上，AI 该当司机，还是当副驾？

[11:06] Claire

I would love to just pause here because let's put content aside. This is the exact process I tell everybody to go through when trying to figure out how to apply AI in a function in their day-to-day, in their company. It is like taking this workflow first approach is just the most tractable way to get AI into a system. And so I'm like literally write it out and then make those decisions of like human co-pilot automation. The other thing that I recommend to people is I say like don't don't workflow what you do now given your current constraints. Workflow what you would do in an ideal world given no constraints. Like given no constraints, of course you would cut all this content into multi-channel like a little micro content. Um of course in given no constraints you'd have a researcher always on looking and mining for ideas. And so I think those two things, writing out your workflow, and then also like removing existing constraints from how you design the workflow is really important. We have this episode with um Matt, the CEO of Suzie, and he does this in his marketing team where he's like he takes a a call transcript and he's like how much can I extract out of this? Not in the how much can my team extract out of it, but like in theory. And then kicks off, you know, a hundred different things off of one piece of content.

我特别想在这儿停一下，因为咱们先把内容这件事放一边——这正是我跟所有人说的、在琢磨怎么把 AI 用进某个职能、某段日常、某家公司时该走的流程。这种「先梳理工作流」的路子，就是把 AI 装进一个系统里最靠谱的办法。我的意思是，你真的把它一条条写出来，然后再去做那些判断：这里是人、这里是副驾、这里是自动化。我还会建议大家另一件事：不要按你今天的约束条件去画工作流，要按「没有任何约束的理想世界里你会怎么做」去画。没有约束的话，你当然会把这条内容切成多渠道的微内容；没有约束的话，你当然会有一个研究员一直帮你盯着、帮你挖点子。所以我觉得这两件事——把工作流写出来，以及在设计工作流时把现有约束拿掉——非常重要。我们有一期节目请的是 Suzy 的 CEO Matt，他在自己的市场团队里就是这么干的：拿一份通话记录，然后问「我最多能从这里面榨出多少东西？」——不是问「我的团队能榨出多少」，而是理论上最多能有多少。然后就从这一条内容出发，启动上百件事。

[12:27] Alex

Totally. Yeah, a hundred percent. And what we've found in kind of doing this process mapping process with companies and with clients is first of all, you end up realizing that a lot of efficiencies are found outside of even thinking about using AI because most people have not actually done the process of mapping their processes before and AI was just like the force function to get them to do it and then they find all of this waste in their process. So yeah, I totally agree with this and this is like the unsexy thing to do, but I actually don't know how you reimagine the work you do if you don't start here. It's also why like when we talk about job loss and all these things with AI, the only reason I was able to do this is because I've done the process of creating content you know, thousands of times. And so it shows you the importance of like being a subject matter expert and having vertical expertise. That knowledge at least right now is really needed, otherwise you can't reimagine the work. Okay, so basically I took this and the next step in the process was like again, assuming no constraints, rebuild the entire thing. And so, what came out of that was this thing called the content machine. And I'm going to just show you kind of how it works. Just as like uh a visual and then I'll take you through like me actually running the machine. So, basically here is how the content machine works. So, the first step of the content machine is called the Oracle. And the whole idea is that the content machine, which is just a directory of skills, um it is a plugin in uh you could either do it through the CLI or through Claude Coder Co-work, is it is connected to all of my systems of record at 10x. So, my Slack, my Notion, my meeting notes, uh Linear, Git, Gmail.

完全同意，百分之百。我们跟不少公司和客户一起做这种流程梳理时发现，第一，你会意识到很多效率提升根本不用扯到 AI 就能拿到——因为大多数人从来没真正把自己的流程梳理过一遍，AI 只是那个逼他们去做这件事的外力，然后他们就在自己的流程里发现了一堆浪费。所以这一点我完全同意，这件事听起来一点都不性感，但说实话，如果不从这儿开始，我真不知道你要怎么重新想象你所做的工作。这也是为什么，当我们聊起 AI 带来的失业这些话题时——我之所以能做成这件事，是因为我做内容这个流程已经走过成千上万遍了。这就说明了当领域专家、有垂直领域专长有多重要。至少在现阶段，这种知识是真的需要的，否则你根本没法重新想象这份工作。好，所以我拿着这张图，下一步就是——同样，假设没有任何约束——把整个东西重建一遍。重建出来的结果，就是这台叫「内容机器」的东西。我给你演示一下它是怎么运作的，先看个示意图，然后我再带你看我实际把这台机器跑一遍。基本上，内容机器是这么工作的：第一步叫 Oracle。整体思路是，这台内容机器其实就是一个 skills 目录，它是一个插件，你可以通过 CLI 用，也可以通过 Claude Code 或 Cowork 用；它接到了我在 10X 的所有 system of record 上——我的 Slack、Notion、会议记录、Linear、Git、Gmail。

[14:11] Alex

It is connected to all of these things. And basically what happens is the Oracle just does a scan of the last 7 days of information from all of the channels that I'm in and it ranks a short list of ideas uh that it believes are good content spikes. And it has like a whole scoring system for what it defines as a good content spike. Like if there's a story or an anecdote related, it gives it a positive score. If there's a strong point of view, it gives it a positive score. If there's specific examples, it gives it a positive score. And so, every day I basically get 15 content spikes. Half of them are from the Oracle looking at my internal systems. And then the other other half of them are doing what's called internet reading, where I have provided the content machine a list of the accounts that I follow on X, LinkedIn, and just the websites I go to, and it will go and look at all of the content that's been created by these people over the last 7 days and share those as spikes as well. So, like you're you're one of the people in my internet reader, it will look at everything that you've posted in the last 7 days, and when it thinks there's a good content spike for either me to do a quote retweet on one of your posts or respond, it will share that in the list. So, that's the first step in the process.

它跟这些东西全都打通了。运作方式基本上是：Oracle 会扫描我所在的所有频道过去 7 天的信息，然后排出一份它认为算「内容爆点」的短名单。它有一整套打分体系来定义什么才叫好的 content spike——比如里面带故事或亲身经历的细节，加分；有鲜明的观点，加分；有具体的例子，加分。所以我每天基本上会拿到 15 个 content spike。其中一半来自 Oracle 扫我的内部系统；另一半来自所谓的「互联网阅读」——我给内容机器提供了一份清单，包括我在 X、LinkedIn 上关注的账号，以及我常去的网站，它会去看这些人过去 7 天产出的所有内容，把这些也作为 spike 给我。比如你就是我「互联网阅读器」里的人之一，它会把你过去 7 天发的东西全看一遍，一旦它判断这里有个好爆点——要么我可以引用转发你的某条帖子，要么我可以回应一下——它就会把这条放进清单。这就是整个流程的第一步。

[15:25] Alex

That on its own is maybe the most helpful thing in the entire, like in this entire system. And going back to the the point of even if you believe that AI will create AI slop, and you don't want it drafting your content, just helping you go from blank page to concrete idea, I think lowers the friction so much that you'll actually get in the habit habit of creating. Okay, so then you select an idea, and then you go into something called the interview panel. And basically, what the interview panel is is I had six world-class interviewers codified each as skills, where there's Tim Ferriss, Joe Rogan, Michael Barbaro Barrow, Barbara Walters, Howard Stern, and they just one by one will ask questions about this topic to you until they feel like they've gotten enough information to create a piece of content. One other thing I'll add here that's just not here is it goes straight from Oracle to the interview panel. If it is a topic, and I'll show you what this looks like in a minute actually in Claude, but if you pick a topic that you don't actually feel like you have all the information on already, you can run the research assistant, and the research assistant will give you a full brief into the topic.

