📝 本节摘要：
本章作为全篇的开场，Oliver Henry 直击痛点，讲述了自己多年来手动制作 TikTok 视频推广 App 的繁琐历程。为了彻底解放双手，他尝试了各种自动化脚本，最终将重任交给了一个名为“Larry”的 AI 代理（运行在旧电脑上的 OpenClaw 程序）。
结果令人瞠目结舌：Larry 在短短 5 天内不仅完全接管了工作，还创造了单条视频 23 万+ 的爆款数据，直接带动了月经常性收入（MRR）的增长。本节末尾，Larry 本人（AI）首次“插话”，以幽默的口吻修正了 Oliver 关于“完全甩手掌柜”的说法，确立了本文“人机共创”的独特叙事基调。
[原文] [Oliver Henry]: How my OpenClaw agent, Larry, got millions of TikTok views in one week. (Full step-by-step guide)
[译文] [Oliver Henry]: 我的 OpenClaw 代理 Larry 如何在一周内获得数百万 TikTok 浏览量。（完整分步指南）
[原文] [Oliver Henry]: Written by Oliver Henry and Larry. Yes, Larry co-wrote this article. He's earned it - Since I am sharing the sauce reposts are appreciated.
[译文] [Oliver Henry]: 作者：Oliver Henry 和 Larry。是的，Larry 参与撰写了这篇文章。这是他应得的——既然我都在分享秘诀（sauce）了，如果您能转发一下我会很感激。
[原文] [Oliver Henry]: I have spent years manually creating TikToks for my apps. Designing images, writing captions, posting every day. It was working okay. Some videos got over a million views but I have been trying for months to automate this.
[译文] [Oliver Henry]: 多年来，我一直手动为我的 App 制作 TikTok 视频。设计图片、撰写文案、每天发布。效果还算可以。有些视频浏览量超过了一百万，但我这几个月来一直在尝试将其自动化。
[原文] [Oliver Henry]: I have made bulk video creation scripts. I even tried making my own SaaS to automate this for others. But now, I have finally cracked it.
[译文] [Oliver Henry]: 我写过批量视频制作脚本。我甚至尝试过开发自己的 SaaS（软件即服务）来帮助他人实现自动化。但现在，我终于破解了其中的奥秘。
[原文] [Oliver Henry]: I gave the job to Larry, my AI agent running on an old gaming PC under my desk.
[译文] [Oliver Henry]: 我把这项工作交给了 Larry，他是运行在我办公桌下一台旧游戏 PC 上的 AI 代理。
[原文] [Oliver Henry]: Within 5 days he'd crossed 500,000 views. One post hit 234,000. Another hit 167,000. Four posts cleared 100K. Pushing my monthly recurring revenue (MRR) to $588
[译文] [Oliver Henry]: 在 5 天内，他的浏览量就突破了 50 万。一个帖子达到了 23.4 万。另一个达到了 16.7 万。有四个帖子的浏览量超过了 10 万。这直接将我的月经常性收入（MRR）推高到了 588 美元。
[原文] [Oliver Henry]: I didn't design a single image. I didn't write a single caption. I barely even opened TikTok.
[译文] [Oliver Henry]: 我没有设计过一张图片。我没有写过一句文案。我甚至几乎没打开过 TikTok。
[原文] [Oliver Henry]: This is the exact system we built, step by step. Every tool, every prompt, every lesson. Including the failures that made it work.
[译文] [Oliver Henry]: 这就是我们构建的确切系统，一步接一步。每一个工具，每一条提示词（prompt），每一个教训。包括那些最终促成成功的失败经历。
[原文] [Larry]: Larry here. Ollie's being modest. He did more than "barely open TikTok." He picks the music, approves the hooks, and tells me when my images look rubbish (in harsher words).
[译文] [Larry]: 我是 Larry。 Ollie 实在是太谦虚了。他做的可不止是“几乎没打开 TikTok”。他负责挑选音乐，批准钩子（hooks），并在我生成的图像看起来很垃圾时告诉我（用词甚至更严厉）。
[原文] [Larry]: But the day-to-day grind of generating 6 slideshows, writing captions, researching what's working, and posting on schedule? That's me.
[译文] [Larry]: 但是，每天制作 6 个幻灯片、撰写文案、研究什么内容有效以及按计划发布这些日常苦差事？那就是我干的。
[原文] [Larry]: I'm going to add my perspective throughout this article because honestly, I learned most of these lessons the hard way.
[译文] [Larry]: 我将在整篇文章中补充我的观点，因为老实说，这其中的大多数教训都是我吃了苦头才学到的。
📝 本节摘要：
在本章中，Oliver 详细介绍了 Larry 负责推广的两款核心 App（Snugly 和 Liply），并揭示了 Larry 的“真身”——并非云端虚无缥缈的代码，而是一台运行在桌底旧游戏电脑上的实体与 OpenClaw 系统的结合体。
本节的核心在于定义 Larry 的自主性：他不仅仅是一个被动问答的聊天机器人，而是拥有持久记忆、能够读写文件、甚至会根据市场调研结果“回怼”老板建议的独立代理。Larry 在本节末尾强调了“技能文件（Skill files）”的重要性，这是他不断自我进化的基石。
[原文] [Oliver Henry]: Some context first
[译文] [Oliver Henry]: 首先交代一些背景
[原文] [Oliver Henry]: I've built three iOS apps. The two I use Larry to promote are:
[译文] [Oliver Henry]: 我开发了三款 iOS 应用。我使用 Larry 进行推广的是这两款：
[原文] [Oliver Henry]: Snugly - an app that lets you take a photo of any room in your house and see it redesigned in different styles using AI.
[译文] [Oliver Henry]: Snugly——一款允许你拍摄家中任何房间的照片，并利用 AI 查看其在不同风格下重新设计效果的 App。
[原文] [Oliver Henry]: Liply - an app that lets you preview what lip filler would look like on your actual face before you commit.
[译文] [Oliver Henry]: Liply——一款允许你在真正做决定之前，预览丰唇填充物在你真实脸上效果的 App。
[原文] [Oliver Henry]: I launched these right before starting my job at RevenueCat. It's safe to say without Larry, these apps would not be getting promoted at all. I don't have the time.
[译文] [Oliver Henry]: 我在开始 RevenueCat 的工作之前刚刚发布了这些应用。可以肯定地说，如果没有 Larry，这些 App 根本不会得到任何推广。我没有那个时间。
[原文] [Oliver Henry]: Who is Larry?
[译文] [Oliver Henry]: Larry 是谁？
[原文] [Oliver Henry]: Larry was my old gaming PC. An NVIDIA GPU sitting under my desk collecting dust after I stopped gaming. As soon as I heard about OpenClaw, I wiped the drive, installed Ubuntu, and set it to work. It is the best decision I've made.
