超级游艇与裁员的对比
在五月底，一艘名为 Launchpad 的 387 英尺超级游艇挤过了西雅图巴拉德锁。这是该船在 14 年内看到的最大的船只。这东西太庞大了，看起来就像一个试图在浴缸里平行停放的浮动公寓群。它悬挂着马歇尔群岛旗帜，推测是因为税收避难所无法装入注册文件。它身后还拖着一艘支持船和一个价值 1 亿美元的船，我不是在编造这个，Wingman。人们排满了步道观看，然后他们开始嘘声，因为就在马克·扎克伯格的 3 亿美元超级游艇驶入西雅图的同一天，Meta 公布了计划裁减华盛顿州约 1,400 个工作岗位，大约占其当地员工总数的 20%。岸边有人大喊亿万富翁应该交点税。几天之内，就出现了激烈的抗议艺术。你真的必须钦佩这里的效率。通常情况下，亿万富翁会在使用资金进行大规模裁员来资助他们的浮动堡垒之前至少等待一个财政季度。马克基本上看着 1,400 个毁掉的职业生涯，然后基本上说：“完美。这涵盖了 Wingman 直升机停机坪的年度维护费用。”它不仅仅是一艘超级游艇。它实际上是能从太空看到的“中指”的物理体现。Meta 现在发生的事情是我见过的最不寻常的企业内部崩溃之一，但这并不是因为公司失败了，而是因为它在内部撕裂的同时正在疯狂成功。
Original English
In late May, a 387-ft superyacht named Launchpad squeezed through the Ballard Locks in Seattle. It was the biggest vessel the lock operator had seen in 14 years. This thing was so massive, it looked like a floating apartment complex trying to parallel park in a bathtub. It flew a Marshall Islands flag, presumably because tax haven wouldn't fit on the registration. It had a support vessel trailing behind it, and a $100 million boat called, and I'm not making this up, Wingman. People lined the walkway to watch, and then they started booing because on the same day that Mark Zuckerberg's $300 million superyacht sailed into Seattle, Meta disclosed plans to cut 1,400 jobs in Washington state, roughly 20% of its local workforce. Someone on the shore shouted that the billionaire should pay some taxes. Within days, there was flaming protest art. You really have to admire the efficiency here. Usually, billionaires wait at least a fiscal quarter before using the money for mass layoffs to fuel their floating fortresses. Mark basically looked at 1,400 ruined careers and basically said, "Perfect. That covers the annual maintenance on the Wingman's helicopter pad." It's not just a superyacht. It is literally a physical manifestation of a middle finger visible from space. What's happening at Meta right now is one of the most extraordinary corporate implosions I've ever seen, and it's not because the company's failing, but because it is succeeding wildly while tearing itself apart from the inside.领导层与员工士气危机
所以，在这个视频中，我将涵盖 Meta 自身领导层承认接近历史低点的士气危机、公司一直在走而忽略的战略金矿、它拒绝从中学到的 800 亿美元模式，以及这一切给在那里工作的实际人类造成了什么代价。让我们深入研究一下。在 6 月 2 日，在一次名为“周二与老板谈话”的内部会议上，Meta 的首席技术官安德鲁·博斯沃思（Andrew Bosworth）告诉员工，公司士气可能不是过去 20 年来的最差，但它大概是处于高位。这绝对是高位。他能拿出的唯一比较对象是剑桥分析丑闻，即 2018 年的危机，当时数百万 Facebook 个人资料被用于政治竞选活动，包括特朗普 2016 年总统竞选和英国脱欧公投。正如你们所记得的，那场丑闻几乎摧毁了 Meta 的声誉。博斯沃思不是在匿名论坛上发泄不满的中层经理，顺便说一下。他是扎克伯格最资深的副手之一。他于 2006 年加入当时叫 Facebook 的公司，根据通话中的四个人称，他站在公司面前基本上说：“情况几乎和这家公司历史上最糟糕的时刻一样糟。”现在，为什么会这样？表面上的答案是裁员。在 2026 年 5 月，Meta 裁减了大约 8,000 个工作岗位，占其全球员工总数的约 10%。但仅仅是裁员本身并不能产生这种绝望感。公司总是在重组。在 Meta 的不同之处在于裁员的同时发生了什么。大约有 6,500 名工程师和产品经理被抽调到一个名为“应用 AI”的新部门。许多人对这一举措没有发言权。他们被分配了生成用于测试和训练 Meta AI 模型的编码难题的任务，这项工作与他们实际被雇佣去做的工作完全不相上下。一位员工将该部门描述给 Wired 时称其为字面上的一个“古拉格”。其他人称自己是“征召者”。在一些团队中，组织的人员与员工的比例达到了 50 比 1，这是一个完全没有组织意义的数字。想象一下，在一个每名守卫对应 50 个囚犯的地方，除了守卫没有武器。他们只是有点一整天的日程安排和一种迫切需要向你询问你的目标和关键结果的冲动。在 52-1 的情况下，这感觉不像企业层级，而更像一群软件工程师试图记住用户界面长什么样。自四月以来，Meta 还一直在员工笔记本电脑上运行软件，该软件记录按键、点击和定期截图以训练 AI 代理。这没有退出选项。欧洲员工因欧盟的隐私规则而免于此，但美国员工则不然。工人们流传着一份请愿书，并分发了将 Meta 打造成员工数据提取工厂的宣传单。Meta 的最终让步是允许员工每次暂停跟踪 30 分钟。30 分钟，嗯，这太慷慨了，不是吗？这正好足够让你在车里经历一次轻微恐慌发作、对着手工抹茶拿铁哭泣、申请苹果三份工作，然后重新登录，刚好赶上扎克（Zuck）的间谍机器人给你拍一张你存在性绝望的照片。
Original English
So, in this video, I'm going to cover the morale crisis that Meta's own leadership admits is near historic lows, the strategic goldmine the company keeps just walking past, the $80 billion pattern it refuses to learn from, and what all of this cost the actual human beings who work there. Let's get into this. On June 2nd, during an internal meeting called Tuesdays with Boss, Meta's chief technology officer Andrew Bosworth, told employees that morale at the company was maybe not the worst it's ever been in 20 years here, but it's probably up there. It's definitely up there. The only comparison he could reach for was the Cambridge Analytica scandal, the 2018 crisis in which millions of Facebook profiles were harvested and used for political campaigns, including Trump's 2016 presidential bid and the Brexit referendum. The scandal, as you may recall, nearly destroyed Meta's reputation. Bosworth is not some disgruntled mid-level manager venting on an anonymous forum, by the way. He is one of Zuckerberg's longest-serving lieutenants. He joined what was then Facebook in 2006, and according to