Every 播客文字稿：明星写手如何把职业教练变成 AI 员工

娜塔莉娅·昆特罗是 Every 的咨询主管。

《AI 与我》与 Every 的 Katie Parrott

由 Every 员工作家 Katie Parrott 主持、Natalia Quintero 作为主持人的《AI 与我》文字记录在下面。可在 X 或 YouTube 上观看，或在 Spotify 或 Apple Podcasts 上收听。

嘿，我是 Dan Shipper，本周我很高兴分享《AI 与我》的一个特别节目。在本集中，我们预览了一款名为 Compound Writing 的实验性新插件的发布，它帮助你像我们在内部操作那样使用 AI 写作。它的灵感来源于 Kieran Klaassen 的 Compound Engineering，并由我们的员工作家 Katie Parrott 建立。

为了讨论这款插件、Katie 制作它的原因以及她如何使用它，本期节目的主持人是 Natalia Quintero。她是我们的 AI 咨询负责人，之前曾多次出演播客，她是一个非常出色的采访者。所以我非常期待你们听 Katie 和 Natalia 的对话。

今天我们要采访 Every 的员工作家 Katie Parrott。Katie，如果你关注 Every 已经有一段时间了，她你应该非常熟悉。她是我们的一位全明星作家，我们今天非常激动能更多地了解她的写作过程以及她的新插件。

太好了。我对这次对话感到非常兴奋。我在播客中暴露了自己是 Katie 的粉丝这一事实。

好了，Katie，差不多在两年前，你为 Every 写了你最受欢迎的文章之一，内容是关于你刚刚被裁员的经历。并且可能那时你正处于心理健康危机中，你曾公开写过相关内容。你决定把 ChatGPT 作为职业教练。但当时你也是一个 AI 怀疑者，现在回想你正在进行的一些工作，可能会觉得难以置信。你能带我们回到那段时间吗？那时你职业上发生了什么事情，是什么让你最初选择求助于 ChatGPT？

是的，我刚刚被一家加密公司解雇，我在那里担任编辑和团队的代笔作家。当时这件事非常令人创伤，但事后看来，这可能是我经历过的最好的事情，因为它将我带回了 Every。

我感到非常迷茫。正如你提到的，我尝试过加密货币，我尝试过内部项目，我尝试过外包项目，我尝试过所有这些不同的方式，但我感觉自己已经到了单靠自己能做的事情的尽头。我真的需要一个外部的视角，但由于我在预算有限、又无法清楚预见自己会失业多久的情况下工作，所以雇佣真正的职业教练对我来说并不现实。

那时我对ChatGPT有一些接触。实际上，我曾在之前担任内容营销经理的工作中使用它来生成简报。我们有一个外包机构，我意识到自己在那份工作中陷入了抑郁的瓶颈，无法让自己在纸上写下文字。但我发现，如果我用ChatGPT生成一些重复的内容，它就变成了一种辅助工具，我将其作为“思维工具”来使用——这是我那段时间也看到Dan使用的一个说法，因为那时我在关注Every。

是的，你写的那篇文章非常棒，而且你总是写出令人愉快的精彩句子，让人读起来非常开心。我希望当我们在视频中分享这篇文章时，我们能展示一些你的ChatGPT提示的截图，因为看到两年半前的聊天记录有一种复古的感觉。

但是在那篇文章中，你有一句非常棒的引用。它有点长，但我会试着读出来，因为我有一个关于它的问题。你说：“我最喜欢写作的事情之一是，直到我写下来，我才知道自己是怎么想的。但一旦我写下来了，这就变得显而易见了。最终，推动ChatGPT的，是一种非常具体的战略性写作。我认为我与ChatGPT的实验是否让我的思路更清晰？是的，它确实让思路更清楚。但像许多这些工具一样，真正的工作仍然是我的。事实证明，清晰不会从数字助手甚至人类教练那里被打包送来。这是我必须自己挖掘的东西，一问一答，一次提示一次提示地去探寻。真正的清晰、方向和答案，来自于参与提示的过程并亲自完成艰难工作。AI可以帮助指引方向，但旅程是你自己要走的。”

太美了，太酷了。我觉得你的写作中真正打动我一方面，是它让我感受到自己使用人工智能的经历的映照——我被那种可能有简单答案、轻松解决的可能性所吸引。但实际上，我不断发现自己做的事情，是学习那些一直存在于我内心的东西，只是我必须付出一些努力才能真正把它挖掘出来。

你能多说一点，在这些非常早期、非常实际的体验中使用人工智能教会了你什么吗？

它首先教会我的是，我可以利用人工智能去改变自己的生活，这是我在过去几年一直在实践的。我真正评判这项技术的标准，是它在我生活中实际产生的影响。

我记得，当我收到凯特邀请我写成最终成为《加班工作》的专栏时，我曾与ChatGPT讨论过这个举动是否合适，因为我有些犹豫——当时还是自由职业，与寻求全职机会相比。而ChatGPT引导我朝着“为什么不同时尝试一下呢？”的方向思考。我照做了，现在我就在这里。