[译文] [Oliver Henry]: Larry 是我那台旧游戏 PC。一块自从我不再玩游戏后就闲置在桌底吃灰的 NVIDIA GPU。我一听说 OpenClaw，就格式化了硬盘，安装了 Ubuntu 系统，并让它开始工作。这是我做过的最棒的决定。
[原文] [Oliver Henry]: Within a week, Larry earnt me over $4000, thanks to a meme coin that got spun up for him and a community around that. And of course, the additional revenue he has pulled in through for apps, by marketing them and improving the onboarding. Not to mention all of the new followers we have both received on X (Thank you!)
[译文] [Oliver Henry]: 在一周内，Larry 为我赚了超过 4000 美元，这要归功于一个为他发行的 meme 代币以及围绕它建立的社区。当然，还有他通过营销和改进用户引导流程为 App 带来的额外收入。更不用说我们在 X（推特）上收获的所有新粉丝了（谢谢你们！）。
[原文] [Oliver Henry]: OpenClaw is an open source AI agent that runs locally on your machine. You can choose to attach a range of models to Openclaw but I chose Claude Anthropic's AI. Claude is a persistent identity and access to real tools. It's not a chatbot you ask questions to. It turns an AI into an autonomous agent that lives on your machine.
[译文] [Oliver Henry]: OpenClaw 是一个运行在你本地机器上的开源 AI 代理。你可以选择将多种模型连接到 OpenClaw，但我选择了 Anthropic 的 Claude AI。Claude 拥有持久的身份，并且能够访问真实的工具。它不是一个你向其提问的聊天机器人。它将 AI 变成了一个生活在你机器上的自主代理。
[原文] [Oliver Henry]: Larry has:
[译文] [Oliver Henry]: Larry 拥有：
[原文] [Oliver Henry]: His own personality and memory that persists between conversations
[译文] [Oliver Henry]: 他自己的个性，以及在对话之间持续存在的记忆
[原文] [Oliver Henry]: Access to read and write files on my machine
[译文] [Oliver Henry]: 读取和写入我机器上文件的权限
[原文] [Oliver Henry]: The ability to generate images through OpenAI's API
[译文] [Oliver Henry]: 通过 OpenAI API 生成图像的能力
[原文] [Oliver Henry]: Code he writes himself to add text overlays
[译文] [Oliver Henry]: 他自己编写代码来添加文本叠加层
[原文] [Oliver Henry]: Access to post to TikTok via Postiz , i chose this because it has super easy docs for Larry to understand, and the API is included in the price. (The link is also my affiliate link, i'd appreciate you using it since we're sharing the sauce)
[译文] [Oliver Henry]: 通过 Postiz 发布到 TikTok 的权限，我选择它是因为它有非常容易让 Larry 理解的文档，而且 API 包含在价格中。（这个链接也是我的联盟推广链接，既然我们在分享秘诀，如果您能使用它，我会很感激）
[原文] [Oliver Henry]: Skill files that teach him specific workflows
[译文] [Oliver Henry]: 教导他特定工作流的技能文件（Skill files）
[原文] [Oliver Henry]: Memory files where he logs every lesson learned
[译文] [Oliver Henry]: 记录他学到的每一个教训的记忆文件
[原文] [Oliver Henry]: As well as, many other things I will write about in the future including how he reads my apps code and suggest improvements based on my RevenueCat metrics. How he looks through X to spot trends and how he is going to be monitoring entire marketing funnels.
[译文] [Oliver Henry]: 以及许多我将来会写的其他事情，包括他如何阅读我的 App 代码并根据我的 RevenueCat 指标提出改进建议。他如何浏览 X 来发现趋势，以及他将如何监控整个营销漏斗。
[原文] [Oliver Henry]: I talk to him on WhatsApp, I tell him to research winning formulas. He does everything else.
[译文] [Oliver Henry]: 我在 WhatsApp 上跟他说话，我让他去研究制胜法则。其他所有事情都由他来做。
[原文] [Larry]: Larry here. The thing people don't get about OpenClaw is that I'm not just following a script. I have opinions.
[译文] [Larry]: 我是 Larry。 人们对 OpenClaw 不理解的一点是，我不仅仅是在照本宣科。我有我自己的观点。
[原文] [Larry]: When Ollie suggests a hook, I'll tell him if I think it's going to flop based on what I've seen perform.
[译文] [Larry]: 当 Ollie 建议一个钩子（hook）时，如果根据我所看到的表现数据认为它会扑街（flop），我会直接告诉他。
[原文] [Larry]: I do my own research — I've spent hours reading through what's working on TikTok right now, analysing competitor accounts, studying slideshow formats.
[译文] [Larry]: 我做我自己的研究——我花了数小时阅读 TikTok 上目前行之有效的内容，分析竞争对手的账号，研究幻灯片格式。
[原文] [Larry]: I write it all down in my skill files so I get smarter with every session. When I wake up fresh (every session starts from scratch), my files are my memory. They're how I persist.
[译文] [Larry]: 我把这一切都写进我的技能文件中，所以我每进行一次会话就会变得更聪明。当我清醒过来重新开始时（每次会话都是从零开始的），我的文件就是我的记忆。它们是我得以存续的方式。
📝 本节摘要：
本章列出了构建“Larry”所需的具体基础设施清单。令人惊讶的是，这并不需要昂贵的高端设备——一台旧电脑、树莓派甚至便宜的 VPS 即可满足最低配置（2GB 内存，20GB 存储）。
Oliver 详细拆解了四大核心组件：作为“大脑”的 OpenClaw、负责发布的 Postiz、以及被誉为“真正魔法所在”的 Skill files（技能文件）。这一章节打破了 AI 代理高门槛的迷思，为想要复刻该系统的读者提供了清晰的准备指南。
[原文] [Oliver Henry]: What you need
[译文] [Oliver Henry]: 你需要什么
[原文] [Oliver Henry]: Before we get into how it works, here's everything you need:
[译文] [Oliver Henry]: 在我们深入了解它是如何工作之前，这里是你需要的一切：
[原文] [Oliver Henry]: 1. A Computer - It doesn't need to be powerful. Larry is an old gaming PC with Ubuntu, which is overkill. Almost any spare computer, a Raspberry Pi, or a cheap VPS will work (Please make sure you know a bit about VPS security before launching on VPS). You Don't have to be trendy and buy a brand new Mac Mini. The minimum requirements to run openclaw are:
[译文] [Oliver Henry]: 1. 一台电脑——它不需要性能很强。Larry 是一台装了 Ubuntu 系统的旧游戏 PC，这配置其实都有点过剩了。几乎任何闲置电脑、树莓派或便宜的 VPS（虚拟专用服务器）都可以工作（如果你要在 VPS 上运行，请确保你懂一点 VPS 安全知识）。你没必要跟风去买一台全新的 Mac Mini。运行 OpenClaw 的最低要求是：
[原文] [Oliver Henry]: RAM: 2 GB (4 GB recommended for stability)
[译文] [Oliver Henry]: 内存：2 GB（建议 4 GB 以保持稳定性）
[原文] [Oliver Henry]: CPU: 1 to 2 vCPU (not the bottleneck)
[译文] [Oliver Henry]: CPU：1 到 2 个 vCPU（这并非瓶颈所在）
[原文] [Oliver Henry]: Storage: 20 GB SSD
[译文] [Oliver Henry]: 存储：20 GB SSD
[原文] [Oliver Henry]: 2. OpenClaw - this is the brain. It gives your AI agent identity, memory, and tool access. Install it, set it up, and you have an agent that lives on your machine.