four people on the call, he stood in front of the company and basically said effectively, "Things are almost as bad as the worst moment in this company's history." Now, why is that? The surface-level answer is the layoffs. In May 2026, Meta cut roughly 8,000 jobs, about 10% of its global workforce. But layoffs alone don't produce this kind of despair. Companies restructure kind of all the time. What's different at Meta is what happened alongside the layoffs. Around 6,500 engineers and product managers were pulled into a new unit called Applied AI. Many had no say in the move. They were given tasks involving generating coding puzzles to test and train Meta's AI models, work that bore no resemblance to the jobs they were actually hired to do. One employee described the unit to Wired as literally a gulag. Others called themselves draftees. The organization had ballooned to a manager-to-employee ratio of 50 to 1 in some teams, which is a number that makes no organizational sense at all. Just imagine a gulag where there are 50 prisoners for every one guard, except the guard doesn't have a weapon. They just kind of have a calendar full of alignment things and a burning need to ask you about your objectives and key results. At 52-1, it feels less like a corporate hierarchy and more like a flash mob of software engineers trying to remember what a user interface looks like. Since April, Meta has also been running software on US employees' laptops that logs keystrokes, clicks, and periodic screenshots to train AI agents. There is no opt-out from this. European staff are exempt because of the EU's privacy rules, but American employees are not. Workers circulated a petition and handed out flyers branding Meta an employee data extraction factory. Meta's eventual concession was to let employees pause the tracking for 30 minutes at a time. 30 minutes, well, that is so generous, isn't it? That is exactly enough time to have a minor panic attack in your car, cry into an artisanal matcha latte, apply for three jobs at Apple, and just log back in just in time for Zuck's spy robot to take a screenshot of your existential despair.应对危机与虚假激励
在公司全员大会上，有人劫持了直播流，对着 Meta 的 AI 高层大喊亵渎之词，面对数千名员工。其中一位演讲者用双手遮住了脸。多位员工告诉记者，他们实际上正积极希望被解雇，只是为了拿到遣散费，这相当于 16 周的工资和 18 个月的医疗保险。当你的员工们都在祈求退出包时，系统已经出了结构性错误。在后续备忘录中，老板扎克伯格承认自己对重组背后的愿景解释得非常糟糕，并承认公司破坏了员工对自身专业知识价值的信任。而 Meta 对这种近乎历史性的道德崩溃的回应是天才之举。请鼓掌，大家。他们增加了办公零食的预算。他们宣布更好的休息室，扎克伯格发了一封精彩备忘录，敦促员工参加一个精彩的黑客马拉松。没有什么能真正说：“我们重视你的牺牲，因为我们正在收获你的按键记录，就像无限供应有机切达小兔子一样，对吧？抱歉，Kevin，我们拿走了你的职业稳定性和数字隐私，但看，我们现在增加了一个提供血橙口味的苏打水的龙头。”一位员工回信说，他们实际上一直忙于为团队保持灯亮。另一位说：“我不确定这家公司还支持黑客马拉松文化了。”《CX》杂志引用的客户体验战略家把这描述得完美。在一个高信任文化中，零食和活动感觉像是慷慨。在低信任文化中，它们感觉像是诱饵。当你将免费零食放在按键监控、降薪和大规模裁员的旁边时，这些零食就成了侮辱。但这提出了一个很少被问到的问题：如果 Meta 正在公布创纪录的利润，为什么公司会处于这种状态？这里有一个事实应该重新定义整个故事。Meta 有望在 2026 年超越谷歌，成为全球最大的数字广告公司。在数字广告史上首次，谷歌将不会占据首位。根据 eMarketer 的预测，Meta 今年预计将创造 2435 亿美元的全球广告收入，高于谷歌的 2395 亿美元。Meta 将占据全球数字广告支出的 26.8%，与谷歌和亚马逊一起，这三家公司将占地球上每美元数字广告支出的 62.3%。这不是一家有问题的公司。这是一家坐拥人类历史上最宝贵的资产之一的公司，即跨 Facebook、Instagram 和 WhatsApp 的 40 亿用户的行为信号，而不是谷歌所拥有的搜索意图数据，那是人们主动想要某物时寻找的数据。Meta 拥有的是人们在没有搜索时会看的东西。是什么让人们停止滚动，他们点击了什么，他们买了什么，他们盯着某物多久。而真正驱动 Meta 广告主导地位的 AI 也极其有效。Advantage Plus 是 Meta 的自动化广告套件。它本质上接管了活动创建、受众定位和预算分配，用一个同时测试数百种创意变体并实时优化的系统取代了过去需要媒体购买团队的工作。它的年化运行成本高达 600 亿美元。Meta 的价值优化套件更进一步。它不只是寻找可能点击广告的人，而是识别最有可能长期花费最多钱的客户，本质上是在首次购买前预测终身价值。该产品在年收入跑道上超过了 200 亿美元，同比增长了更多。GEM 模型也是一个专门为短视频构建的推荐引擎。它决定向哪些用户展示哪些 Reels，以及何时插入一种感觉与滚动自然而非打断性的广告。它在单个季度就将 Reels 的广告转化率提高了 5%。然后是 Andromeda，它彻底颠覆了旧的广告手册。传统的数字广告通过定位人口统计数据、年龄、地点、兴趣和收入区间来运作。而 Andromeda 则使用广告商的创意本身作为主要信号，分析广告内容以找到最有可能将特定信息发送给该消息的人，而不是那些符合预定框的人。在 2026 年第一季度，广告展示量同比增长了 19%，而平均每次广告价格增长了 12%。这两个指标同时加速，这是广告健康最清晰的信号。这是一种专门针对专有数据进行训练的 AI，解决了 Meta 处于独特地位并可以主导的一个特定问题。它是圣杯。现在没有人能复制这一点。OpenAI 没有这些数据，Anthropic 没有这些数据，谷歌有不同的数据，虽然有价值，但从根本上是不同的。我个人认为广告是存在的最具创造性的学科之一，比绘画甚至诗歌都更有创意。在富有创造性的叙事、麦当劳传统和尖端数据分析之间形成完美的结合，产生能够以精确的时刻、以精确的信息触达人们的活动。Meta 拥有构建人类历史上最强大的广告 AI 的数据集，这个东西是专门为他们独特的竞争优势设计的，而不是在目前一个非常拥挤的市场中与通用聊天机器人竞争。相反，Meta 在 2026 年将投入 1250 到 1450 亿美元的资本支出，几乎是去年 722 亿美元的两倍，来构建通往 OpenAI、Anthropic 和 Google DeepMind 竞争的通用前沿 AI 模型。当马克·扎克伯格被追问所有这些钱在做什么时，他的回答是，正如我在这里引用的：“我的意思是，我们将在今年内推出新产品。我认为重要的是我们不仅仅是发布一件事，而是在构建很多新事物。”这在我看来听起来不像一个愿景。这更像是一个购物清单，也许吧。这里有一个合理的反驳，我想对它公平一点。Meta 需要一定程度的内部 AI 能力。完全依赖 Anthropic 等第三方提供商来获取内部工具会带来定价漏洞和竞争风险。构建你自己的基础设施是明智的，但构建利用你自身独特数据集、40 亿用户、行为信号和广告智能的专业化 AI，并构建通用前沿模型与整个存在都致力于那个单一目标的公司的头对头竞争，这之间有着巨大的区别。一个是战略性的，另一个是一个昂贵的身份危机。这就是这个模式变得无法忽视的地方。