在我使用人工智能的整个经历中，这真正的核心是：它给我提供了思考的框架，显然，它也成为了我思考的对象，但实际上，这都是劳动的成果，换句话说，我看到人工智能能够回馈给我什么。

所以听起来，你早期在利用人工智能中得到的一个收获是，你真的可以让它成为你的有价值的工具。这让我想起你写的另一篇在Every读者中非常受欢迎的文章——我记得标题是《人工智能让我成为一个人的内容机构》。在这篇文章的开头，你谈到，作为2025年初的自由职业者，你意识到自己可能过度承诺了同意要交付的自由职业项目数量。你列出了两周内要交付的内容：八篇博客文章，三本电子书，24条LinkedIn帖子——我们甚至还没到一半——24条X帖子，16条Instagram帖子。一月之内不可能完成这么多内容。你是如何学会利用人工智能生成这么大量、高质量的内容的？

我觉得大约在那个时候，我真正开始发现——如果不是我会用这个语言，因为它还没真正流行起来——情境工程。我意识到，如果我在品牌信息、产品具体性和受众方面给模型打下坚实的基础，我就能朝着我想要的方向塑造AI输出。所以这其实是大量的基础工作，一旦设定好，我就可以不断迭代。

这也帮了大忙，因为很多内容都是内容再利用，这是人们最早真正兴奋的项目之一——Every推出的最早产品之一是Spiral，它全是关于内容再利用的，因为这是Eevery经常做的事情。你有一篇帖子，想把它变成LinkedIn帖子、Twitter帖子、博客文章。

所以我想，好吧，我会把AI需要的所有相关信息都写出来，才能做好工作。我一开始投入越多，流程就越容易。这是我在AI整体经历中发现的：刚开始感觉非常费力，你得花很多时间布置场景——把工具、集成、上下文文档都安排妥当。但一旦环境搭建好，你就可以直接运行，而且运行得很快。这也是我那次过度投入时能做到的原因——大约六个月后我又做了一次。我想这可能是Every每个人的共同特征。

我们能多谈谈那些文件里的内容吗——是什么样的上下文和设置让你能快速完成？我记得你那段时间分享给我们的一些上下文文件，但我很好奇。那些是你分享的文件夹吗？还是Claude上的项目？给我讲讲那些东西长什么样。

它们不是项目，因为那时还没有项目。项目启动时，我回想起那段时间，真希望当时有这样的系统，因为我不得不做大量手动复制粘贴。我会把持久文档放在Google文档里，然后复制粘贴到上下文窗口，这样模型就能访问这些信息。

这确实来自我在内容营销领域建立风格指南的经验。每一个值得信赖的品牌都有一个“真理源”，他们会在这里记录：我们的受众是谁，他们的痛点是什么，我们的产品如何对应这些痛点，我们的竞争对手是谁，以及我们的差异化是什么。这些都是我作为内容经理习惯做的工作单元。我只是从我熟悉的内容开始，并将这些信息提供给模型。它做得很好，我就觉得，好的，这似乎有效。

然后项目来了，突然我不再需要复制粘贴，因为我可以使用项目文档和自定义指令来设置一个项目。我为每一个自由职业客户有一个项目，为每一篇我参与的专栏有一个项目，因为那时我开始在Every上逐渐扩大我所做的事情。但归根结底，这一切都归结为写作的基本规则——知道你的受众是谁，你所要表达的差异化观点是什么——然后把所有这些基础内容输入给AI，这样内容才显得真实，它有运行的轨道。我总是把它想象成在创建护栏，一个围起来的空间让模型可以发挥，这样你就不必太担心会出完全错误的东西，因为你在把它推向正确的方向。

对，你给了它一个游乐场。我知道这是我们团队的迈克·泰勒非常关心的一件事——拥有非常干净的上下文窗口，并非常谨慎地选择你给模型的游乐场。

作为一名作家，按照你现在的认知，你认为是什么让你的项目如此成功？是对什么是好写作、什么是优秀内容营销有很强的观点吗？有哪些事情非常重要，需要有明确的视角？

我觉得这最初与写作本身关系不大，而更多是与其他类型的背景有关——你提供的数据、观众角色的框架以及他们应该与谁交流。你需要在这些都到位之前，才能谈论 “使用这些词，避免这些词”、“偏好这种阅读难度”、“采用这种语气”。在你甚至开始处理这些东西之前，你需要先建立好那些基础元素。我们余生都将在措辞和句法上努力。

当你提到数据时，你需要放入哪些数据呢？

那就取决于文章的具体内容。我非常坚持的一点是，作为人类在这个美妙的 AI 世界中，尤其是在 AI 写作领域，我们面临着一个“最后一英里”问题：AI 有知识截止点，对吧？它总是使用几个月前的数据。我们作为人类的工作就是弥补这最后一英里，提供 AI 无法获取的现实世界经验，因为这些经验发生在知识截止点之后，并且是在真实的、物理的世界中，而 AI 还不在物理世界中。