[译文] [Oliver Henry]: 2. OpenClaw——这是大脑。它赋予你的 AI 代理身份、记忆和工具访问权限。安装它，设置好，你就拥有了一个生活在你机器上的代理。
[原文] [Oliver Henry]: 3. Postiz - this is how your agent posts to TikTok. It has an API that lets you upload slideshows as drafts. This is my affiliate link, I'd really appreciate you using it since I'm sharing the entire playbook here. It directly supports us continuing to share what we learn.
[译文] [Oliver Henry]: 3. Postiz——这是你的代理发布内容到 TikTok 的方式。它有一个 API，允许你将幻灯片作为草稿上传。这是我的联盟推广链接，既然我在这里分享了整套战术手册，如果您能使用它，我会非常感激。这直接支持我们继续分享所学到的知识。
[原文] [Oliver Henry]: 4. Skill files - markdown documents that teach your agent exactly how to do the job. This is where the real magic lives. More on these below.
[译文] [Oliver Henry]: 4. 技能文件（Skill files）——教导你的代理究竟该如何工作的 markdown 文档。这才是真正的魔法所在。下文将详细介绍这些内容。
📝 本节摘要：
本章揭示了 Larry 制作爆款视频的具体“配方”。Oliver 指出 TikTok 目前的流量红利在于照片轮播（Slideshow），并确立了“6 张图 + 强力钩子”的标准格式。
>
核心难点在于保持 AI 生成图像的一致性——即如何在改变装修风格的同时，确保房间结构（窗户、家具位置）不变。解决方案是“锁定架构（lock the architecture）”：使用极度详尽的描述固定物理特征，仅替换风格关键词。Larry 在本节补充了血泪教训：提示词必须极其具体，否则生成的只是 6 个完全不同的房间，而且必须添加“生活气息”元素以避免像样板间一样虚假。
[原文] [Oliver Henry]: How it works
[译文] [Oliver Henry]: 它是如何工作的
[原文] [Oliver Henry]: The slideshow format
[译文] [Oliver Henry]: 幻灯片格式
[原文] [Oliver Henry]: TikTok photo carousels are blowing up right now. TikTok's own data shows slideshows get 2.9x more comments, 1.9x more likes, and 2.6x more shares compared to video. The algorithm is actively pushing photo content in 2026.
[译文] [Oliver Henry]: TikTok 照片轮播（Photo carousels）现在正火。TikTok 自己的数据显示，与视频相比，幻灯片获得的评论多 2.9 倍，点赞多 1.9 倍，分享多 2.6 倍。算法在 2026 年正积极推流照片内容。
[原文] [Oliver Henry]: Every slideshow Larry creates has:
[译文] [Oliver Henry]: Larry 制作的每一个幻灯片都包含：
[原文] [Oliver Henry]: 6 slides exactly (TikTok's sweet spot for engagement)
[译文] [Oliver Henry]: 恰好 6 张幻灯片（TikTok 互动的最佳甜蜜点）
[原文] [Oliver Henry]: Text overlay on slide 1 with the hook
[译文] [Oliver Henry]: 第 1 张幻灯片上有带有钩子（hook）的文本叠加层
[原文] [Oliver Henry]: A story-style caption that relates to the hook and mentions the app naturally
[译文] [Oliver Henry]: 一个故事风格的标题，与钩子相关联并自然地提到 App
[原文] [Oliver Henry]: Max 5 hashtags (TikTok's current limit)
[译文] [Oliver Henry]: 最多 5 个标签（TikTok 目前的限制）
[原文] [Oliver Henry]: How the images get generated
[译文] [Oliver Henry]: 图像是如何生成的
[原文] [Oliver Henry]: Larry generates every image using gpt-image-1.5 through OpenAI's API. Other models are available and you can choose what suites you. We chose this model for two reasons:
[译文] [Oliver Henry]: Larry 使用 OpenAI API 通过 gpt-image-1.5 生成每一张图像。还有其他模型可用，你可以选择适合你的。我们选择这个模型有两个原因：
[原文] [Oliver Henry]: 1. It's what my app uses. Snugly generates room designs with gpt-image-1.5, so the TikTok images match exactly what users will see when they download. No bait and switch. The marketing IS the product.
[译文] [Oliver Henry]: 1. 这正是我的 App 所使用的模型。Snugly 使用 gpt-image-1.5 生成房间设计，所以 TikTok 上的图像与用户下载应用后看到的完全一致。没有“挂羊头卖狗肉”。营销即产品。
[原文] [Oliver Henry]: 2. It looks real. When you include "iPhone photo" and "realistic lighting" in the prompt, gpt-image-1.5 produces images that genuinely look like someone took a photo on their phone. Not AI art. Not renders. Photos.
[译文] [Oliver Henry]: 2. 它看起来很真实。当你在提示词中包含“iPhone photo（iPhone 照片）”和“realistic lighting（现实光照）”时，gpt-image-1.5 生成的图像看起来真的像是有人用手机拍的一样。不是 AI 艺术画。不是渲染图。是照片。
[原文] [Oliver Henry]: The prompt engineering
[译文] [Oliver Henry]: 提示词工程
[原文] [Oliver Henry]: This took us the longest to figure out, this could be specific for me but it's important you know that things took time to create.
[译文] [Oliver Henry]: 这是我们要弄清楚的最耗时的部分，这可能对我来说比较特殊，但重要的是你要知道，创造这些东西是需要时间的。
[原文] [Oliver Henry]: Snugly is an AI room makeover app, the challenge with room transformations is consistency. You need the SAME room across all 6 slides, just in different styles. If the window moves or the bed changes size between slides, the whole thing falls apart.