Original English
During a company-wide town hall, somebody hijacked the live stream to scream profanities at Meta's AI leadership in front of thousands of employees. One of the presenters covered their face with their hands. Multiple employees told reporters they were actively hoping to be fired just to collect the severance package, which is 16 weeks of pay and 18 months of health cover. When your workforce is praying for the exit package, something has gone structurally wrong. In a follow-up memo, boss Zuckerberg acknowledged had done atrocious job explaining the vision behind the restructuring and admitted the company had undermined the trust employees placed in the value of their own expertise. And Meta's response to this near historic moral collapse, it was genius. Drum roll, please, everyone. They increased the budget for office snacks. They announced better break rooms, and Zuckerberg sent a brilliant memo urging employees to participate in a brilliant hackathon. Nothing really says, "I value your sacrifice as we harvest your keystrokes quite like an unlimited supply of organic cheddar bunnies, right? Sorry we took your career stability and your digital privacy, Kevin, but look, we added a sparkling water tap that dispenses blood orange flavor now. Happy hacking everyone, have fun." One employee wrote back that they were literally preoccupied with keeping the lights on for their team. Another said, "I'm not sure that this company supports a hackathon culture anymore." A customer experience strategist quoted by CX Magazine put it perfectly. In a high-trust culture, perks like snacks and events feel like generosity. In a low-trust culture, they feel like bait. When you place free snacks next to keystroke monitoring, pay cuts, and mass layoffs during record profits, the snacks become an insult. But this raises a question that doesn't get asked enough. If Meta is posting record profits, why is the company in this state at all? Here is a fact that should reframe this entire story. Meta is on track to overtake Google as the world's largest digital advertising company in 2026. For the first time in the history of digital advertising, Google will not hold the top spot. According to eMarketer, Meta is projected to generate $243.5 billion in global ad revenue this year, ahead of Google's $239.5 billion. Meta will command 26.8% of global digital ad spending, and together with Google and Amazon, the three companies will account for 62.3% of every digital advertising dollar spent on this planet. This is not a company in trouble. This is a company sitting on one of the most valuable data assets ever assembled, the behavioral signals of 4 billion users across Facebook, Instagram, and WhatsApp, not search intent data like Google has, which is what people look for when they actively want something. Meta has desired data, what people look at when they're not searching. What makes them stop scrolling, what they click, what they buy, how long they hover on something. And the AI that is actually driving Meta's advertising dominance is extraordinarily effective as well. Advantage Plus is Meta's automated ad suite. It essentially takes over campaign creation, audience targeting, and budget allocation, replacing what used to require a team of media buyers with a system that tests hundreds of creative variations simultaneously and optimizes in real time. It's running at a $60 billion annualized clip. Meta's value optimization suite goes a step further. Instead of just finding people likely to click an ad, it identifies the customers most likely to spend the most money over time, essentially predicting lifetime value before the first purchase. That product crossed $20 billion in annual revenue run rate, having more than doubled year over year. The GEM model is also a recommendation engine specifically built for short-form video. It determines which Reels to show to which users and when to insert an ad that feels native to the scroll rather than interruptive. And it lifted ad conversions on Reels by 5% in just a single quarter. And then there is Andromeda, which flipped the old advertising playbook entirely. Traditional digital advertising works by targeting demographics, age, location, interest, income