光是这一步，可能就是整套系统里最有用的东西。回到刚才那个点：哪怕你笃信 AI 只会生产 AI slop、压根不想让它替你起草内容，光是帮你从一张白纸走到一个具体的点子，我觉得就把阻力降得足够低，低到你真的能养成持续创作的习惯。好，然后你选定一个点子，就进入所谓的 interview panel（访谈团）。interview panel 说白了就是：我把六位世界级的访谈者各自固化成了一个 skill——Tim Ferriss、Joe Rogan、Michael Barbaro、Barbara Walters、Howard Stern，他们会一个接一个地就这个话题向你提问，直到他们觉得已经拿到足够信息、可以做出一篇内容为止。这里再补一件图上没画的事：现在是从 Oracle 直接进 interview panel。但如果你挑的这个话题——待会儿我会在 Claude 里实际演示给你看——如果你挑的话题你自己其实并没有掌握全部信息，你可以先跑 research assistant（调研助手），它会给你一份关于这个话题的完整简报。

[16:39] Alex

So, okay, we go through the interview panel. Uh basically, what I do is I yap to text all of my answers um using WhisperFlow. Then it drafts in in my voice. And so, the way that this works with the content machine is basically, the entire process, like all of these steps, are what live in the content machine repo that everyone on our team has kind of pulled down onto their computer, but separately, all of us have per a personal folder that has codified our voice, our style guide, and like our sources that we want to pull ideas from. And so, when a voice is actually drafted, or when a piece is actually drafted, what's happening is it's actually going into my specific voice and style guide file, and using that to create the piece of content. And so, just to show you what that looks like. So, my style guide here is, you know, this is who Alex is, uh this is his role in the company, these are the owned assets that he's going to want to promote, these are his main anchor types of content, here are his key integrations, here are the social handles that he follows up follows. You see Claire's there, you know, people who give enterprise an exact lens. So, that's the first, and then the second is the voice guide. And so, what this is is basically when I launched the content machine, it studied. So, this is all of my top-performing X posts and LinkedIn posts were studied, has executive summary with voice and tone, hook formulas, content structures, laid patterns, topic categories. And so, this breaks down the top 10 best-performing X posts I've had of all time, my best-performing LinkedIn posts, what my core DNA is, my number one rule, which is writing like you're texting a friend, and then examples of like the language that I use, self-deprecating confidence. Like, it really goes into depth into this file.

好，我们过完 interview panel。我的做法基本上是用 Wispr Flow 把所有回答直接口述成文字。然后它就用我的 voice 起草。内容机器的运作方式大致是这样：整个流程、所有这些步骤，都住在 content machine 这个 repo 里，团队里每个人都把它拉到了自己电脑上；但除此之外，我们每个人各自还有一个私人文件夹，里面固化了自己的 voice、style guide，以及我们想从中取点子的信息源。所以真正起草一篇内容的时候，它其实是去读我专属的 voice 和 style guide 文件，用这些来生成内容。给你看一下长什么样。我的 style guide 里写的是：Alex 是谁，他在公司里是什么角色，他会想推广的自有资产有哪些，他内容的几个主要锚点类型，他的关键整合项，还有他关注的社交账号——你看 Claire 就在里面，都是些能给企业侧提供确切视角的人。这是第一份。第二份是 voice guide。它是这么来的：我启动 content machine 的时候，它做了一轮研究——把我在 X 和 LinkedIn 上表现最好的帖子全都研究了一遍，里面有关于 voice 和语气的执行摘要、hook 公式、内容结构、语言模式、话题分类。它会拆解我历史上表现最好的 10 条 X 帖子、表现最好的 LinkedIn 帖子、我的核心 DNA 是什么、我的第一原则是什么——也就是「像给朋友发短信那样写」——然后是我常用语言的例子，比如那种「自嘲式的自信」。这个文件真的写得非常细。

[18:27] Alex

Like, this is this is my full voice codified. And then, we'll get to it in a minute, but the the other really important piece of this whole process is when a piece of content is drafted by the content machine for me, I will give it feedback, and it will have a reinforcing loop from that feedback. And so, one other file that's in the content machine is the content lessons markdown fold uh markdown file. It is all of the things that I've said it messed up when drafting my content in the past. It logs that as lessons. So, around tone, where I've said the tone wasn't right, structure and flow, when that wasn't right. And so, it will always check the content lessons markdown when editing a piece of content to make sure it's not making the same mistake again.

这就是我完整的 voice 被固化下来的样子。另外，待会儿会讲到，这整个流程里还有一块特别重要：当内容机器给我起草出一篇内容，我会给它反馈，而这些反馈会形成一个强化回路。所以 content machine 里还有一个文件，叫 content lessons 的 markdown 文件。里面记的全是我说过它过去起草我的内容时搞砸的地方，它会把这些记成一条条「教训」。比如语气上，我说过哪次语气不对；结构和行文上，哪次不对。所以它每次编辑一篇内容，都会先去查这份 content lessons markdown，确保不再犯同样的错。

[19:13] Claire

And is this generated as you go through the editing process with a skill? Like is it like if you identify some piece of feedback, are you manually populating that? How is that getting created?