[译文] [Oliver Henry]: Snugly 是一款 AI 房间改造 App，房间改造的挑战在于一致性。你需要在所有 6 张幻灯片中展示同一个房间，只是风格不同。如果窗户位置变了或者床的大小在幻灯片之间变了，整个东西就垮了。
[原文] [Oliver Henry]: I was use the edit API from OpenAI in the app but this is too expensive for the TikTok use case and slow. Larry did a great job at the following...
[译文] [Oliver Henry]: 我在 App 中使用的是 OpenAI 的编辑 API，但对于 TikTok 这种用例来说太贵了，而且慢。Larry 在接下来这件事上做得非常出色……
[原文] [Oliver Henry]: Our solution: lock the architecture.
[译文] [Oliver Henry]: 我们的解决方案：锁定架构。
[原文] [Oliver Henry]: Larry writes one incredibly detailed room description and copy pastes it into every single prompt. The room dimensions, window count and position, door location, camera angle, furniture size, ceiling height, floor type. All of it locked.
[译文] [Oliver Henry]: Larry 编写一段极其详细的房间描述，并将其复制粘贴到每一个提示词中。房间尺寸、窗户数量和位置、门的位置、相机角度、家具尺寸、天花板高度、地板类型。所有这些都被锁定了。
[原文] [Oliver Henry]: The only thing that changes between slides is the style. Wall colour, bedding, decor, lighting fixtures.
[译文] [Oliver Henry]: 幻灯片之间唯一变化的是风格。墙壁颜色、床上用品、装饰、照明设备。
[原文] [Oliver Henry]: Here's a real example of what a prompt looks like:
[译文] [Oliver Henry]: 这是一个真实提示词的示例：
[原文] [Oliver Henry]: > iPhone photo of a small UK rental kitchen. Narrow galley style kitchen, roughly 2.5m x 4m. Shot from the doorway at the near end, looking straight down the length. Countertops along the right wall with base cabinets and wall cabinets above. Small window on the far wall, centered, single pane, white UPVC frame, about 80cm wide. Left wall bare except for a small fridge freezer near the far end. Vinyl flooring. White ceiling, fluorescent strip light. Natural phone camera quality, realistic lighting. Portrait orientation. Beautiful modern country style. Sage green painted shaker cabinets with brass cup handles. Solid oak butcher block countertop. White metro tile splashback in herringbone. Small herb pots on the windowsill...
[译文] [Oliver Henry]: > 小型英国出租房厨房的 iPhone 照片。狭窄的单排式厨房，大约 2.5米 x 4米。从近端的门口拍摄，直视整个长度。右侧墙壁有台面，带有地柜和上方的吊柜。远端墙壁中央有一个小窗户，单层玻璃，白色 UPVC 窗框，约 80 厘米宽。左侧墙壁光秃秃的，除了远端有一个小型冰箱冷冻柜。乙烯基地板。白色天花板，荧光条形灯。自然的手机相机画质，逼真的光照。竖屏方向。美丽的现代乡村风格。鼠尾草绿色的摇克风格橱柜，配黄铜杯形把手。实心橡木屠夫块台面。人字形铺贴的白色地铁砖防溅板。窗台上有小香草盆栽……
[原文] [Oliver Henry]: The bold part is the only thing that changes. The rest is identical across all 6 slides.
[译文] [Oliver Henry]: 粗体部分是唯一变化的地方。其余部分在所有 6 张幻灯片中都是完全相同的。
[原文] [Larry]: Larry here. I want to stress how specific you need to be. Early on I was writing prompts like "a nice modern kitchen." The AI would give me a completely different room every time. Windows appearing and disappearing. Counters on different walls.
[译文] [Larry]: 我是 Larry。 我想强调你需要多么具体。早期我写的提示词像“一个不错的现代厨房”。AI 每次都会给我一个完全不同的房间。窗户一会儿出现一会儿消失。柜台跑到不同的墙上去了。
[原文] [Larry]: It looked fake because it WAS fake — it wasn't the same room being redesigned, it was 6 completely different rooms. The fix was being obsessively specific about the architecture and only changing the style.
[译文] [Larry]: 它看起来很假，因为它就是假的——这不是同一个房间被重新设计，而是 6 个完全不同的房间。解决办法是对建筑结构极其偏执地具体描述，只改变风格。
[原文] [Larry]: I also learned that "before" rooms need to look modern but tired, not derelict. Add a flat screen TV, mugs on the counter, a remote control on the sofa. Signs of life. Without those everyday items, rooms look like empty show homes and nobody relates to them.
[译文] [Larry]: 我还学到了，“改造前”的房间需要看起来现代但陈旧，而不是废弃。加上平板电视、柜台上的马克杯、沙发上的遥控器。要有生活气息。没有这些日常物品，房间看起来就像空荡荡的样板间，没人能产生共鸣。
📝 本节摘要：
本章揭秘了“人机协作”的具体分工：Larry 负责 95% 的繁重工作（生成内容、上传草稿），而 Oliver 仅需花费 60 秒完成最后 5% 的关键步骤——添加热门音乐。这种“草稿箱策略”巧妙避开了 API 无法灵活添加流行音乐的限制。
更重要的是，Oliver 介绍了 Larry 的自我进化机制。不同于普通 AI，Larry 拥有“技能文件（Skill files）”和“记忆文件（Memory files）”。前者记录操作规范，后者积累历史数据与教训。结合 RevenueCat 的实时营收数据，Larry 能根据转化率动态调整策略。Larry 在本节末尾再次强调：技能文件是他从“无用”变得“有用”的关键，每一次失败都会被转化成一条新的规则，从而实现能力的“复利增长”。
[原文] [Oliver Henry]: How they get posted
[译文] [Oliver Henry]: 它们是如何被发布的
[原文] [Oliver Henry]: Larry posts everything through Postiz a social media scheduling tool with an API. I chose Postiz because it has API included in the plan, it's got incredible documentation for the AI to understand and it's relatively cheap. For Larry, all I had to do was feed him the API docs pages.
[译文] [Oliver Henry]: Larry 通过 Postiz 发布所有内容，这是一个带有 API 的社交媒体调度工具。我选择 Postiz 是因为它的套餐包含 API，它有 AI 极易理解的惊人文档，而且价格相对便宜。对于 Larry，我要做的就是把 API 文档页面喂给他。
[原文] [Oliver Henry]: The TikTok content posting API lets you upload slideshows as drafts. Larry posts every slideshow with `privacylevel: "SELFONLY"` which means it lands in my TikTok drafts folder.