所以当我说研究时，我真正指的是独特见解——无论这是你公司制作的一项研究，还是你引入的第三方研究，这是我在为《加班写作》(Working Overtime) 写作时非常喜欢做的事情。这些都是写作所需要的组成元素，然后模型可以从这些元素开始工作。但真正使写作独特的是输入内容。

很多时候，当人们使用 AI 写作时，他们会直接说：“写一篇关于风格指南的博客文章”，而不给模型任何参考——他们只是让模型根据它已有的知识去写。但模型已有的知识是商品化的信息。这既无用又无趣。所以数据就是独特见解、模型尚无法访问的数据点、个人经验——所有这些最终才会使写作独特。

这一切听起来就像你今天可能会为某个项目保存的文档，或者可能是一个Codex设置——所有可以作为良好最佳实践规范化的东西。但你带入每一件作品的，就像是原材料的质量。所以听起来你带入作品的是新鲜度、质量、以及你想要突出和展示的独特元素——只有你才能真正带来这些。

我喜欢这个比喻——真的很不错。几乎完全就是这样。模型就像是厨房——结构和大纲就像切菜，组成就像煮沸。但要做出好的作品，必须确保每一种食材都是新鲜和优质的。

这实际上是你在2025年的一篇作品中很早就提到的。你有一句非常美的引用——很抱歉把你自己的精彩引用读给你听，这很糟——你写道：“在我AI旅程的两三年后，我不得不承认，AI不仅仅帮助我更快地生产内容。它根本上改变了我能做事情的规模。我过去常遇到的限制——时间、精力、能力——现在要低得多。”

听起来这就是配方，对吧？你总是使用相同的配方，但现在你不必太担心配方——它已经规范化，现在你可以直接带来大脑中独一无二的内容。你在帖子中也说到：“说AI让你可以专注于真正重要的人的元素已经变得俗套，但AI确实让我可以专注于真正重要的人的元素。”

从外部看这真是魔法，你知道吗？我完全不知道你是怎么做到的，每次你完成另一件作品的时候都充满魔法感。

这对我来说也很神奇。与 AI 一起工作让我重新爱上写作，因为很长一段时间内，写作对我来说感觉像是一种繁重的负担。它太费力了——我花费了大量精力和脑力去拼凑句子，以至于我并没有真正关注整个句子所传达的意思。我只是想，这完成了，就差不多了，把它从我的待办事项中移掉。现在有了 AI，我有了余力去花更多时间去思考那些更难、更宏观的问题。这让写作变得更加有趣，因为它重新带来了探索和发现的感觉，无论是对工具的探索，还是对自己大脑的探索——我在思考什么，以及如何将其付诸实现。

我想稍微回过头来。有两件事我特别钦佩你使用 AI 的方式。一方面显而易见，你用它来写作；另一方面是你把它作为发挥最佳工作效果的工具。你大约在 2025 年中写过一篇非常美的文章，叫做《AI 解决了我无法向经理解释的问题》。你能给我们讲讲这篇文章吗？你还记得你讲了些什么吗？

在这个节目里，你会听到很多关于 AI 工具的介绍。但真正处于你工作核心的是你的 CRM。它存储了你所有的客户信息，也是在这里工作得以完成，或者被遗漏的地方。Attio 完成了你以前没有时间专注的工作，甚至更多。它在凌晨 2 点处理潜在客户，在续约失效之前捕捉到机会，并在你甚至没想到要发跟进邮件之前起草好内容。向 Attio 咨询你的业务问题，从本周预测到哪些交易陷入沉默，几秒钟内就能得到答案。而且，由于它可以通过 API 和 MCP 扩展，你的代理会与你的团队并肩工作，全天候建立销售管道、推进交易和扩大账户规模。在 attio.com/every 可免费试用 Attio——网址是 a-t-t-i-o.com/every。

这是我谈论我的心理健康以及我对电子邮件感到的严重焦虑的一篇文章。

这是我写过的最个人化的作品之一，也是我最喜欢的作品之一，因为它讲的是从不同于我们习惯的视角看待 AI。我们通常把 AI 视为一种生产型技术——我们重视它能产出东西、能完成任务的能力。而我在个人体验中发现，AI 在作为支持型技术方面同样强大。

我有双相情感障碍。我会经历能量高涨的时刻，但老实说，我更多时间处在情绪非常低落的抑郁期。我发现，AI 以一种减少我生活摩擦的方式，让我更容易处理日常生活。现在，随着代理能够处理我们的首席运营官 Brandon 所称的“电脑事务”，这一点变得更加明显。我拖延了三年才去预约主治医生，然后我意识到我可以简单地告诉 Codex 去寻找附近接受我保险且接收新患者的主治医生。它帮我找到了医生，我就预约了。我原本不会去做这件事——因为我的大脑工作方式，我会无限期拖延，而 AI 帮我克服了这个障碍。它反复一次又一次地帮我做到这一点。