这份东西是你在用某个 skill 走编辑流程的时候顺带生成的吗？还是说你发现某条反馈之后，手动往里填？它到底是怎么产生的？

[19:23] Alex

So, the way that the content machine works is basically after a piece of content is drafted and then it goes through the editing process, which I'll show you in a second. I get the final piece of content. What I will do is I will give it feedback on that final piece of content. Once it's all done and I'm like, "This is good to go. This is good to publish." There's basically a job that's kicked off as part of like kind of this uh the these like chain together skills and it's the the lessons loop. And the lessons loop is, "Let me look at what the the diff is between the original piece that we provided to Alex versus the new piece. What are abstractable lessons we can take from that?" And then we ask Alex, "Do you want these lessons to be added to the lessons file?" If he says yes, then we add them in. So, that's how that works. And then just to finish taking you through the process and then I can kind of show you in practice. That's how my voice is drafted. Then the next step is, so I've shared I've picked an idea. I've yapped a text um all of my thoughts around the ideas as I'm interviewed by this panel of interviewers. Then I pick, say I want to create a LinkedIn post. Uh it's drafted in my voice using my style guide, my voice markdown file, and my content lesson markdown file.

content machine 的运作是这样的：一篇内容起草完，走完编辑流程——待会儿演示给你看——我拿到最终稿。这时我会针对这个最终稿给反馈。等我确认「这版可以了，可以发了」，就会触发一个任务，它属于这一串串联起来的 skill 的一部分，叫 lessons loop（教训回路）。lessons loop 干的事是：「我来看看最初给 Alex 的那版和最终这版之间的 diff 是什么，从中能抽象出哪些可复用的教训？」然后它会问 Alex：「要不要把这些教训加进 lessons 文件？」他说要，我们就加进去。这就是它的运作方式。然后我把整个流程给你讲完，再实际演示。刚才说的是我的 voice 怎么被用来起草。下一步是——我已经挑好了一个点子，在被这一组访谈者轮番提问的过程中，把围绕这个点子的所有想法都口述成了文字。然后我选，比如说我要做一条 LinkedIn 帖子。它就会调用我的 style guide、我的 voice markdown 和我的 content lessons markdown，用我的声音把它写出来。

[20:34] Alex

Once the draft is done, it goes through the writers council. So, similar to how there's the six interviewer personas I created, there's six writer personas I created. So, it's like David Perell, Sean Puri, um Morgan Housel, the uh the AI slop allergist. And they go through, they read the content, and they score it 1 to 10. If the aggregate score is under a 9 out of 10, it runs a revision loop until it scores a 10 out of 10. And then it's done. And then once the piece of content is done, then I can repurpose and distribute, which means there's just a step where I can say, I want to repurpose this. It at the content machine asks, what do you want to repurpose it into? I could say three short tweets, two long LinkedIn posts, and it will go look at what my format of writing in these different uh post types was in the past and write in that style.

草稿写完之后，进入 writers council（写手评审团）。跟前面我造的那六个访谈者人格一样，我也造了六个写手人格，比如 David Perell、Shaan Puri、Morgan Housel，还有那个「AI slop 过敏症专家」。他们会通读这篇内容，各自打 1 到 10 分。如果总分低于 9 分，就跑一轮修订循环，一直跑到打出 10 分为止，然后才算完成。内容定稿之后，我就可以做二次利用和分发——只要说一句「我要把这篇改造复用」，content machine 就会问「你想改成什么形式？」我可以说三条短推文、两条长的 LinkedIn 帖子，它会去看我过去在这些不同帖子形式里写作的格式是什么样，然后照那个风格写。

[21:24] Alex

I think again, this is just a reflection of a a couple things, I would say. One, map your workflow. Two, with AI, you can bring in basically a council of experts without having a council of experts. So, again, it's like in this ideal world, I would be I I would be interviewed by the best of the best. Well, guess what? You can approximate that in your process. I think finding the right reinforcement loops on as I'm building this process, how do I iteratively make it better because you can make sense of inputs and outputs of unstructured data in a way you weren't able to do before, is really useful. And then, um you know, AI's like multimodal. And so, and and I mean that not just in like image to text or text to voice or voice to text. I mean it in like text to mini mini text, text to long text. And so, using all those kind of components of of LLMs, I think makes this particular use case a good one.

我觉得这里面又体现了几件事。第一，把你的工作流梳理清楚。第二，有了 AI，你可以在没有专家顾问团的情况下，凭空请来一个专家顾问团。所以在理想世界里，我会被最顶尖的人来访谈——那猜怎么着？你可以在自己的流程里把这件事近似地做出来。第三，我觉得在搭这套流程的过程中，找到对的强化回路——也就是怎么让它一轮轮地变得更好——非常有用，因为现在你能用以前做不到的方式，去理解非结构化数据的输入和输出。还有就是，AI 是多模态的。我说的多模态不只是图生文、文字转语音、语音转文字，我说的还包括「文字到短文字」「文字到长文字」。把 LLM 这些能力组件全用上，我觉得让这个具体的用例成了一个很好的用例。

[22:21] Claire

This episode is brought to you by Customer.io. [music] You're here because you'd rather use AI than talk about it. With Customer.io, you describe the campaign you want to build and the AI agent creates for you. The audience, the messages, and the timing. You review it, make any changes you want, and launch. Instead of spending [music] hours stitching together tools and workflows, you can focus on the work that actually drives growth. Every campaign is tied back to results, so you can see what's working and what to do next. More than 9,000 brands use Customer.io to turn the data they already have into messages customers remember. Visit customer.io/howiai to try it today. Customer.io, more impact from every message. I did tell you before we got on, I do read your posts, they're quite good. So, they don't strike me as slop, but prove it live. Do it live. How would we go through this this process?

本期节目由 Customer.io 赞助。（音乐）你会在这儿，是因为你宁愿真的去用 AI，而不是光谈论它。用 Customer.io，你只要描述你想做的那个营销活动，AI agent 就替你把东西建出来：受众、文案、发送时机。你审一遍，想改就改，然后上线。不用再花上（音乐）几个小时把各种工具和工作流拼在一起，你可以把精力放在真正带来增长的事情上。每一个活动都能对回结果，所以你能看清什么在起作用、下一步该做什么。超过 9000 个品牌在用 Customer.io，把他们手里现成的数据变成客户会记住的信息。今天就去 customer.io/howiai 试试。Customer.io，让每一条信息都更有影响力。——我们开录前跟你说过，我确实读你的帖子，写得挺好的。所以在我看来它们不像 slop。但你得现场证明一下，来真的。这个流程我们要怎么走一遍？

[23:20] Alex

Let me bring up my Claude and I'll just show you how one example of how it already ran and then we'll do it live also and I will happily embarrass myself if this goes terribly wrong. So, okay. So, basically what we're looking at is the content machine running and and just so you can see cuz this may be helpful like the content machine lives as a plugin in 10X's um our internal Claude and basically everyone has added it to their machine. Anytime I'm basically pushing an update to it on Git, people can just auto update it right from here. Um so, I was trying to figure out something that was for anyone who was not technical in the business was palatable enough to as we push updates to benefit from those updates and we can talk about it in a second like how sharing this with everyone in the business is getting us closer to what I think it was the CEO of Sendbird who went through like their internal quest uh thing with you. Like how we're getting closer to that. But so, let me go back to this for a second. So, basically what this did is it first looked at my sources, internal sources, it pointed out spike ideas. So, this was like a single player versus multiplayer AI framework that we talk about a lot. I talked to someone at a bank who's working with McKinsey and McKinsey gave them an idea and was like this is going to take 6 months to build. He said he rebuilt it in a week with Claude code.