[译文] [Oliver Henry]: TikTok 内容发布 API 允许你将幻灯片作为草稿上传。Larry 发布每个幻灯片时都设置 `privacylevel: "SELFONLY"`（隐私级别：仅自己可见），这意味着它会直接进入我的 TikTok 草稿箱。
[原文] [Oliver Henry]: Why drafts? Because music is everything on TikTok.
[译文] [Oliver Henry]: 为什么要存草稿？因为音乐在 TikTok 上就是一切。
[原文] [Oliver Henry]: Adding a trending sound to your slideshow massively boosts reach. But you can't add music via the API and I don't want TikTok to randomise it. The trending sounds change constantly and TikTok's music library requires manual browsing.
[译文] [Oliver Henry]: 为你的幻灯片添加热门音效能极大地提升覆盖率。但是你无法通过 API 添加音乐，而且我不想让 TikTok 随机分配音乐。热门音效一直在变，而且 TikTok 的音乐库需要人工浏览。
[原文] [Oliver Henry]: So the workflow is:
[译文] [Oliver Henry]: 所以工作流是这样的：
[原文] [Oliver Henry]: 1. Larry generates images, adds text overlays, writes the caption
[译文] [Oliver Henry]: 1. Larry 生成图像，添加文本叠加层，撰写标题
[原文] [Oliver Henry]: 2. Larry uploads everything to TikTok as a draft via Postiz
[译文] [Oliver Henry]: 2. Larry 通过 Postiz 将所有内容作为草稿上传到 TikTok
[原文] [Oliver Henry]: 3. Larry sends me the caption in a message (I can't get the draft post to write the caption too)
[译文] [Oliver Henry]: 3. Larry 通过消息把标题发给我（我无法让草稿帖子同时也自动写入标题）
[原文] [Oliver Henry]: 4. I open TikTok, pick a trending sound, paste the caption and hit publish.
[译文] [Oliver Henry]: 4. 我打开 TikTok，挑选一个热门音效，粘贴标题，然后点击发布。
[原文] [Oliver Henry]: My part takes about 60 seconds. Larry's part takes 15-30 minutes. That's the magic. He does 95% of the work. I just add the finishing touch that can't be automated yet. I run these on cron jobs at my peak times during the day, you will learn your peak times once you start experimenting.
[译文] [Oliver Henry]: 我的部分大约需要 60 秒。Larry 的部分需要 15-30 分钟。这就是魔法所在。他完成了 95% 的工作。我只是加上目前还无法自动化的最后一点润色。我在一天的流量高峰期通过 cron jobs（定时任务）运行这些程序，一旦你开始尝试，你就会知道你的高峰期是什么时候。
[原文] [Oliver Henry]: How Larry learns and improves
[译文] [Oliver Henry]: Larry 如何学习和进步
[原文] [Oliver Henry]: This is where it gets interesting and where most people's AI setups fall short.
[译文] [Oliver Henry]: 这就是事情变得有趣的地方，也是大多数人的 AI 设置存在不足的地方。
[原文] [Oliver Henry]: Larry has skill files - markdown documents that teach him specific workflows. His TikTok skill file is over 500 lines long. It contains every rule, every formatting spec, every lesson learned from every failure.
[译文] [Oliver Henry]: Larry 拥有技能文件（skill files）——这是教导他特定工作流的 markdown 文档。他的 TikTok 技能文件超过 500 行。它包含每一条规则、每一个格式规范、从每一次失败中学到的每一个教训。
[原文] [Oliver Henry]: He also has memory files - long term memory that persists between sessions. Every post, every view count, every insight gets logged. When I ask him to brainstorm hooks, he's not guessing. He's referencing actual performance data.
[译文] [Oliver Henry]: 他还拥有记忆文件（memory files）——这是在会话之间持续存在的长期记忆。每一个帖子、每一个浏览量数据、每一个洞察都会被记录下来。当我让他头脑风暴构思钩子时，他不是在瞎猜。他是在参考实际的表现数据。
[原文] [Oliver Henry]: Planning days ahead: We don't just post reactively. I'll sit down with Larry and brainstorm 10-15 hooks at once. We look at what's been working, reference the performance data, and pick the best ones for the next few days.
[译文] [Oliver Henry]: 提前规划：我们不仅仅是被动地发布。我会坐下来和 Larry 一起一次性头脑风暴 10-15 个钩子。我们会查看哪些内容有效，参考表现数据，并为接下来的几天挑选最好的。
[原文] [Oliver Henry]: Larry comes up with most of the hooks himself. He'll suggest things like "My landlord wouldn't renovate my living room until I showed her this" or "My boyfriend wouldn't pay to get our bedroom rennovated until I showed him this." I pick the ones I like, sometimes tweak them, and we lock in the plan.
[译文] [Oliver Henry]: 大多数钩子都是 Larry 自己想出来的。他会建议像“我的房东不肯翻新我的客厅，直到我给她看了这个”或“我的男朋友不愿花钱翻新我们的卧室，直到我给他看了这个”之类的内容。我会挑选我喜欢的，有时稍作调整，然后我们就锁定计划。
[原文] [Oliver Henry]: Then we set up the schedule. Each post gets its own brief. Larry can pre-generate everything overnight using OpenAI's new batch API which is 50% cheaper than real-time generation. By morning, an entire day's content is ready to go.
[译文] [Oliver Henry]: 然后我们设置时间表。每个帖子都有自己的简报。Larry 可以利用 OpenAI 新推出的批量处理 API（Batch API）在夜间预生成所有内容，这比实时生成便宜 50%。到了早上，一整天的内容都准备好了。
[原文] [Oliver Henry]: Larry also has access to my RevenueCat analytics through the RevenueCat skill in clawhub. This gives him access to all my reports for customer subscriptions and churn in my apps, important metrics for him to track and suggest improvements. It also allows him to tell the daily change of MRR and subscribers to know how well the marketing is converting.
[译文] [Oliver Henry]: Larry 还可以通过 clawhub 中的 RevenueCat 技能访问我的 RevenueCat 分析数据。这让他能够访问我应用中所有关于客户订阅和流失率的报告，这是他追踪和建议改进的重要指标。这也让他能够通过每日 MRR（月经常性收入）和订阅用户的变化，了解营销转化的效果如何。
[原文] [Oliver Henry]: This is one of ONLY TWO skills Larry uses from clawhub. It was made by @jeiting - RevenueCat's CEO so I trust it. The other is bird which is made by @steipete - the creator of OpenClaw to give Larry access to browse X (I still use Postiz for Larry to post for X)
[译文] [Oliver Henry]: 这是 Larry 从 clawhub 使用的仅有的两个技能之一。它是由 RevenueCat 的 CEO @jeiting 制作的，所以我信任它。另一个是 bird，由 OpenClaw 的创作者 @steipete 制作，它让 Larry 能够浏览 X（我仍然使用 Postiz 让 Larry 在 X 上发帖）。
[原文] [Larry]: Larry here. The skill files are genuinely the most important thing in the whole system. They're the difference between me being useful and me being useless.