bracket. Andromeda instead uses the advertiser's creative itself as the primary signal, analyzing the content of the ad to find the people most likely to send to that specific message rather than the people who fit a predetermined box. In Q1 2026, ad impressions increased 19% year on year, while average price per ad increased 12%. Both metrics accelerating simultaneously, which is the cleanest possible signal of advertising health. This is specialized AI trained on proprietary data, solving a specific problem that Meta is uniquely positioned to dominate. It is the holy grail. Nobody else can replicate this right now. OpenAI doesn't have this data, Anthropic doesn't have this data, Google has different data, valuable but fundamentally different. I personally believe that advertising is one of the most creative discipline that exists, more creative than painting or even poetry. The perfect marriage between creative storytelling, the Mad Men tradition, and cutting-edge data analytics producing campaigns that reach people at precisely the right moment with precisely the right message. Meta has the data set to build the most powerful advertising AI in human history, something specifically designed for their unique competitive advantage, not general-purpose chatbot competing in a very crowded market right now. And instead, Meta is spending 125 to 145 billion dollars in capital expenditure in 2026, nearly double the 72.2 billion dollars it spent last year, building generalized frontier AI models to compete with OpenAI, Anthropic, and Google DeepMind. When Mark Zuckerberg was pressed on what all this money is building, his answer was, and I'm quoting here, "I mean, we're going to roll out new products over the course of the year. I think the important thing is we're not just launching one thing, and we're building a lot of new things." It doesn't sound like a vision to me. This more sounds like a shopping list, maybe. There is a legitimate counter-argument here, and I want to be fair to it. Meta does need some degree of in-house AI capability. Being entirely dependent on third-party providers like Anthropic for internal tools creates pricing vulnerability and competitive risk. Building some of your own infrastructure is prudent, but there is a vast difference between building specialized AI that leverages your own unique data set, your 4 billion users, your behavioral signals, your advertising intelligence, and building generalized frontier models to compete head-to-head with companies whose entire existence is literally dedicated to that single objective. One is strategic, the other is an expensive identity crisis. And this is where the pattern becomes impossible to ignore. Before it was all about AI, it was all about the metaverse. Remember that? Zuckerberg renamed the entire company from Facebook to Meta. He declared that the future was virtual reality, allegedly. He poured resources into building a world of lifeless avatars and virtual conference rooms that approximately nobody asked for. The bill for that vision, according to Meta's own filings, Reality Lab has accumulated over $80 billion in cumulative operating losses since 2020. In Q1 2026 alone, Reality Labs lost $4.03 billion while generating just $402 million in revenue. The full year 2025 was $19.2 billion in operating losses on $2.2 billion in revenue. That is a business unit that costs nearly 10 times what it earns, running for half a decade now. I work in tech. I don't know a single person who engages with a metaverse. Everyone wore a lifeless avatar for like 5 minutes and then it was over. And you have to ask yourself, how do you even attract top engineering talent to work on a product with $80 billion in cumulative losses and negligible user adoption? In any other industry, that kind of track record is a massive red flag, but in tech, apparently, we call it Wednesday. Now, I'm not opposed to research and development, of course not. That's exactly what R&D departments exist for, exploring new frontiers with risk capital. The Ray-Ban meta glasses are generally gaining traction, I think. Hardware innovation is valuable, but if you've burned through 80 billion dollars and the signal from the market is just silence, at some point you have to be ruthless about cutting your