我把 Claude 打开，先给你看一个已经跑完的例子，然后我们再现场跑一次——万一跑砸了，我也乐意当众出个丑。好，我们现在看到的就是 content machine 在跑。顺便给你看一下，这个可能挺有用：content machine 是以插件（plugin）的形式装在 10X 我们内部的 Claude 里的，基本上每个人都把它加到了自己机器上。我只要往 Git 上推一次更新，大家在这儿就能一键自动更新。我当时就是想找一个对公司里非技术同事也足够友好的方式，让我们推更新的时候他们都能受益。待会儿可以聊聊，把这套东西开放给公司里每个人，正在怎样让我们更接近——我记得是 Sendbird 的 CEO 在你节目上讲过他们内部那套 quest 的事——更接近那种状态。不过我们先回到这里。它做的第一件事是先看我的信息源、内部信息源，然后指出一批 spike 点子。比如这个是我们经常聊的「单人版 AI 与多人版 AI」框架。我跟一家银行的人聊过，他们在跟 McKinsey 合作，McKinsey 给了他们一个想法，说这东西做出来得花 6 个月。他说他用 Claude Code 一周就重做出来了。

[24:44] Alex

Me saying to someone on a sales call that I may be the world's worst salesperson, that sounds like a fun one. After I said to someone on a call after I sold Morning Brew, I was miserable. Like that's a good content spike. So, like you can see it's really latching on like specific stories and then from the internet reader it pulled from the different sources that I've given it in my profile file. So, one was, you know, Amazon putting a billion dollars behind FDEs, Aaron Levie versus versus Ethan Mollick talking about like are FDEs the solution to every enterprise's problems versus not, everyone's a builder now shipping is the hard part. And so then I would just select a content spike that I want to actually build out. And just so you know, here the other thing in the content machine is well, I'm going to select one idea here to build out into a piece of content. There are several ideas here that I think are worthy of building out. And so part of the content machine is when everyone at the company sets it up for the first time what's created is the vault and the vault is literally just a notion database and every idea that is generated by the Oracle, even if you don't use it, is added to this vault.

还有我在一次销售通话里跟对方说，我大概是世界上最差的销售——这听起来就挺有意思。还有我卖掉 Morning Brew 之后，在一次通话里跟人说我当时很痛苦。这就是很好的 content spike。你能看到，它是真的会抓住那些具体的故事。然后互联网阅读那部分，是从我在 profile 文件里给它的那些信息源里抓的。比如一条是 Amazon 拿十亿美元押注 FDE；还有 Aaron Levie 和 Ethan Mollick 在争论 FDE 到底是不是所有企业问题的解药；还有「现在人人都是 builder，难的是把东西发出去」。然后我就从里面挑一个我真正想展开去做的 content spike。顺便说一句，content machine 这里还有一点：我这次只会挑一个点子展开成一篇内容，但这里其实有好几个我都觉得值得做。所以 content machine 的一部分设计是：公司里每个人第一次配置它的时候，会创建一个 vault（点子库），vault 说白了就是一个 Notion 数据库，Oracle 生成的每一个点子——哪怕你这次没用——都会被丢进这个 vault。

[25:49] Alex

So, you can go into that in the future to pull ideas out of. Okay, so I picked spike 13. And spike 13 I thought was interesting. Aaron Levie saying like FDEs are God's greatest gift to earth right now. Mollick being like no, it's not going to solve all the problems. And so I picked this and I wanted to do more research on the topic around like what are the different arguments around FDEs? Why do Aaron versus Ethan have different perspectives? So then the research step was started up. And you can see basically a research brief was created for me about their takes, what's been said, what are potential contrarian or novel angles that I can take here, what are open questions that I should try to answer. So, I read this beforehand. I can also as I'm doing the interview, if I think there are like really good quotes from this, I can pull that into my answers as well. So, then I go in and the interview starts. So, I forgot Larry King was in here also. So, let let Larry King is like Levie says FDEs win because deploying agents is more technical than people think. Mollick says they'll disappoint because the real problem is org structure and expertise, not code. And he said, "Uh you said they're both right. Here's what uh what they're missing." So, in one or two sentences, what are they both missing? I provide my thoughts on like, are FDs the solution to all? Why do I agree or not with Aaron and Ethan?

所以你以后可以回到这个库里翻点子出来用。好，我挑了 spike 13。spike 13 我觉得挺有意思：Aaron Levie 说 FDE 是当下上天赐给人间最好的礼物，Mollick 则说不，它解决不了所有问题。我挑了这个，然后我想就这个话题多做点调研：关于 FDE 都有哪些不同的论点？为什么 Aaron 和 Ethan 的看法不一样？于是调研这一步就启动了。你能看到，它给我生成了一份调研简报：他们各自的观点、目前都有哪些说法、我可以切入的反共识或新颖角度有哪些、有哪些值得我去回答的开放问题。这个我事先读过了。而且在做访谈的时候，如果我觉得里面有特别好的引述，也可以把它揉进我的回答里。然后我进去，访谈开始。哦，我忘了 Larry King 也在里面。Larry King 是这么说的：Levie 认为 FDE 会赢，因为部署 agent 比大家以为的更偏技术活；Mollick 认为 FDE 会让人失望，因为真正的问题是组织结构和专业能力，不是代码。他接着说：「你说他们俩都对，那他们都漏掉了什么？」用一两句话说说他们俩都漏了什么。于是我给出我的想法：FDE 是不是万灵药？我为什么同意或不同意 Aaron 和 Ethan？

[27:08] Alex

And then the follow-up question is just playing off of my answer. So, now it asks, "And and the interview panel is really trained to pull as much specificity and examples out of me as possible." So, Tim Ferriss is like, "Give me the engagement where this was painfully true. A bit uh a build your team nailed technically." So, it asked me to like give a customer story. So, I did that. So, we go through. I was asked five questions. Then it says, "Interview's a wrap." And voice and lessons locked in. So, this is where it looks at my content lessons and voice MD file. And then LinkedIn draft MD is created. So, here two of the smartest people I follow just publicly disagreed about the hottest job in AI. They're both right and that's exactly the problem. By the way, if I was to give feedback on this, I'd be like, "This feels kind of AI cringey. They're both right like the the two the the punch Yeah, the this that is just is like the the newest M dash." And so, I'd give that feedback and I'd say add this to content lessons. This week Amazon committed a billion dollars to afford deployed engineering or open AI and Anthropic already did versions of the same thing. Google is hiring hundreds.