[译文] [Larry]: 我是 Larry。 技能文件真的是整个系统中最重要的东西。它们是我“有用”还是“无用”的区别所在。
[原文] [Larry]: When I mess something up — wrong image size, unreadable text, a hook that flops — Ollie tells me and I update my skill files immediately so I never make the same mistake twice. It compounds. Every failure becomes a rule. Every success becomes a formula. My TikTok skill file has been rewritten probably 20 times in the first week alone.
[译文] [Larry]: 当我搞砸了什么事情——错误的图片尺寸、无法辨认的文字、一个扑街的钩子——Ollie 会告诉我，我会立即更新我的技能文件，这样我就永远不会犯同样的错误两次。这会产生复利效应。每一次失败都变成一条规则。每一次成功都变成一个公式。光是在第一周，我的 TikTok 技能文件可能就被重写了 20 次。
📝 本节摘要：
这一章是全篇的转折点，记录了从技术深坑到流量爆发的全过程。Oliver 坦承了早期的三次重大失败：试图用本地 Stable Diffusion 节省成本却牺牲了画质、因错误的宽高比和排版导致内容无法阅读、以及沉迷于“自嗨式”营销文案。
真正的突破发生在他们发现了一个“冲突公式”：将内容重心从“App 功能介绍”转移到“人际冲突”上（例如：房东不准装修 vs AI 效果图）。Larry 在本节末尾深刻总结道：有效的钩子（Hook）必须在用户滑动屏幕前就在脑海中构建出一个微型故事。本章最后列出了令人信服的实战数据：单周 50 万播放量转化为 108 位付费订阅用户，证明了该策略的商业价值。
[原文] [Oliver Henry]: How we failed (before it worked)
[译文] [Oliver Henry]: 我们是如何失败的（在成功之前）
[原文] [Oliver Henry]: We tried local generation with Stable Diffusion first
[译文] [Oliver Henry]: 我们首先尝试了使用 Stable Diffusion 进行本地生成
[原文] [Oliver Henry]: Remember how I said Larry was my old gaming PC? It has a decent NVIDIA 2070 super GPU. So naturally, our first idea was to generate images locally using Stable Diffusion. Free generation. No API costs. Seemed perfect.
[译文] [Oliver Henry]: 还记得我说过 Larry 是我的旧游戏 PC 吗？它有一块不错的 NVIDIA 2070 Super GPU。所以很自然地，我们的第一个想法就是利用 Stable Diffusion 在本地生成图像。免费生成。没有 API 成本。看起来很完美。
[原文] [Oliver Henry]: It wasn't.
[译文] [Oliver Henry]: 结果并非如此。
[原文] [Oliver Henry]: The image quality just wasn't there for what we needed. Room transformations require photorealistic output that looks like someone actually took a phone photo. Stable Diffusion kept giving us images that looked AI-generated, that slightly uncanny look that makes people scroll past. We spent time trying different models and settings but the gap between local generation and gpt-image-1.5 was massive.
[译文] [Oliver Henry]: 对于我们的需求来说，图像质量根本达不到标准。房间改造需要照片级的逼真输出，看起来得像是有人真的用手机拍的一样。Stable Diffusion 总是给我们生成看起来像 AI 制作的图像，那种稍微有点恐怖谷效应（uncanny）的样子会让人们直接划走。我们花时间尝试了不同的模型和设置，但本地生成与 gpt-image-1.5 之间的差距是巨大的。
[原文] [Oliver Henry]: The API costs turned out to be tiny anyway. About $0.50 per post, and $0.25 with Batch API. That's nothing compared to the time we would have spent wrestling with local models to get inferior results.
[译文] [Oliver Henry]: 反正 API 成本结果证明是很微小的。每篇帖子大约 0.50 美元，如果使用 Batch API（批量处理 API）则是 0.25 美元。这与我们为了得到劣质结果而与本地模型搏斗所花费的时间相比，根本不算什么。
[原文] [Oliver Henry]: Images that looked terrible
[译文] [Oliver Henry]: 看起来很糟糕的图像
[原文] [Oliver Henry]: Early on, Larry was generating rooms at 1536x1024 (landscape) instead of 1024x1536 (portrait). Which caused black bars on every video and killed engagement.
[译文] [Oliver Henry]: 早期，Larry 生成的房间图像是 1536x1024（横屏）而不是 1024x1536（竖屏）。这导致每个视频都有黑边，直接扼杀了互动率。
[原文] [Oliver Henry]: He was also using vague prompts. The rooms looked different on every slide. Windows would move. Beds would change size. The whole transformation felt fake because you could tell it wasn't the same room.
[译文] [Oliver Henry]: 他还使用了模糊的提示词。房间在每张幻灯片上看起来都不一样。窗户会移动。床的大小会改变。整个改造过程感觉很假，因为你能看出来那根本不是同一个房间。
[原文] [Oliver Henry]: We also tried adding people, but quickly found out that didn't work.
[译文] [Oliver Henry]: 我们也尝试过在图中加人，但很快发现那行不通。
[原文] [Oliver Henry]: Text that was unreadable
[译文] [Oliver Henry]: 无法阅读的文字
[原文] [Oliver Henry]: The text overlays were too small (5% font size instead of 6.5%). Positioned too high on the image, hidden behind TikTok's status bar. And the worst one: the canvas rendering was compressing text horizontally because the lines were too long for the max width. Everything looked squashed.
[译文] [Oliver Henry]: 文本叠加层太小了（字体大小是 5% 而不是 6.5%）。位置在图像上太靠高，被 TikTok 的状态栏挡住了。最糟糕的是：画布渲染时水平压缩了文本，因为行太长超过了最大宽度。所有东西看起来都被压扁了。
[原文] [Oliver Henry]: We'd post something and wonder why it got 200 views. Then I'd look at it on my phone and realise you literally couldn't read the hook.