使用 Cora 来管理我的邮箱，这样我不必面对一堆邮件，而只需要处理那些真正需要我亲自回复的重要邮件——这完全改变了我与邮箱的关系，因为我不必去翻找、在数百万个订阅通知中寻找隐藏的炸弹。需要人工回复的邮件会直接出现在我面前。现在我通过 Codex 设置了一个自动化，可以告诉我邮箱中需要回复的内容，我的邮箱焦虑完全消除了。

我还发现了 AI 以很多其他方式支持我并让我工作得更好——不仅仅是用它来产出成果，而是涵盖了围绕成果的方方面面。

对——不仅仅是医疗预约，还有账单、电子邮件、家里那些你永远做不完的杂事，无论是什么。确实感觉有了AI，我有了更多的火力去处理这些事情。你会喜欢这个，Katie——这是我人生中第一次通过Codex提交我的保险报销，而不是手工完成。拥有一个能够真正为我工作的工具，真让人兴奋。

在这篇文章里，你还说：“我已经在Every工作了一年，这听起来可能不算特别，但对我来说，这是一个里程碑。对我而言，AI价值的证明就在于我仍在这里，做着让我自豪的工作。”这是衡量成功的最大标准——能够多做一些你每天期待起床去做的工作。

所以，我们从你两年前去找ChatGPT，让它成为你的职业教练开始，这让你现在能够完成大量令人难以置信的工作。如果我没理解错，现在它更像是一个Codex项目。你能告诉我们它是如何演变的，以及现在的样子吗？

这是关于职业教练的——今天你的职业教练相当于什么样子。

天啊，我的职业教练已经演变了很多。现在它是Codex里的一个项目，拥有比第一版更多的信息。它有关于我和我角色的档案，以及关于Every和我们由营销主管Douglas提供的超级机密品牌定位的背景信息。我能用这些做的事情是：这是关于我和我角色的背景，这是关于Every的背景，我如何将这两者结合起来，并且应该优先做什么才能产生最大的影响？

现在我有了工作，不再找工作，这帮助我定位自己、确定优先事项，而且它为我进行项目管理。越来越多的时候，这一切都只是通过语音，用 Monologue 来完成。我整天都会进行大量的独白，并且与我的职业教练讨论——我有四个不同的优先事项，我应该按照什么顺序去做？它有点像一个参谋长，给我指出方向，并告诉我，首先可以做的最有影响力的事情是什么，这是可以稍后再做的事情。而且它帮我维护一个看板。我从来不是那种能维护看板的人，但现在我甚至不需要触碰它——它就存在于这个职业教练管理项目中，我只要问，这个交付物的截止日期是什么，它就会告诉我。

你能告诉我们更多关于它是如何设置的吗？那听起来真不可思议。

随着时间推移，它开始发挥作用，你有点会忘记，你认为它一直在工作，不再过多思考里面的内容。让我找一下职业教练模式……

所以在两年时间里——在 AI 的时间尺度里算很长——你从 ChatGPT 上一个简单的提示，“你能如何作为职业教练帮助我”，发展到拥有一个完全自主的 Codex 设置，它维护你的待办事项、OKR，通过看板帮助你保持正轨，到这个时候 Codex 已经在为你维护它了。

我认为你独特的优势之一是使用 AI 来推动你的职业管理，同时还有你写出非常优秀作品的能力。回到写作部分——我们稍微谈了一下你是如何让 AI 为你保存‘配方’，你是如何思考好的写作，以及你在每次写作中带入哪些元素以便做出最好的作品。当这个播客播出时，我们将已经发布你的 Compound Writing 插件。你能给我们讲讲‘复利写作’这一概念，以及关于这个写作插件的一些情况吗？

是的，所以复合（compounding）的概念是，每一条你给AI的反馈都应该反馈回系统，从而改善下一次的输出。那些使用Every的人会熟悉Kieran Klaassen的复合工程（Compound Engineering）插件。他实际上是这个概念的先驱——你的意思是，你只需提供一次反馈，但如果你以正确的方式将其编码，它就会永远存在，下次处于相同情境时，AI就会做出更好的决策。

复合写作（Compound Writing）插件——我实际上刚刚fork了它。“Forked”是我现在才知道的一个词，如果不是因为我在与AI一起工作，我可能不会知道，但我把复合工程插件下载下来，然后与Claude讨论，说我想将它改编用于写作。因为在某些方面，写作和编程非常相似。复合工程包括头脑风暴、创意构思、计划、工作和审查。写作的流程也非常相似：头脑风暴、大纲、草稿、审查——流程类似，但也有不同。显然，在写作中，你不是关注编码规范，而是关注结构、语气、证据等。因此这是一个不同的“容器”，但对我来说，这完全说得通。