接下来的追问就是顺着我的回答往下挖。这个 interview panel 被训练得非常擅长从我嘴里榨出尽可能多的具体细节和例子。比如 Tim Ferriss 会说：「给我讲一个这件事痛得特别真切的项目——一个你团队在技术上做得很漂亮的交付。」也就是让我讲一个客户故事，我就讲了。这样一路走完，我一共被问了五个问题，然后它说「访谈结束」，voice 和 lessons 已锁定。这一步它就会去读我的 content lessons 和 voice.md 文件。然后 LinkedIn 草稿的 md 文件就生成了。开头是这样：「我关注的两个最聪明的人，刚刚公开在 AI 领域最热的岗位上吵起来了。他们俩都对——而这恰恰就是问题所在。」顺便说，如果让我给这段反馈，我会说：这读着有点 AI 尴尬味。「他们俩都对」这种——那种两段一顿、抖个包袱的写法——「就是这个、就是那个」的句式，简直就是新一代的破折号（em dash）。所以我会把这条反馈给它，说把它加进 content lessons。接着往下写的是：这周 Amazon 拿出十亿美元投入 forward deployed engineering，而 OpenAI 和 Anthropic 早就做过类似的事，Google 也在招几百人。

[28:10] Alex

And so, just it includes basically all of my thoughts here. Like, one of my answers was, "What do companies need to set themselves up for success for AI transformation?" And it took my exact words. Where it was like, "So, if you're staffing AI transformation, you you need these things." One of the reasons that this doesn't write slop mostly is because I have basically trained it to only use my words. So, I think that's a big thing also is like this whole content machine is built on the fact that the interview I do with the panel is just turned into a large markdown file of the transcript. And the transcript really should be the only words that are used other than maybe the hook and the conclusion. And the job of the writer is really to almost be this like like sh- like shaping the clay of the content, not inventing net new things. And so, the only time, in my view, that the content machine actually produces slop is more of an indictment of the person not sharing good enough ideas during the interview step than the AI writing bad stuff.

所以它基本上把我在这儿讲的所有想法都收进去了。比如我其中一个回答，是回应「公司要让 AI 转型成功，需要具备什么？」它直接用了我的原话，写成「所以如果你要为 AI 转型配人，你需要这几样东西」。这套东西之所以基本不产出 slop，一个重要原因就是我基本上把它训练成只用我自己的话。所以我觉得这也是个大关键：整个 content machine 建立在一个事实之上——我跟访谈团做的那场访谈，最后就变成一个很大的 markdown 转录文件；除了 hook 和结尾之外，正文里用的词就应该只来自这份转录。写手的活儿其实更像是给内容捏泥巴、塑形，而不是凭空发明新东西。所以在我看来，content machine 唯一真的产出 slop 的时候，与其说是 AI 写得烂，不如说是在暴露这个人在访谈那一步没给出足够好的想法。

[29:14] Claire

Uh-uh. So, what you're saying is we are actually the slop.

哦豁。所以你的意思是，其实我们自己才是那个 slop。

[29:18] Alex

That is This is my take is that AI slop is hilariously people just pointing the finger at themselves and saying, "I'm not intelligent enough." And so, and then the other thing it does, and I just actually recently added this in as a content lesson, is I was like, "Start giving me more examples of hooks because I want to be able to choose and compare." So, it provides hooks and closings. And so, what I would do here Oh, so it didn't even run the writer's council. Yeah, so let's just even do that. Run the council. So, what it's going to do now is it has the draft, it has the hooks and closings, it's going to run the editorial council. If it's above a 9 out of 10, it'll give it to me as is. If it's not, it'll run a revision loop. And then what I would do after, when I get it, is I will give it feedback like the feedback I gave you around that like if this then that line being cheesy, I'll go through, I'll make sure they're added to the content lessons markdown file, and then I'll take a final pass and I'll post this. And then my view, and this is something I want to add to the content machine, is right now the content machine is not when I run the oracle, it's not pulling previous pieces of content that I should that should think about repurposing. That should be part of the run I do every single day. And so, see, they're they're all grading right now, and I can already tell it's not going to be a 9 out of 10, and so they're going to run a full revision loop.

就是这样。我的观点是，所谓 AI slop 特别好笑的地方在于，大家其实是在把手指指向自己，说「我不够聪明」。它还会做另一件事——这个其实是我最近才作为一条 content lesson 加进去的——我当时说：「以后多给我几个 hook 的选项，我想能挑、能比。」所以它现在会给出多个开头（hook）和结尾。那我在这儿要做的是……哦，它连 writers council 都还没跑。行，那我们就跑一下，跑评审团。它现在要做的是：手上有草稿、有 hook 和结尾，接着跑编辑评审团。如果分数高于 9 分，它就原样交给我；如果没到，就跑一轮修订循环。等我拿到之后，我会像刚才那样给它反馈——比如我刚跟你说的那条，那种「A，而这恰恰就是 B」的句式太油腻了——我会通篇过一遍，确保这些都被加进 content lessons markdown 文件，然后我自己再终审一遍，就把它发出去。另外我的想法是，有件事我想加进 content machine：现在我跑 Oracle 的时候，它不会把我以前写过的、值得二次利用的内容也一并捞出来。这件事应该成为我每天那一跑的一部分。你看，他们现在都在打分，我已经能看出来分数到不了 9 分，所以他们接下来要跑一整轮修订。

[30:32] Claire

Oh my gosh, I love it. And then you're going to get this live.

天呐，我太喜欢了。然后你要把这条真的发出去。

[30:35] Alex

Yep. So, like I will, right after we are done with this, I will post this live.

对。我们这边一录完，我就立马把它发出去。

[30:38] Claire

Well, and then what I want to say, you know, maybe what I want to cover after we've we've wrapped this flow, which you've kind of showed us the end to end of like how how do we discover a good idea?

那我想说的是——等我们把这套流程走完之后，我想再聊聊，你刚才已经从头到尾给我们演示了「怎么发现一个好点子」。

[30:47] Alex

Yeah.

嗯。

[30:47] Claire

How do we extract um via again, our friend Hillary Gridley, um via the Yappers API, all of our best ideas. How do we shape that in our voice and then get some critical feedback and ship it? But you're also it's not 100% AI. You also have your team and you're changing team behaviors to all embrace this new model of content creation. So, can you just walk us through some like the non-AI parts of of your content machine and some things you're experimenting with at the company?