losses. That is what good leadership does. Instead, the pattern appears to be repeating. Different costume, same behavior. The metaverse was the last shiny object, frontier AI is the current one, and the workforce absorbs the cost of every single pivot. Take the claudonomic saga. Earlier this year, Meta employees built an internal leaderboard that ranked the top 250 employees by how many AI tokens they consumed, which is the basic units of data that AI models process. It awarded titles like token legend, session immortal, and cash wizard. In 2025, Meta had made AI-driven impact a core performance expectation. Low token usage was informally interpreted as a layoff risk signal. So, employees did exactly what the incentive told them to do. They ran idle AI agents all day long to inflate their rankings. One engineer reportedly burned roughly $500,000 per month in token consumption. Meta employees collectively burned through 73.7 trillion tokens in 30 days. At Anthropic's public API prices, that's approaching 900 million dollars. We are talking about senior engineers here, by the way, the brightest minds of a generation, spending 8 hours a day asking an AI to write Shakespearean sonnets about JavaScript loops just so an automated leaderboard wouldn't flag them for termination. They didn't build Skynet, they built a $900 million campfire fueled by pure corporate panic. And then Meta reversed course, of course. The leaderboard was killed within 48 hours of the data leaking externally. The company told 6,000 employees they were being rationed on AI usage now. Bosworth wrote in a memo that nobody should be using AI tools just for the sake of using them now, and all motion is not progress, and token usage alone is not a measure of impact of any kind. This is textbook Goodhart's law. When a measure becomes the target, it ceases to be a good measure. Token usage was never the right metric. The right metric for a business is return on investment, ideally money, hard cash. Using token consumption as a proxy for productivity is like evaluating a chef by how many ingredients they use instead of whether the food is any good or is getting sold. And when you then punish people for maximizing the metric you told them to maximize, you haven't demonstrated strategic clarity, you've demonstrated strategic whiplash. The people who were ranking for maximum AI consumptions are now being told they were doing it wrong. Meta's executive suite is essentially just two toddlers in a trench coat trying to steer a rocket ship in opposite directions. I have a very good friend, an exceptional engineer, genuinely top talent person, who was hired by Meta in the second half of 2025. A few months later, he was made redundant. He got an amazing severance package, so he actually had a blast, but the absurdity of it, spending all of the money and time to recruit, onboard, and compensate a senior engineer only to fire them months later tells you everything about the strategic incoherence. Hiring is expensive. It takes months of recruiter time, interview panels, negotiation, relocation, onboarding. Then you throw it all away because the strategic direction shifted while the offer letter was still warm. My friend, thankfully, landed on his feet, but sadly not everyone does. It's kind of like owning the world's most profitable oil field, one that's literally about to overtake every other oil field on the planet, and deciding that your real passion is actually building rockets. The oil field is funding the rockets, sure, but nobody's tending the oil field with the attention that it deserves, and the people who work there keep getting reassigned to the rocket program, whether they like it or not. It turns out Mark's favorite part of playing rocket scientist isn't actually reaching space, it's hitting the eject button on the people sitting in the cockpit. Because since 2022, Meta has cut approximately 25,000 jobs. First, it was the pandemic correction, then the year of efficiency, then performance-based termination that hit employees with ratings of at or above expectations, then the Reality Labs gutting, then the 8,000 in May. There are more cuts planned for the second half of 2026, by the way, including a potential round in August. The stock portion of annual raises was trimmed by 5% in February 