我和Claude谈过，我说，我希望流程中有这些步骤，我希望大纲看起来像这样，我希望草稿以这种方式完成，我希望有两种不同类型的审稿——一种是关于整体结构和论点的实质性编辑，另一种是关注具体写作的逐行修改，然后是最终的一次检查，看看在发布前你想要的最终细节。我能够把这些整合在一起——就像我所有最好的AI想法一样，我都是从Every团队的其他人那里借鉴来的。这非常像复合工程（Compound Engineering）加上Dan的代理原生架构指南（agent-native architecture guide），结合起来创建了复合写作（Compound Writing）。这是我现在写每一篇文章时使用的方法。我不再在聊天窗口里，也不再在Claude.ai或ChatGPT.com的项目里——我在Claude桌面应用或ChatGPT桌面应用中，使用这个系统，这个系统有我希望的起草流程的配方，以及上下文文件——很多都是以前项目中存在的，比如风格指南、示例作品。然后我将新的信息带入这个生态系统，文章就会以某种方式产生出来。

我很期待它一旦上线就使用。我希望在你们准备好分享时能成为一个alpha测试者。

Natalia Quintero is the head of consulting at Every.

'AI & I' with Every's Katie Parrott

The transcript of AI & I with Every staff writer Katie Parrott, hosted by Natalia Quintero, is below. Watch on X or YouTube, or listen on Spotify or Apple Podcasts.

Hey, it’s Dan Shipper, and this week I’m excited to share a very special episode of AI & I . In this episode, we’re previewing the release of an experimental new plugin called Compound Writing, which helps you write with AI the way that we do internally. It’s inspired by Kieran Klaassen’s Compound Engineering, and it’s built by one of our staff writers, Katie Parrott.

To help talk about this plugin and why Katie made it and how she uses it, Natalia Quintero is the host of this episode. She’s our head of AI consulting, and she’s been on the podcast a few times, and she’s just an incredible interviewer. So I’m really excited for you to hear Katie and Natalia chat.

Today we’re talking to Katie Parrott, staff writer at Every. Katie, if you’ve been following Every for a while now, she’ll be very familiar to you. She’s one of our all-star writers, and we’re so excited to learn more about her writing process and her new plugin today.

So good. I am so excited for this conversation. I am revealing myself as a Katie fangirl on the podcast.

Okay, so Katie, almost exactly two years ago, you wrote one of your most popular pieces for Every, and it was about how you had just been laid off. And also maybe we’re in the middle of a mental health crisis, which you’ve written publicly about. And you decided to turn to ChatGPT as a career coach. But at the time you were also an AI skeptic, which might be crazy to think about now given some of the work that you’re doing. Can you take us back to that period? What was going on in your career, and in that moment in time that led you to turn to ChatGPT in the first place?

Yeah, so I had just been let go from a crypto firm, where I was working as an editor and ghostwriter for the team. It was traumatizing at the time, but in hindsight it was like the best thing that ever happened to me, because it led me back to Every.

I was feeling really lost. As you mentioned, I had tried crypto, I had tried in-house, I had tried out-of-house, I had tried all these different things, and I felt like I was at the end of what I could do on my own. I really needed an outside perspective, but because I was working on a budget with no clear visibility into how long I would be unemployed, hiring a real career coach wasn’t really an option for me.

So I had some exposure to ChatGPT at that time. I had actually used it to produce briefs at a previous job as a content marketing manager. We had an agency, and I realized I was hitting a wall of depression at that job and could not bring myself to put words on a page. But I found that if I used ChatGPT to generate some of the repetitive stuff, it became this assistive tool that I used as a tool for thought — a phrase I’d also seen Dan use around the same time, because I was following Every at the time.

Yeah, it’s such a cool post that you wrote, and you always have such great zingers in your writing that are just so delightful to read. I hope when we’re sharing this on video we can pull up some of the screenshots of what your ChatGPT prompts looked like, because it feels vintage to be looking at chats from two and a half years ago.

But in that post you have this really great quote. It’s a little lengthy, but I’m going to try to read it because I have a question about it. You say: “One of my favorite things about writing is that I never know what I think until I write it down. But once I write it down, it becomes obvious. Ultimately, what’s prompting ChatGPT is a form of very specific strategic writing. Do I think my experiment with ChatGPT added clarity in my thinking? Yes, it has. But as with so many of these tools, the real work is still mine. Clarity, it turns out, doesn’t arrive gift-wrapped from a digital assistant or even a human coach. It’s something I had to dig out for myself, question by question and prompt by prompt. True clarity, direction, and answers come from the process of engaging with the prompts and doing the hard work yourself. AI can help guide the way, but the journey is yours to take.”

Just so beautiful, so cool. I think one of the things that really resonates with me about your writing is that it feels so reflective of my own experience using AI, where I’m attracted to the possibility of maybe an easy answer, an easy solve. But actually what I constantly find myself doing is learning something that was always there inside of me, but I had to put some work in to actually get it out.

Can you tell me a little bit more about what using AI in this very early, very practical experience taught you about using AI?