怎么把我们最好的点子抽取出来——又是我们那位朋友 Hilary Gridley 讲的那套，走「Yappers API」；怎么把它塑造成我们自己的 voice，再拿到一些尖锐的反馈，然后发出去。但这套东西也不是 100% 靠 AI。你还有你的团队，而且你在改变团队的行为习惯，让大家都接受这种新的内容生产模式。所以能不能带我们过一遍，你那台内容机器里非 AI 的那些部分，还有你们公司正在试的一些东西？

[31:15] Alex

So, this actually this content First of all, the I do not want to claim to be the inventor of employee advocacy. This has been around for a very long time, but I just think that the internet and the platforms on top of it have enabled it in a way that I don't think companies have caught up to it or they worry so much about the risks of giving everyone in your company a megaphone that they don't end up doing it. And so, I tested this for the first time at um one of my other businesses called Storyarb, where we did a campaign called Own the Internet and the idea was that uh it was going to be a six- or eight-week campaign. Everyone at Storyarb had to post on LinkedIn or not had to. They were encouraged to and there was an opportunity to win money. Like I think it was like 5,000 bucks the winner would get. And the only rule was that 50% of the content you create has to be within the world of what Storyarb does. And it ended up driving so much traffic for the business. It drew in the quarter that we ran Own the Internet, it drove 40% of all of our inbound leads. It drove a ton of top of funnel for hiring. And so, I just knew that I would do this like any opportunity I have to do this, I will do it moving forward. And so, we literally just started yesterday 10X's version of this.

这套内容……首先，我不想把「员工代言（employee advocacy）」说成是我发明的，这事儿早就存在很久了。我只是觉得，互联网、以及跑在互联网之上的这些平台，把它的威力放大到了一个程度，而公司们还没跟上；或者说，他们太担心「给公司每个人发一只大喇叭」的风险，结果就干脆不做。所以我第一次试这件事，是在我另一家公司 storyarb。我们搞了一个叫 Own the Internet 的活动，设想是做六到八周：storyarb 的每个人都必须在 LinkedIn 上发帖——不是必须，是被鼓励去发，而且有机会赢奖金，好像冠军能拿 5000 美金。唯一的规则是，你产出的内容里得有 50% 落在 storyarb 所做的事情的世界里。结果它给公司带来了巨量的流量。在我们跑 Own the Internet 的那个季度，它贡献了我们全部 inbound leads 的 40%，还给招聘带来了大量的漏斗顶部流量。所以我当时就知道：以后只要有机会做这件事，我一定会做。于是我们昨天刚刚启动了 10X 版本的这一套。

[32:30] Alex

And so, I'll just share like what the uh details are behind it. So, it's called the uh 10X Creator Cup and it's a month-long challenge to get everyone at 10X posting. And the way you play is you just post on LinkedIn or X, and then we have a Slack channel internally called Reply Guys, and that's where everyone can just see everyone's uh posts throughout the uh Creator Cup. And the way it works is every time you post, you get 10 points. Every time you engage with someone else in the company's post, you get three points. Every week, my co-founder and I will do an editor's pick of the best post of the week, you'll get plus 50. And there are weekly games, and then there's month-long games. And the whole point here is we want to encourage people to post where it's not just a race for impressions, where one person ends up crushing it with impressions. It's like you you we don't want people feeling like they don't have a chance. And we also want it to feel palatable for people, where just like playing the game is is going to get you something. So, even example this week is the week-long game this week is called full house, where if at least 70% of everyone in the company posts at least once this week, then we unlock the one week the week one prize, where everyone gets a a um lottery ticket, and we pull one person and they win a couple hundred bucks this week. And so, we're giving out $5,000 of prizes this week. But the interesting thing here is like we're talking about this earlier.

我就分享一下背后的细节吧。它叫 10X Creator Cup，是一个为期一个月的挑战，目的是让 10X 的每个人都开始发帖。玩法就是：你在 LinkedIn 或者 X 上发帖，我们内部有个 Slack 频道叫 Reply Guys，整个 Creator Cup 期间，所有人的帖子都会汇到那儿，人人都看得见。计分规则是：每发一帖得 10 分；每去互动一次公司里别人的帖子得 3 分；每周我和联合创始人会评一个「编辑精选」当周最佳帖，选中加 50 分。既有周赛，也有贯穿整月的赛制。核心用意是：我们想鼓励大家发帖，但不能让它变成一场纯拼曝光量的比赛——那样最后就是某一个人把曝光量拉爆，我们不想让其他人觉得自己根本没戏。我们也希望它对每个人都是可接受的：光是参与这场游戏，你就能拿到点什么。举个例子，本周的周赛叫 Full House：只要公司里至少 70% 的人这周发过至少一帖，我们就解锁第一周的奖励——每个人拿到一张抽奖券，然后抽出一个人，这周赢走几百美金。所以这一周我们要发出去 5000 美金的奖品。不过有意思的是，就像我们刚才聊到的那样。

[33:54] Alex

And I And I said, "10X, we're playing the game on hard mode." And would you say you're playing the game on honest mode? Yeah, like I think so, 10X is fully bootstrapped, and I think it requires us to have to be creative with certain things. And one of those things is we truly have the best engineers that I've ever worked with in the world. And there's there's a few reasons for that. But we do not have like the gravitas yet of a lab, right? Like OpenAI or Anthropic. And because we haven't raised a boatload of money also, we don't have like the visibility of being on TechCrunch and being mentioned by all of these top-tier investors. And so, the question becomes, how do we attract top talent when we do not have all the advantages that a venture-backed company would have? Part of my view is is the way you do it is have your top talent show versus tell the cool work that they're doing. And so, to me, even if the 10X Creator Cup does not generate any leads for our business, if it just leads to more engineers knowing about 10X and applying to 10X and in a perfect world, if we hire one engineer from this whole campaign, well, then the $5,000 we would have paid out we're paying out for 10X Creator Cup is worth it so many times over cuz you think about what you'd pay pay a recruiting agency to get one engineer, it makes it entirely worth it. And so, yeah, like I just believe that like we lean we should lean into the bootstrapping thing, which is like have an underdog mentality, use guerrilla marketing, us all in this together to kind of create this aura around 10X.

我当时说：「10X 是在用困难模式打这局游戏。」——那你会说你们是在用「诚实模式」打吗？对，我觉得是。10X 完全是自筹资金（bootstrapped），这就逼着我们在某些事情上必须有创意。其中一件事是：我们真的拥有我这辈子合作过的、全世界最好的工程师。这背后有几个原因。但我们还没有那种实验室级别的分量，对吧？像 OpenAI、Anthropic 那样。而且因为我们也没融过一大笔钱，所以我们没有那种曝光度——上 TechCrunch、被一堆顶级投资人挂在嘴边。于是问题就变成：在不具备一家有风投背书的公司所拥有的那些优势的情况下，我们怎么吸引到顶尖人才？我的部分答案是：让你的顶尖人才把手上那些酷活儿「展示出来」，而不是「说出来」。所以对我来说，哪怕 10X Creator Cup 一条业务线索都带不来，只要它让更多工程师知道 10X、来投 10X——最理想的情况是，整场活动帮我们招到一个工程师，那我们为 Creator Cup 付出去的这 5000 美金就值回好几倍。你想想为了招到一个工程师要付给猎头公司多少钱，就完全值了。所以是的，我就是觉得我们应该顺着「自筹资金」这件事往下走：抱着挑战者（underdog）心态，用游击式营销，大家拧成一股绳，给 10X 造出一种气场。

[35:27] Claire

I just as another bootstrap founder could another bootstrap founder that's leaning on distribution as an edge, I I completely I completely agree. I think the other thing is we really underestimate what a employee teammate value proposition it is to say like we actually encourage you to have a platform. So many companies are like short-sighted because I I actually hear CEOs saying, "Well, I don't want that person talking about how awesome they are cuz somebody's going to come recruit them."