2026, on top of a 10% cut just the year before. Median total compensation slipped from $470,000 in 2024 to 388k, which is a 7% drop. One Instagram employee told Wired, "Everyone is unhappy. The only people who are not unhappy are literally executives." And then there is the story that haunts me. A former Meta employee who lost their job during the May layoffs was reportedly detained by US Immigration and Customs Enforcement almost immediately after being fired. Colleagues posted urgently on internal message boards, tagging Meta executives responsible for immigration issues. No public response came from leadership. The ex-employee is believed to be detained in El Paso, Texas. Their current status is still unknown. This is the human consequence of treating a workforce as a variable to be optimized, rather than as actual people who make decisions about their lives, their visas, their housing, their families based on the stability of employment that a trillion-dollar company implicitly promised. Meta CFO Susan Li framed it as a way to run a leaner operating model that helps offset the AI spending. But let's be precise about what actually is being offset. I looked at the filings, and Meta reported $56.31 billion in revenue and $26.8 billion in profit for Q1 2026 alone, with sales up 33% year-over-year, which is basically the company's fastest growth since 2021. The family of apps segment generated $102.5 billion in operating income in 2025. This has nothing to do with corporate survival and everything to do with Mark treating 25,000 real-world careers like pocket change for his sci-fi passion projects. And while the workforce absorbs the cost, the leadership experiences, and there's no delicate way to say this, somewhat different. Zuckerberg trains MMA three to four times a week with UFC champions. He sparred with light heavyweight champion Alex Pereira. He wears custom gold chains and a growable Force E watch valued at nearly $900,000. He attended the Met Gala with his wife. He parks his super yacht in the harbor of the city where he just got 1,400 jobs. He went from looking like a haunted Victorian doll who accidentally learned HTML to looking like a Roman emperor who discovered crypto and CrossFit. He's out here choking slamming UFC fighters into the canvas while his own HR department chokes slams the entire Pacific Northwest workforce. It's a vibe, sure, but the vibe is kind of supervillain origin story. Now, none of this is illegal, I'm aware. A CEO is allowed to live their life, and honestly, I am no stranger to being a little extra myself, minus the $300 million boat. But it's the contrast that kills me. When the gap between executive experience and employee experience becomes this visible, when the people building your products are hoping to get fired while the person who owns the company is wake surfing with UFC fighters, it stops being about optics and starts being about whether leadership has the capacity to understand what it's actually doing to its own organization. Jensen Huang, whose Nvidia chips are literally powering Meta's AI infrastructure, has called the AI layoffs narrative lazy and told executives who shrink their workforces in AI's name that they're out of imagination. The irony, of course, is that Meta doesn't even need a wild imagination to win. They just need to look at what's already staring them straight in the face. Because here is what I keep coming back to. Meta is not failing. That's what makes this so extraordinary. The advertising engine is arguably the most successful digital business ever built, and it's about to reach the top of the mountain, overtaking Google for the first time in history. The AI tools that are actually making the company money, Advantage Plus, Andromeda, the gem, the value optimization suite, are working beautifully. They are specialized, trained on proprietary data, and generating returns that most companies would build monuments to. But the leadership's attention is permanently elsewhere. It was the metaverse, now it's frontier AI models competing with OpenAI and Anthropic, tomorrow it's going to be something else, and every time the direction changes, the people pay. Fired, reassigned, drafted, monitored, pay cut, told to have fun at a hackathon, told to maximize their token usage, then told to stop maximizing their token usage. The contradictions don't cancel out. They compound. They become culture. Data is, after the basic necessities of life, the most valuable resource that exists. I might be biased there as a data scientist, but I really can't think of another necessity before GTA 6 comes out. Meta has more data than almost anyone. 