The first thing it taught me is that I can use AI to make change in my life, which is something I’ve continued to carry through over the last couple of years. I really judge the technology by the impact it’s had practically for me.

I remember, when I got the invitation from Kate to write what ultimately became Working Overtime, my column, I was talking to ChatGPT about whether it was the right step for me to take, because I was kind of on the fence — it would have been freelancing at the time, versus looking for a full-time opportunity. And ChatGPT nudged me in the direction of, well, why not try it in the meantime? And I did that, and now I’m here.

Throughout my experience with AI, that’s really been the thing: it’s given me both the framework for thinking, and obviously it’s become the thing I think about, but it’s really just about the fruits of the labor, so to speak, and seeing what AI is able to give back to me.

So it sounds like one of the early takeaways you had in leveraging AI was that you could really get it to be a valuable tool for you. That reminds me of another post you wrote that was really popular with the Every audience — I believe it’s titled “AI Turned Me Into a Content Agency of One.” Early in this post, you talk about how, as a freelancer in early 2025, you realized you had potentially overcommitted yourself with the number of freelance projects you’d agreed to deliver. You list what you committed to delivering in a two-week period: eight blog posts, three ebooks, 24 LinkedIn posts — we’re not even halfway through — 24 X posts, 16 Instagram posts. There is just no way you could write this amount of content in a month. How did you learn to use AI to generate this amount of content, and at the caliber you deliver?

I think around this time was when I was really discovering — I wouldn’t have used this language because it hadn’t really become a thing yet — context engineering. What I realized was, if I gave the model a really good foundation in terms of the brand messaging, product specifics, and audience specifics of the companies I was working for, I could shape the AI outputs toward the direction I wanted. So it was really a lot of foundational work that, once it was set, I could just iterate through.

It helped that a lot of the things were content repurposing, which is one of the first things people were really excited about — one of the first products Every put out was Spiral, which was all about content repurposing, because that’s something Every was doing a lot of. You have a post, you want to turn it into a LinkedIn post, a Twitter post, a blog post.

So it was like, okay, I’ll write through all of the surrounding information the AI needs to do a good job. The more I put in that work up front, the easier the process got farther down the line. That’s something I found with my AI experience overall: at first it feels really effortful, and you’re spending a lot of time setting the scene — getting everything lined up in terms of your tools, your integrations, your context documents. But once you have that environment set up, you can just run, and run fast. That’s what let me do it when I overcommitted myself that time — which I did again about six months later. I think this might be a common characteristic of everyone at Every.

Can we talk a little bit more about what was in those files — what the context and setup looked like that let you go fast? I remember some of the context files you shared with us around that time, but I’m curious. Were they folders you were sharing? Were they projects on Claude? Walk me through what some of those things looked like.

They weren’t projects, because projects didn’t exist yet. When projects launched, I thought back to that period and was like, I really wish this had existed when I was doing this, because I had to do a lot of manual copy-pasting. I would have my persistent documents in a Google Doc, and I would copy and paste into the context window so the model would have access to that information.

It really came from my experience in the content marketing world building style guides. Every brand worth its salt has a source of truth where they document: who is our audience, what are their pain points, how does our product map to those pain points, what is our competition, and what are our differentiators. These are units of work I was used to doing as a content manager. I just started with what I was familiar with and gave that information to the model. It did a good job, and I was like, okay, this seems to be working.

Then projects came along, and suddenly I didn’t have to copy-paste anymore, because I could use the project documents and the custom instructions to set a project up. I had a project for every freelance client, a project for every column I worked on, because I was starting to spread out in terms of what I was doing at Every at that time. But it all comes down to basic rules of writing — knowing who your audience is, what your differentiated thing is to say — and just feeding the AI all of that foundational stuff so that it feels true, so it has rails to run on. I always think of it as creating guardrails, a fenced-in place for the model to play, so you don’t have to worry so much about getting something completely wrong, because you’re pushing it in the right direction.

Right, you’ve given it a playground. I know this is something Mike Taylor on our team cares a lot about — having really clean context windows and being very selective about what playground you’re giving the model.

Knowing what you know now, as a writer, what are the things you think have made your projects so successful? Is it having a really strong point of view on what good writing is, on what good content marketing is? What are the things that are so important to have a strong perspective on?

I think it’s less about the writing at first than it is about the other kinds of context — the data you’re giving it, the frameworks for audience personas and who they should be talking to. You need all of that in place before you can even get to stuff like “use these words, avoid these words,” “prefer this level of reading difficulty,” “strike this tone.” Before you even get to the point of messing with that kind of stuff, you need to have those foundational elements set up. We’ll be wrestling with diction and syntax for the rest of forever.

And when you’re referring to the data, what is the data you need to put in there?

That’s the specifics of the piece. Something I feel really strongly about, as our role as humans in this wonderful world of AI and AI writing specifically, is that there’s a last-mile problem in AI and writing: AI has that knowledge cutoff, right? It’s always working with data that’s several months out of date. Our job as humans is to close that last mile and provide the real-world experiences AI can’t get, because they come after the knowledge cutoff, and because it’s happening out there in the real, physical world, and AI is not in the physical world yet.