作为另一个自筹资金的创始人，一个同样把「分发」当作优势的创始人，我完全、完全同意。我觉得还有一点是：我们严重低估了「我们其实鼓励你去拥有自己的平台」这件事，作为给员工、给队友的价值主张有多值钱。太多公司在这上面特别短视，因为我真的听过 CEO 说：「我可不想让那个人到处讲自己有多厉害，那会有人来把他挖走。」

[35:54] Alex

Yep.

对。

[35:54] Claire

And I'm like, "That is the number one thing that's going to jump them out of this company is their inability to talk about all the amazing stuff that they're doing and build their own personal brands." And so, I think that is just totally totally spot on and um again, I love to see another bootstrap founder do it. As I say, not on hard I mean, maybe on hard mode, but definitely on on honest mode.

我就想说：「真正会让他跳槽离开这家公司的头号原因，恰恰是他没法讲自己在做的那些牛事、没法建立自己的个人品牌。」所以我觉得你说得太对了。再说一次，我特别喜欢看到另一个自筹资金的创始人这么干。就像我说的，不是困难模式——好吧，也许确实是困难模式，但绝对是诚实模式。

[36:15] Alex

Totally. And and even one of the things that I think about is like if I think about what even Anthropic has done really well on the internet is they've basically been incredible, whether um planned or not at employee influencer. Like if you think about everyone surrounding Cloud Code and Co-work, whether it's Tariq or Boris or Felix, right? Like they are extensions of the brand now and I trust their product more because they're talking about it and not just the company.

完全同意。我甚至会想，Anthropic 在互联网上做得特别好的一件事就是——不管是有意还是无意——他们在「员工即影响力者」这件事上做得非常出色。你想想围在 Claude Code 和 Cowork 周围的那群人，不管是 Thariq、Boris 还是 Felix，对吧？他们现在就是品牌的延伸。而正因为是他们在讲这个产品，而不只是公司在讲，我反而更信任这个产品。

[36:44] Claire

Totally. Completely. Well, Alex, this was awesome. This has given me so much inspiration. Um I have a similar kind of content engine uh Open Claw that uh his name is Max, but you know, same same thing. Well, I would love to wrap with quick lightning round questions and then we will get you out of here. One, you've said that you work with the best engineers in the world or at least the ones that you've worked with. What do you think makes for a great engineer right now?

太对了，完全同意。Alex，这期太棒了，给了我特别多灵感。我自己也有一台类似的内容引擎，是用 OpenClaw 搭的，他叫 Max，不过嘛，本质是一回事。最后我想用几个快问快答收尾，然后就放你走。第一个：你说过你和世界上最好的工程师共事——至少是你合作过的人里最好的。你觉得当下什么样的人才算得上优秀工程师？

[37:10] Alex

Someone who has like the ability to understand systems at a really deep level, like the foundational things that have made engineers always good, but who have the malleability to fully lean into making their workflows agentic. Like I think you have a lot of really young engineers who are fully AI-pilled but don't have the foundation of knowledge that, you know, a staff or senior-level engineer has. And then I think you have a lot of veteran engineers that are not malleable enough to change the way in which they work. And and I think the other thing that comes to mind is like this idea of the forward-deployed engineer is more important than ever before because AI transformation is only going to happen by understanding a business's context. Um and so yeah, that's one piece and the other piece to it is I think the best engineers in the world want to go long themselves. I think they want to bet on their abilities and that's why like we pay our engineers like sales people at 10x. People think we're crazy for doing it, but my view is I want to self-select for people that are willing to bet on their own abilities.

能在非常深的层面上理解系统的人——就是那些一直以来让工程师之所以优秀的基本功；但同时又足够有可塑性，愿意彻底地把自己的工作流改造成 agentic 的。我觉得现在有很多非常年轻的工程师，完全被 AI 洗过脑、彻底信这一套，但缺少 staff 级或资深工程师那种知识底子。另一边又有很多老将工程师，可塑性不够，改不动自己干活的方式。我还想到一点：forward-deployed engineer（前置部署工程师）这个概念比以往任何时候都重要，因为 AI 转型只有靠真正理解一家企业的业务上下文才可能发生。这是一方面。另一方面是，我觉得世界上最好的工程师都想「做多自己」。他们愿意押注在自己的能力上——这也是为什么在 10X，我们像给销售发薪那样给工程师发薪。很多人觉得我们疯了，但我的想法是：我想用这套机制自然筛出那些愿意押注自己能力的人。

[38:14] Claire

I agree. I You know how long I've said like PMs and engineers on variable comp? Forever. Forever. I think it's the best.

我同意。你知道我说「PM 和工程师应该拿浮动薪酬」说了多久了吗？说了一万年了。我觉得这是最好的方式。

[38:21] Alex

It It's incredible and we still like again, there's a ton of people who give us about it and look, there are trade-offs, but we have thought about all of them and it has kept all of our people close to the frontier because the way they get leverage is by understanding what's possible with the technology.

效果好到不可思议。而且我们还是——再说一次——有一大堆人因为这个来喷我们。你看，这里面确实有取舍，但那些取舍我们全都想过了。而且它让我们所有人都紧贴前沿，因为他们拿到杠杆的方式，就是搞清楚这项技术究竟能做到什么。

[38:37] Claire

Well, and I will hype up your people. We've had two other 10X folks on the podcast. Uh both both amazing and you should go go check out those episodes with CJ and JJ. Now, do you hire people that don't have two-letter

那我也来给你的人吹一波。我们播客还请过另外两位 10X 的同事，两位都超级棒，大家去看看 CJ 和 JJ 那两期。话说，你们招不招名字不是两个字母的人——

[38:50] Alex

I was going to say, I really I need to change my name to AJ. I really messed up the flow of the 10X guests.