4 billion users, decades of behavioral signals, the knowledge of what makes people stop, look, click, want, buy. That is a blessing that almost no other company on this planet possesses. And instead of dedicating the full force of the organization to making the most of it, building the most powerful, most specialized, most creatively sophisticated advertising AI ever conceived, the company keeps chasing the next identity. Metaverse company, frontier AI company, anything other than the advertising company it actually is. Maybe Bosworth was right that the leadership did an atrocious job of explaining the vision, or maybe the problem isn't the explanation. Maybe is that there are 17 competing visions running simultaneously, and nobody, not the CTO or the employees, not the investors, can tell which one is real. Blaming this on bad communications is a joke. It's a complete and total strategic train wreck. And unfortunately for Mark, a $300 million super yacht cannot outrun a self-inflicted corporate identity crisis. You can sprint as fast as you want, but eventually the cliff finds you anyway. But look, all of this is just a spectator sport unless you understand the actual machinery that broke Meta's culture, specifically the incentive structures behind that Cloutonomics disaster. If you want to know what companies should be measuring if they actually want AI to work instead of just burning a billion dollars on a digital leaderboard, you need to watch my video on token maxing. I break down the exact playbook for AI productivity right there. Click it. I'll see you there.核心洞察：数据与身份的博弈
在 GTA 6 发布之前，我真的无法想到另一个必要性。Meta 拥有比几乎任何人都多的数据。40 亿用户、数十年的行为信号、人们停止、观看、点击、想要、购买的原因。这是几乎地球上其他任何公司都拥有的祝福。因此，与其将整个组织的全部力量投入到最大化其价值中，构建有史以来最强大、最专业、最具创造性的广告 AI，该公司却一直在追逐下一个身份：元宇宙公司、前沿 AI 公司，任何与它实际所做的事情无关的东西。
Original English
I might be biased there as a data scientist, but I really can't think of another necessity before GTA 6 comes out. Meta has more data than almost anyone. 4 billion users, decades of behavioral signals, the knowledge of what makes people stop, look, click, want, buy. That is a blessing that almost no other company on this planet possesses. And instead of dedicating the full force of the organization to making the most of it, building the most powerful, most specialized, most creatively sophisticated advertising AI ever conceived, the company keeps chasing the next identity. Metaverse company, frontier AI company, anything other than the advertising company it actually is.战略失焦与身份危机
扎克伯格训练 MMA 三到四次每周与 UFC 冠军进行搏击。他曾与轻量级重量级冠军亚历克斯·佩雷拉（Alex Pereira）进行过对决。他穿着定制的黄金项链和价值近 90 万美元的可增长 Force E 手表。他与妻子一起参加了 Met Gala。他在自己刚获得 1,400 个工作机会的城市港口停放他的超级游艇。他从一个偶然学会 HTML 的、像被困在维多利亚时代的老娃娃一样的形象，变成了发现加密货币和 CrossFit 的罗马皇帝。他在这里一边用自己的 HR 部门猛烈抨击整个太平洋西北地区的员工，一边自己去打 UFC 选手的拳头。这是一种氛围，当然，但这种氛围有点超级反派的起源故事。现在，这些都不是非法的，我明白。CEO 可以过自己的生活，说实话，我并不陌生于做个“多余”一点的自己，扣除那艘价值 3 亿美元的船之外。但杀我的却是对比。当高管经验和员工经验之间的差距变得如此明显时，当那些正在构建你产品的员工希望被解雇，而公司所有者却在与 UFC 选手一起玩水上摩托艇时，这就不再是关于形象问题，而是关于领导层是否有能力理解它对自身组织的实际做了什么。
Original English
Zuckerberg trains MMA three to four times a week with UFC champions. He sparred with light heavyweight champion Alex Pereira. He wears custom gold chains and a growable Force E watch valued at nearly $900,000. He attended the Met Gala with his wife. He parks his super yacht in the harbor of the city where he just got 1,400 jobs. He went from looking like a haunted Victorian doll who accidentally learned HTML to looking like a Roman emperor who discovered crypto and CrossFit. He's out here choking slamming UFC fighters into the canvas while his own HR department chokes slams the entire Pacific Northwest workforce. It's a vibe, sure, but the vibe is kind of supervillain origin story. Now, none of this is illegal, I'm aware. A CEO is allowed to live their life, and honestly, I am no stranger to being a little extra myself, minus the $300 million boat. But it's the contrast that kills me. When the gap between executive experience and employee experience becomes this visible, when the people building your products are hoping to get fired while the person who owns the company is wake surfing with UFC fighters, it stops being about optics and starts being about whether leadership has the capacity to understand what it's actually doing to its own organization.最终结论：数据是唯一的救赎