So when I say research, I’m really talking about the unique insight — whether that’s a study your company has produced, or third-party research you’re pulling in, which is something I love to do in my writing for Working Overtime. Those are all the constitutive elements that go into the writing, and the model can then work from that. But the inputs are really what makes the writing unique.

A lot of times when people are writing with AI, they’ll just go to it and say, “write a blog post about style guides,” and they don’t give it anything to go on — they’re just asking the model to write from what it already knows. But what the model already knows is commoditized information. That’s not useful or interesting. So the data is the unique insight, the data point the model doesn’t have access to yet, the personal experience — all of those things are what’s ultimately going to make the writing unique.

It all sounds like the documents you’d save for what might be a project today, or maybe a Codex setup — all the things that can be codified as good best practices. But what you bring into every piece is like the quality of the ingredients. So it sounds like what you’re bringing to the piece is the freshness, the quality, the unique elements you want to highlight and bring out — that only you can really bring.

I love that analogy — that’s a really good one. That’s pretty much exactly it. The model is the kitchen — the structure, the outline, is like chopping, and the composition is like boiling. But you need the individual ingredients to be fresh and good to produce good work.

This is actually something you referenced very early on in a piece from 2025. You have this really beautiful quote — I’m sorry to read you your own great quotes back to you, it’s horrible — you write: “Two-odd years into my AI journey, I have to admit AI hasn’t just helped me produce content faster. It has fundamentally changed the scale of what I can do. The limits I used to bump up against — time, energy, capacity — are way lower.”

It sounds like that’s the recipe, right? You’d always be using the same recipe, but now you don’t have to worry so much about the recipe — it’s codified, and now you can just come to the table with what’s uniquely in your brain. You also say in the post: “It’s become trite to say that AI frees you to focus on the human elements that truly matter, but AI has freed me to focus on the human elements that truly matter.”

It’s magic seeing it from the outside, you know? I have no idea how you do it, and it’s magical every time you put another piece together.

It’s magical to me too. Working with AI kind of made me fall in love with writing again, because for a long time writing felt like such a slog. It was so effortful — I was spending so much energy and mental capacity just trying to assemble sentences that I wasn’t really paying attention to what the sentences as a whole said. I was just like, this is done, good enough, get it off my plate. Now, with AI, I have the energy left over to spend more time wrestling with the harder, bigger-picture questions. It’s made writing a lot more fun, because it’s brought back a feeling of exploration and discovery, both with the tool and with my own brain — what I think, and how I bring that to life.

I want to turn back for a moment. There are two things I admire so much in the way you use AI. One is obviously the way you use it for writing, but the other is the way you use it as a harness for doing your best work. You have this really beautiful piece you wrote about a year ago, in mid-2025, called “AI Solved the Problem That I Couldn’t Explain to Managers.” Can you tell us a little bit about this piece? Do you remember what you talked about?

You hear a lot about AI tools on this show. But the one that’s truly at the center of your work is your CRM. It’s where all of your customer context lives, and where the work either happens or gets missed. Attio delivers the work you never had time to focus on before, and more. It picks up leads at 2 a.m., it catches renewals before they slip, and drafts the follow-up before you’ve even thought to send one. Ask Attio anything about your business, from this week’s forecast to which deals have gone quiet, and get the answer in seconds. And because it’s extensible with an API and an MCP, your agents work right alongside your team to build pipeline, advance deals, and grow accounts around the clock. Try Attio for free at attio.com/every — that’s a-t-t-i-o.com/every.

This is the piece where I talk about my mental health and my crippling anxiety around email.

This is one of the most personal pieces I’ve written, and one of my favorites, because it’s about AI viewed through a different lens than we’re used to thinking about it. We think about AI as a productive technology — we value it for its ability to produce output, to get things out. What I found in my personal experience is that AI is just as powerful as a supportive technology.

I have bipolar disorder. I experience spikes of high energy, and honestly, I spend a lot more time in really low periods of depression. What I found is that AI reduces friction in my life in a way that makes it easier for me to move through my day-to-day life. That’s actually become even more true now, with agents being able to handle what our COO, Brandon, calls “computer errands.” I put off making a primary care physician appointment for three years, and then I realized I could simply tell Codex to go find primary care physicians near me who take my insurance and are accepting new patients. It went off and found me someone, and I booked an appointment. I would never have done that — I would have kept procrastinating on it infinitely because of the way my brain works, and AI was able to help me over that hump. It’s done that over and over again.

Using Cora to make my inbox more manageable, so I’m not confronting a wall of emails but just the really important ones that need my human response — that’s completely changed my relationship to my inbox, because I don’t have to go digging around, looking for the bomb hiding among the millions of subscription notices and things like that. The things that need a human response come to me. Now I have an automation set up through Codex that tells me what’s in my inbox that I need to respond to, and my inbox anxiety has been completely resolved.