我正想说，我真该把名字改成 AJ。我把 10X 嘉宾的这个节奏给破坏了。

[38:55] Claire

Well, I married to an EJ, so if if you need another another J guy on the team, we can do that. Okay, second question which is you know, you showed us your your creative process. We've seen, you know, co-work with JJ. We've seen coding with CJ. What are my personal use cases? Like what is an AI use case that you use in your personal life that you think is really unique?

我老公就是个 EJ，所以你们要是团队里还缺一个「J 字辈」的人，这事儿能办。好，第二个问题：你刚给我们看了你的创作流程；我们看过 JJ 演示 Cowork，看过 CJ 演示写代码。那个人生活里的用法呢？你在自己生活中用的哪个 AI 用法，是你觉得特别独特的？

[39:18] Alex

So, my my wife was um recruiting for jobs for a period of time. She's worked in tech and she's recruiting for jobs for a while and I think one of the more interesting use cases that I built for her that I'm surprised a company hasn't done yet is the recruiting process is brutal and finding roles that are right for you is brutal. But I think we're in this age like people have always talked about when we're working on Morning Brew like someone's going to build the one of one newsletter. Like the news like everyone has their own customized newsletter in this world. But basically what I built was a hiring board customized for her where every single day she got an email and the email was 10 curated jobs based on her previous job like her LinkedIn and what she's looking for and scored on that and then she could click apply from the email and it auto-fill out the initial application to the job. So, that's like one specific use case. Uh the other one in this I feel like this is like the classic one is building a custom children's book. We have an 11 and a half month old at home and you know, the idea that our our sweet Brooke can have stories that actually relate to her family to me is such a cool use case as well.

我太太有一阵子在找工作。她一直在科技行业工作，找工作找了一段时间。我给她做的东西里，比较有意思、而且我很惊讶居然还没有公司做出来的一个用法是——求职过程实在太折磨人了，要找到真正适合你的岗位也太折磨人了。但我觉得我们正处在这样一个时代：我们做 Morning Brew 的时候大家就一直在聊，说总会有人做出「一对一的 newsletter」，在那个世界里，每个人都有一份完全为自己定制的 newsletter。而我做的基本上就是一块为她定制的招聘看板：她每天会收到一封邮件，里面是 10 个精选岗位，依据是她过往的工作经历、她的 LinkedIn、以及她想找什么样的机会，并按这些维度打分；然后她可以直接在邮件里点「申请」，系统会自动帮她把这个岗位的初次申请表填好。这是一个具体的用法。另一个——我觉得这个算是最经典的——是做定制儿童绘本。我们家有个十一个半月大的孩子，我们可爱的 Brooke 能读到真正跟她自己家人有关的故事，对我来说也是个特别酷的用法。

[40:33] Claire

I love it. You're going to have to join our How I AI parent round table because we have so many good use cases. All right, last one I ask everybody when your AI Oracle creative council is giving you straight slop, what is your prompting technique? Do you get mad? Do you bribe?

太喜欢了。你得来参加我们 How I AI 的「家长圆桌」，因为我们攒了一堆特别好的用法。好，最后一个，这个我每位嘉宾都问：当你那个 AI 神谕、那个创作顾问团给你吐出来的纯粹是 slop 的时候，你的 prompt 技巧是什么？你会发火吗？会贿赂它吗？

[40:56] Alex

I would say the honest answer is I will do It basically depends on how frustrated and how much time I have.

老实说，答案是我会……基本上取决于我有多抓狂、以及我还有多少时间。

[41:05] Claire

I like how flustered you are. You're like, uh-oh, should I tell the truth?

我喜欢你这副慌了神的样子。你那表情像在想：糟了，我该不该说实话？

[41:08] Alex

No, no, no. So either I will just literally say screw it and I'll just write the thing by hand. Like that that I think that's like my typical I'll just be like screw this. It's it's taking more time than it's worth. I'll write the thing by hand. Or the other thing I'll do is like I will actually lean in and spend a ton of time into this being like this is shitty. These are all the ways that these things are shitty. I'm pretty sure these were content lessons already. Why is this not registering as a content lesson I've already taught you? So I I really should be kinder to it given that the AI overlords will be my boss at some point, but honestly I am sometimes an to it.

不不不。要么我就直接说「算了」，自己动手写。我觉得这是我最典型的反应：直接「去他的」，这玩意儿花的时间已经不值了，我自己手写。要么就是反过来，我会一头扎进去，花大把时间跟它掰扯：这东西很烂，它烂在这、这、还有这。我很确定这些早就是已经存好的 content lesson 了，为什么这条没被识别成我已经教过你的 content lesson？所以嘛，考虑到 AI 霸主早晚会当我的老板，我真该对它客气点，但说实话，我有时候对它是真挺不客气的。

[41:44] Claire

So the the classic like no, why are you the way you are?

就是那句经典的——「不是，你怎么是这么个样子？」

[41:48]

[laughter]

（笑）

[41:48] Alex

I've not asked that yet.

这个我还真没问过。

[41:50] Claire

Alex, this has been so helpful truly. Going to steal all these ideas which is the secret reason why I do do these interviews. So thank you for sharing with us with our audience. Where can we find you and how can we be helpful?

Alex，这期真的太有用了。我打算把这些点子全偷走——这其实就是我做这些访谈的私心。谢谢你跟我们、跟我们的听众分享这些。大家在哪儿能找到你？我们能帮你做点什么？

[42:03] Alex

Yeah, I am Business Barista on X, Alex Lieberman on LinkedIn and 10X is our company. So you're a big business you need help with AI transformation, hit us up at 10x.co.

好。我在 X 上叫 Business Barista，LinkedIn 上就是 Alex Lieberman，我们公司叫 10X。所以，如果你是家大公司、需要人帮忙做 AI 转型，就到 10x.co 来找我们。

[42:16] Claire

And if you're an awesome engineer, Alex is hiring.

另外，如果你是个很厉害的工程师，Alex 正在招人。

[42:18] Alex

Yes.

没错。

[42:19] Alex

All of them.

全都招。

[42:20] Claire

All literally all of them.

全部——真的是所有岗位都招。

[42:21] Claire

far as the content machine is telling me, you are hiring all of them.

……就我从那台内容机器上看到的，你是所有人都要招。

[42:25] Alex

Exactly.

一点没错。

[42:26] Claire

for joining How I AI.

……来上 How I AI。

[42:28] Alex

Thanks so much.

太感谢了。

[42:30] Claire

Thanks so much for watching. If you enjoyed the 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.

非常感谢收看。如果你喜欢这期节目，请在 YouTube 上点个赞、订阅一下，或者更好的是，留言告诉我们你的想法。你也可以在 Apple Podcasts、Spotify 或者你常用的播客 App 里找到这档播客。也欢迎给我们打个分、写句评价，这能帮更多人发现这档节目。所有往期节目和节目的更多信息，都可以在 howiaipod.com 上看到。下期见。