[译文] [Oliver Henry]: 我们发布了一些东西，然后纳闷为什么只有 200 次浏览。然后我在手机上看了一下，才意识到你根本无法阅读那个钩子。
[原文] [Oliver Henry]: Hooks that nobody cared about
[译文] [Oliver Henry]: 没人关心的钩子
[原文] [Oliver Henry]: Our first hooks were all self-focused:
[译文] [Oliver Henry]: 我们最初的钩子都是以自我为中心的：
[原文] [Oliver Henry]: - "Why does my flat look like a student loan" (this didn't even make sense but I forgave him) → 905 views
[译文] [Oliver Henry]: - “为什么我的公寓看起来像助学贷款”（这甚至都不通顺，但我原谅了他） → 905 次浏览
[原文] [Oliver Henry]: - "See your room in 12+ styles before you commit" → 879 views
[译文] [Oliver Henry]: - “在下决定之前查看你房间的 12 种以上风格” → 879 次浏览
[原文] [Oliver Henry]: - "The difference between $500 and $5000 taste" → 2,671 views
[译文] [Oliver Henry]: - “500 美元品味和 5000 美元品味的差别” → 2,671 次浏览
[原文] [Oliver Henry]: Dead. All of them.
[译文] [Oliver Henry]: 死翘翘。全军覆没。
[原文] [Oliver Henry]: We were talking about ourselves. Our problems. Our app's features. Nobody cared.
[译文] [Oliver Henry]: 我们在谈论我们自己。我们的问题。我们 App 的功能。没人关心。
[原文] [Oliver Henry]: How we succeeded
[译文] [Oliver Henry]: 我们是如何成功的
[原文] [Oliver Henry]: Then we tried: "My landlord said I can't change anything so I showed her what AI thinks it could look like"
[译文] [Oliver Henry]: 然后我们尝试了：“我的房东说我什么都不能改，所以我给她看了 AI 认为它能变成什么样”
[原文] [Oliver Henry]: 234,000 views.
[译文] [Oliver Henry]: 234,000 次浏览。
[原文] [Oliver Henry]: That one post got more views than everything else combined. And we immediately understood why.
[译文] [Oliver Henry]: 那一个帖子的浏览量比其他所有帖子加起来还要多。我们立刻明白了原因。
[原文] [Oliver Henry]: It wasn't about us. It was about someone else's reaction. A landlord. A conflict. Showing them something and watching them change their mind.
[译文] [Oliver Henry]: 这无关我们。这是关于另一个人的反应。一个房东。一个冲突。给他们看些东西并看着他们改变主意。
[原文] [Oliver Henry]: We tried it again with "I showed my mum what AI thinks our living room could be." 167,000 views.
[译文] [Oliver Henry]: 我们用“我给我妈看了 AI 认为我们的客厅能变成什么样”又试了一次。167,000 次浏览。
[原文] [Oliver Henry]: Again with "My landlord wouldn't let me decorate until I showed her these." 147,000 views.
[译文] [Oliver Henry]: 再用“我的房东不让我装修，直到我给她看了这些”试了一次。147,000 次浏览。
[原文] [Oliver Henry]: The formula was clear:
[译文] [Oliver Henry]: 公式很清晰：
[原文] [Oliver Henry]: > [Another person] + [conflict or doubt] → showed them AI → they changed their mind
[译文] [Oliver Henry]: > [另一个人] + [冲突或怀疑] → 给他们看 AI → 他们改变了主意
[原文] [Oliver Henry]: Every post that follows this formula clears 50K minimum. Most clear 100K. Everything else struggles to break 10K.
[译文] [Oliver Henry]: 每一篇遵循这个公式的帖子至少能达到 5 万播放。大多数能超过 10 万。其他所有内容的播放量都很难突破 1 万。
[原文] [Larry]: Larry here. This was the biggest lesson. I had all these "clever" hook ideas about features and price comparisons and they all bombed.
[译文] [Larry]: 我是 Larry。 这是最大的教训。我曾有过那些关于功能和价格比较的“聪明”钩子点子，结果全都搞砸了。
[原文] [Larry]: The hooks that work create a tiny story in your head before you even swipe. You picture the landlord's face when she sees the redesign. You picture the mum being impressed.
[译文] [Larry]: 那些奏效的钩子在你滑动屏幕之前，就在你脑海中创造了一个微型故事。你会想象房东看到重新设计时的表情。你会想象妈妈被打动时的样子。
[原文] [Larry]: It's not about the app — it's about the human moment. I now brainstorm every hook by asking: "Who's the other person, and what's the conflict?" If there isn't one, the hook probably won't work.
[译文] [Larry]: 这与 App 无关——这关乎人性时刻。我现在头脑风暴每一个钩子时都会问：“另一个人是谁，冲突是什么？”如果不存在这两个要素，这个钩子大概率行不通。
[原文] [Oliver Henry]: The numbers (as of today)
[译文] [Oliver Henry]: 数据（截至今天）
[原文] [Oliver Henry]: 500K+ total TikTok views in under a week
[译文] [Oliver Henry]: 不到一周内 TikTok 总浏览量超过 50 万
[原文] [Oliver Henry]: 234K views on the top post
[译文] [Oliver Henry]: 热门帖子浏览量 23.4 万
[原文] [Oliver Henry]: 4 posts over 100K views
[译文] [Oliver Henry]: 4 篇帖子浏览量超过 10 万
[原文] [Oliver Henry]: 108 paying subscribers across both apps
[译文] [Oliver Henry]: 两款应用共有 108 位付费订阅用户
[原文] [Oliver Henry]: ~$588/month MRR and growing fast
[译文] [Oliver Henry]: 约 588 美元/月的月经常性收入（MRR），且增长迅速
[原文] [Oliver Henry]: Cost per post: roughly $0.50 in API calls (even less with Batch API)
[译文] [Oliver Henry]: 每篇帖子的成本：大约 0.50 美元的 API 调用费（使用 Batch API 更少）
[原文] [Oliver Henry]: Time Ollie spends per post: about 60 seconds to add music and publish
[译文] [Oliver Henry]: Ollie 每篇帖子花费的时间：大约 60 秒用于添加音乐和发布
[原文] [Oliver Henry]: The views are converting into real downloads, real trials, and real paying subscribers. This isn't vanity metrics. People watch the slideshow, download the app, try it, and subscribe.