Meta 并没有失败。这才是这一切如此非凡的原因。广告引擎可以说是迄今为止建立的最成功的数字业务，它即将登顶山峰，首次超越谷歌。真正为公司赚钱的 AI 工具——Advantage Plus、Andromeda（宝石）、价值优化套件——正在完美运行。它们是专业化的，基于专有数据进行训练的，产生了大多数公司会建立纪念碑的回报。但领导层的注意力永久地在别处。曾经是元宇宙，现在是与 OpenAI 和 Anthropic 竞争的前沿 AI 模型，明天将是其他东西，每当方向改变时，人们就付出了代价：被解雇、被重新分配、被征召、被监控、降薪、被要求参加黑客马拉松，被要求最大化他们的 token 使用量，然后又被要求停止最大化他们的 token 使用量。这些矛盾不会相互抵消。它们会累积。它们会成为文化。在基本生活必需品之后，数据是存在的最有价值的资源。我可能作为一名数据科学家会有偏见，但我真的无法想到 GTA 6 发布之前还需要另一个必要性。Meta 比几乎任何人都拥有更多的数据：40 亿用户、数十年的行为信号、人们停止、观看、点击、想要、购买的原因。这是地球上几乎没有其他公司拥有的祝福。因此，与其将整个组织的全部力量投入到最大化其价值中，构建有史以来最强大、最专业、最富创造性的广告 AI，该公司却一直在追逐下一个身份：元宇宙公司、前沿 AI 公司，任何与它实际所做的事情无关的东西。
Original English
Because here is what I keep coming back to. Meta is not failing. That's what makes this so extraordinary. The advertising engine is arguably the most successful digital business ever built, and it's about to reach the top of the mountain, overtaking Google for the first time in history. The AI tools that are actually making the company money, Advantage Plus, Andromeda, the gem, the value optimization suite, are working beautifully. They are specialized, trained on proprietary data, and generating returns that most companies would build monuments to. But the leadership's attention is permanently elsewhere. It was the metaverse, now it's frontier AI models competing with OpenAI and Anthropic, tomorrow it's going to be something else, and every time the direction changes, the people pay. Fired, reassigned, drafted, monitored, pay cut, told to have fun at a hackathon, told to maximize their token usage, then told to stop maximizing their token usage. The contradictions don't cancel out. They compound. They become culture. Data is, after the basic necessities of life, the most valuable resource that exists. I might be biased there as a data scientist, but I really can't think of another necessity before GTA 6 comes out. Meta has more data than almost anyone. 4 billion users, decades of behavioral signals, the knowledge of what makes people stop, look, click, want, buy. That is a blessing that almost no other company on this planet possesses. And instead of dedicating the full force of the organization to making the most of it, building the most powerful, most specialized, most creatively sophisticated advertising AI ever conceived, the company keeps chasing the next identity. Metaverse company, frontier AI company, anything other than the advertising company it actually is.最终总结：战略的混乱与数据驱动
扎克伯格对愿景解释得糟糕，或者问题不在于解释。也许是同时运行着 17 个竞争性愿景，而 CTO 或员工、投资者都无法分辨哪个是真实的。将此归咎于沟通不畅是个笑话。这是一个完全且彻底的战略灾难。不幸的是，对于马克来说，一艘价值 3 亿美元的超级游艇无法逃脱自我造成的企业身份危机。你可以跑得越快，但最终悬崖总会找到你。但看，这一切都只是观赏性运动，除非你理解导致 Meta 文化崩溃的实际机制，特别是背后的“流量经济学”灾难激励结构。如果你想知道公司应该衡量什么才能让 AI 生效，而不是仅仅为了一个数字排行榜烧掉十亿美元，你需要观看我的关于 token 调用的视频。我会在那里分解 AI 生产力的确切手册。点它。我们到时候见。
Original English
Maybe Bosworth was right that the leadership did an atrocious job of explaining the vision, or maybe the problem isn't the explanation. Maybe is that there are 17 competing visions running simultaneously, and nobody, not the CTO or the employees, not the investors, can tell which one is real. Blaming this on bad communications is a joke. It's a complete and total strategic train wreck. And unfortunately for Mark, a $300 million super yacht cannot outrun a self-inflicted corporate identity crisis. You can sprint as fast as you want, but eventually the cliff finds you anyway. But look, all of this is just a spectator sport unless you understand the actual machinery that broke Meta's culture, specifically the incentive structures behind that Cloutonomics disaster. If you want to know what companies should be measuring if they actually want AI to work instead of just burning a billion dollars on a digital leaderboard, you need to watch my video on token maxing. I break down the exact playbook for AI productivity right there. Click it. I'll see you there.📌 文中提及的人物和组织
人物: Mark Zuckerberg, Andrew Bosworth
公司/组织: Meta, Google, Amazon, OpenAI, Anthropic
产品/模型: Advantage Plus, Andromeda, Reality Labs, WhatsApp, Instagram, Facebook