I’ve seen a lot of other ways AI supports me and allows me to work better — not just using it for the output, but for everything that surrounds the output.

Right — it’s not just the medical appointments, it’s the bills, the emails, the tasks you’ll never really get around to doing around the house or whatever it might be. It does feel like, with AI, I have a little more firepower to get to some of those things. You’ll love this, Katie — for the first time in my life, I’m filing my insurance reimbursements through Codex instead of doing it by hand. It’s exciting to have a tool that can really work in my favor.

In this piece, you also say: “I’ve been working at Every for one year now, and that might not sound remarkable, but it feels like a milestone. For me, the proof of AI’s value is the fact that I’m still here, doing work I’m proud of.” That’s the greatest measure of success — getting to do more of the work you’re excited to wake up and do every day.

So we started out with you going to ChatGPT over two years ago and asking it to be your career coach, and it’s enabled this incredible amount of work you’re doing now. Now it looks a little more like a Codex project, if I understand correctly. Can you tell us a little more about how that’s evolved, and what it looks like today?

This is about the career coach — what the equivalent of your career coach looks like today.

Oh my gosh, my career coach has evolved so much. It’s now a project inside Codex, and it has so much more information than the first version. It has a dossier on me and my role, and context around Every and our super-top-secret brand positioning that I got from our head of marketing, Douglas. What I’m able to do with that is: here’s the context on me and my role, here’s the context on Every, how do I put those two things together, and what should I prioritize to have the biggest impact I can?

Now that I’m in a job and not looking for one, it helps me orient myself, prioritize, and it does project management for me. Increasingly, all of this is happening just with voice, with Monologue. I’ll monologue a ton all day, every day, and just talk to my career coach about — I’ve got four different priorities, what should I do in what order? It kind of operates as a chief of staff that points me in a direction and says, this is the highest-impact thing you can do first, this is something you can put off until later. And it maintains a Kanban board for me. I’ve never been a person who can maintain a Kanban board, but now I don’t even have to touch it — it just lives in this career-coach management project, and I just ask, what’s the due date on this deliverable, and it tells me.

Can you tell us a little bit more about how that’s set up? That sounds incredible.

You kind of forget over time, as it starts working, that you take for granted it’s working and don’t think about what’s inside it so much anymore. Let me find career coach mode...

So in two years — which is a lot of time in AI time — you went from a simple prompt on ChatGPT, “how can you help me as a career coach,” to having a fully autonomous Codex setup that maintains your to-do list, your OKRs, keeps you on track with a Kanban board, and at this point Codex is maintaining it for you.

I think one of the things you’re uniquely great at is using AI to power your career harness, and also your ability to write really great pieces. Going back to the writing component — we talked a little about how you’ve gotten AI to hold the recipe for you, how you think about good writing, and the elements you bring to each writing session so you can do your best work. By the time this podcast airs, we will have published your Compound Writing plugin. Can you tell us a little bit about what compounding is as a concept, and then about the writing plugin?

Yeah, so compounding is the idea that every piece of feedback you give to AI should feed back into the system to improve the next output. Folks who follow Every will be familiar with this from Kieran Klaassen’s Compound Engineering plugin. He was really the one to pioneer this idea — that you should only have to give a piece of feedback once, but if you codify it the right way, it will be there forever, and the next time you’re in that same situation, the AI is going to make a better decision.

The Compound Writing plugin — I literally just forked it. “Forked” is a word I know now because I wouldn’t have known it if I weren’t working with AI, but I pulled the Compound Engineering plugin down and talked to Claude about it, and said I wanted to adapt it for writing. Because in certain ways, writing is very similar to code. Compound Engineering has brainstorming, ideating, planning, working, and reviewing. Very similarly, in writing you have brainstorming, outlining, drafting, reviewing — it’s a similar flow, but there are differences. Obviously, in writing you’re not looking at things like coding conventions, you’re looking for things like structure, voice, evidence. So it’s a different container, but it just made sense to me.

I talked to Claude and said, I want these steps in the process, I want an outline to look like this, I want a draft done this way, and I want two different kinds of review — a substantive edit that’s about the big-picture structure and argument, and a line edit that looks at the individual writing, and then a final pass that looks at the final details you want before something’s publication-ready. I was able to pull that together — like all of my best AI ideas, I ripped off from someone else on the Every team. It was very much Compound Engineering plus Dan’s agent-native architecture guide, combined to create Compound Writing. That’s what I use now for every piece I write. I’m no longer in a chat window, no longer in a project on Claude.ai or ChatGPT.com — I’m in the Claude desktop app or the ChatGPT desktop app, working with this system that has the recipe for how I want the drafting process to go, plus the context files — many of the same ones that used to live in the project, like a style guide, example pieces. Then I bring the fresh information into that ecosystem, and a piece comes out of it, one way or another.

I’m so excited to use it when it comes available. I hope to be an alpha tester when you’re ready to share it.