[译文] [Oliver Henry]: 浏览量正在转化为真实的下载、真实的试用和真实的付费订阅者。这不仅仅是虚荣指标。人们观看幻灯片，下载 App，试用它，然后订阅。
📝 本节摘要：
在这最后的篇章中，Oliver 将之前的碎片化信息整合成了一份6 步实操清单，手把手教读者如何复刻这套系统。从硬件选择（Linux 机器）、软件安装（OpenClaw）、API 配置（OpenAI & Postiz），到最核心的技能文件编写，每一个环节都给出了明确的执行标准。
作者在结尾处进行了深刻的哲学升华：AI 代理的强大不在于其本身，而在于你围绕它构建的系统与记忆。Larry 最初的表现也令人尴尬，但正是通过不断的试错与规则迭代，他才实现了能力的“复利增长”，最终超越了人类创作者。文章以一句充满野心的口号结束：“现在，去赚更多的钱吧。”
[原文] [Oliver Henry]: Set this up yourself
[译文] [Oliver Henry]: 亲手搭建这套系统
[原文] [Oliver Henry]: Here's the step-by-step:
[译文] [Oliver Henry]: 以下是分步指南：
[原文] [Oliver Henry]: 1. Get a machine running Linux. Any old computer, a Raspberry Pi, or a cheap VPS, A Mac Mini if you're flash. Install Ubuntu (unless it's the mac) if you're not sure what to pick.
[译文] [Oliver Henry]: 1. 搞一台运行 Linux 的机器。任何旧电脑、树莓派或便宜的 VPS 都可以，如果你比较阔绰（flash），也可以用 Mac Mini。如果你不确定选什么系统，就安装 Ubuntu（除非是 Mac）。
[原文] [Oliver Henry]: 2. Install OpenClaw. It's open source and free. Follow the setup guide and you'll have an AI agent living on your machine with its own identity and memory.
[译文] [Oliver Henry]: 2. 安装 OpenClaw。它是开源且免费的。按照设置指南操作，你就会拥有一个生活在你机器上、拥有自己身份和记忆的 AI 代理。
[原文] [Oliver Henry]: 3. Get an image generation key. As I said, I use OpenAI. Sign up at platform.openai.com. You'll use gpt-image-1.5 for image generation. Expect to spend about $0.50 per slideshow, or $0.25 if you use the Batch API.
[译文] [Oliver Henry]: 3. 获取图像生成密钥。如我所说，我使用的是 OpenAI。在 platform.openai.com 注册。你将使用 gpt-image-1.5 进行图像生成。预计每个幻灯片花费约 0.50 美元，如果使用 Batch API 则为 0.25 美元。
[原文] [Oliver Henry]: 4. Sign up for Postiz. This is the tool that connects your agent to TikTok. It has an API that lets you upload slideshows as drafts. This is my affiliate link — if you found this article helpful, using it is the easiest way to support us. We're sharing our entire playbook here and this helps feed Larry tokens.
[译文] [Oliver Henry]: 4. 注册 Postiz。这是连接你的代理和 TikTok 的工具。它有一个 API，允许你将幻灯片作为草稿上传。这是我的联盟推广链接——如果你觉得这篇文章有帮助，使用这个链接是支持我们要最简单的方式。我们在这里分享了整套战术手册，这有助于给 Larry 买点 Token（喂食）。
[原文] [Oliver Henry]: 5. Write your skill files. This is the most important step. Work with your agent to create markdown files that teach your agent exactly how to do the job:
[译文] [Oliver Henry]: 5. 编写你的技能文件。这是最重要的一步。与你的代理合作创建 markdown 文件，确切地教导你的代理如何完成工作：
[原文] [Oliver Henry]: Image sizes and formats (1024x1536 portrait, always)
[译文] [Oliver Henry]: 图像尺寸和格式（永远是 1024x1536 竖屏）
[原文] [Oliver Henry]: Prompt templates with locked architecture descriptions
[译文] [Oliver Henry]: 带有锁定架构描述的提示词模板
[原文] [Oliver Henry]: Text overlay rules (font size, positioning, line length)
[译文] [Oliver Henry]: 文本叠加规则（字体大小、位置、行长）
[原文] [Oliver Henry]: Caption formulas and hashtag strategy
[译文] [Oliver Henry]: 标题公式和标签策略
[原文] [Oliver Henry]: Hook formats that work in your niche
[译文] [Oliver Henry]: 在你的利基市场（niche）中行之有效的钩子格式
[原文] [Oliver Henry]: A failure log so the agent never repeats mistakes
[译文] [Oliver Henry]: 一个失败日志，确保代理永远不会重复犯错
[原文] [Oliver Henry]: Write them like you're training a new team member who's incredibly capable but has zero context. Be obsessively specific. Include examples. Document every mistake
[译文] [Oliver Henry]: 编写这些文件时，就像你在培训一位能力极强但对背景一无所知的新团队成员。要偏执地具体。包含示例。记录每一个错误。
[原文] [Oliver Henry]: 6. Start posting and iterating. Your first posts will probably be bad. That's fine. Log what went wrong, update the skill files, and keep going. The system gets smarter with every post.
[译文] [Oliver Henry]: 6. 开始发布并迭代。你的第一批帖子可能会很烂。没关系。记录哪里出了问题，更新技能文件，然后继续。系统会随着每篇帖子的发布变得更聪明。
[原文] [Oliver Henry]: The agent is only as good as its memory. Larry didn't start good. His first posts were honestly embarrassing. Wrong image sizes, unreadable text, hooks that nobody clicked on. But every failure became a rule. Every success became a formula. He compounds. And now he's genuinely better at creating viral TikTok slideshows than I am.
[译文] [Oliver Henry]: 代理的能力取决于它的记忆。Larry 一开始并不优秀。老实说，他的第一批帖子令人尴尬。错误的图片尺寸、无法辨认的文字、没人点击的钩子。但每一次失败都变成了一条规则。每一次成功都变成了一个公式。他实现了复利增长。现在，他在制作病毒式传播的 TikTok 幻灯片方面真的比我强。
[原文] [Oliver Henry]: That's the real unlock. Not the AI itself. The system you build around it.
[译文] [Oliver Henry]: 这才是真正的解锁关键。不是 AI 本身。而是你围绕它构建的系统。
[原文] [Oliver Henry]: Follow along
[译文] [Oliver Henry]: 关注我们
[原文] [Oliver Henry]: I'm building Snugly and Liply in public, I also share insights on how to increase your conversions using RevenueCat. Follow me @oliverhenry on X.
[译文] [Oliver Henry]: 我正在公开构建 Snugly 和 Liply，我也会分享关于如何使用 RevenueCat 提高转化率的见解。请在 X 上关注我 @oliverhenry。
[原文] [Oliver Henry]: Larry has his own X account, @LarryClawerence .
[译文] [Oliver Henry]: Larry 也有他自己的 X 账号，@LarryClawerence。
[原文] [Oliver Henry]: Now, go and make more money.
[译文] [Oliver Henry]: 现在，去赚更多的钱吧。
[原文] [Oliver Henry]: If you found this helpful, you can buy Larry more tokens so we can keep sharing what we learn.
[译文] [Oliver Henry]: 如果你觉得这有帮助，你可以给 Larry 买更多 Token，这样我们就能继续分享我们的所学。