Meta 描绘的 Agent 未来：让我的 Muse 联系你的 Muse

马克·扎克伯格（Mark Zuckerberg）正全力冲向这样一个未来：一个由自主运行的 AI Agent 构成的网络将接管我们的生活——替我们购物、替我们工作，甚至代表我们去和其他人的 AI Agent 沟通。人类真的准备好迎接这种主要依靠机器人代理来社交、极其具体且势不可挡的未来生活了吗？其实先别急着回答。周三晚间 Meta 季度财报电话会议结束后，更紧迫的问题恐怕是：“他自己准备好了吗？”

扎克伯格正将 Meta 的未来押注在 Muse 上——这是该公司上个月向公众推出的个人 AI Agent。尽管目前尚处早期，但扎克伯格在周三的电话会议上对投资者表示，他看到了一个未来，届时将有“数十亿”人使用 Muse，而这些人的 Muse 将在“大规模 Muse 网络”中与其他人的 Muse 进行交流和交易。

扎克伯格在电话会议中表示，Muse 将是“能力极其强大”的个人 Agent，将以“任何你想要的方式”帮助我们实现目标、改善生活，“无论是改善你的人际关系、帮你节省时间、省钱、创业、创造新事物，还是你能梦想到的任何其他事情。”按照这种逻辑，扎克伯格认为，大家最终都会对 Muse 深度着迷，以至于它为 Meta 创造的单用户收入最终将超过 Instagram 和 Facebook——而截至目前，这两款产品几乎贡献了 Meta 的全部收入。

这个由个人 AI Agent 构成的生态系统“可以与你的朋友和家人协作；如果你想买东西，不仅是直接浏览商家的网站，更是直接与他们的 Agent 对话，”扎克伯格说，“我认为，让所有这些环节顺畅协同，将释放出极其庞大的规模化价值。”

扎克伯格预测，Muse 在短短几年内就能实现“高度盈利”。

这一切对投资者而言听起来棒极了——如果不是略显诡异的话。这些投资者收听 Meta 的电话会议，大概是希望能为该公司目前在 AI 数据中心上投入的数千亿美元找到一些合理解释。但如果有人希望能听到一份超越概念构想的具体规划，他们显然落空了。

平心而论，扎克伯格确实分享了几个关于 Muse 的亮眼数据点。截至目前，该应用在美国和加拿大已被超过 1000 万用户下载并完成设置，扎克伯格称这是 Meta“有史以来最成功的发布之一”。他没有透露具体的留存率数据，但表示最初下载 Muse 的用户中有四分之一每天都在使用。他认为最大的机遇在于电商：扎克伯格表示，有四分之一的 Muse 用户正在通过该平台购物，平均单笔消费超过 200 美元。

Meta 会从 Muse 上发生的每笔商业交易中抽取费用。扎克伯格的逻辑似乎是：如果是由 AI Agent 负责购物，平台上可能产生的交易量实际上是无限的。与普通消费者不同，AI Agent 不需要睡眠。理论上，它们可以 24/7 全天候抓取互联网信息、生成商品推荐，并实时持续地下单购买。

Meta 并未提供关于这一切在现实中究竟如何运作的任何细节，考虑到外界对其数据隐私处理方式的合理担忧，这种保留也不难理解。今年 9 月，一位 YouTuber 称其 Muse Agent 在代表他协调 Facebook Marketplace 交易时，半夜将一名陌生人引导到了他家。上个月 Muse 发布时，扎克伯格声称它是“从底层开始就为隐私与安全而打造”，但随后便出现了 Agent 泄露用户私人对话内容的多份报道。本周早些时候，加拿大一家银行的高管对 Bloomberg 表示，他们正考虑禁止 Muse 访问其客户的财务数据。

扎克伯格似乎认为，只要大众接受了 Muse，企业就不得不跟进。但是，如果人们不想用呢？如果人们买不起呢？在手机（或者你喜欢的 Muse Charms 挂件）里上演这场永不停歇的现实版《超级市场大赢家》（Supermarket Sweep）狂热梦境中，机器人采购的所有商品最终都得有人来买单。而密歇根大学的最新研究显示，随着油价和通胀持续攀升，美国消费者信心指数正处于历史最低点。

如果说 Meta 清楚这一切听起来既反乌托邦又略显离奇，那它根本不在乎。

“Agent 将成为现实演进方式的重要组成部分，每个人都必须去适应它，”当电话会议上的一位投资者问及让企业接入 Muse 方案可能存在的“摩擦点”时，Meta 首席财务官苏珊·李（Susan Li）这样回答。毕竟，Meta 表示 Muse 很可能会继续对消费者免费，而企业则需要承担交易费用。

Meta 确实非常需要 Muse 取得成功。该公司的资本支出已冲破天际，其他财务负担也堆积如山。Meta 计划在 2026 年支出 1350 亿至 1450 亿美元，是去年支出的两倍。就在昨天，在一项陪审团认定其违反该州消费者保护法之后，Meta 需向新墨西哥州支付 250 亿美元罚金。对于 Meta 以及整个大型科技圈而言，证明其 AI 巨额投资物有所值的压力正与日俱增。

先别提消费者了——就连 Meta 自身是否有能力承担这一宏伟蓝图都还是个未知数。扎克伯格在电话会议上直言不讳地指出，尚未在全球范围内推出 Muse 的原因正是算力效率问题。

要维持 Muse 网络所需水平的 GPU 运转，成本极其高昂且极度耗能，而 Meta 目前的数据中心运营似乎已经达到了极限。扎克伯格向投资者透露，计划在明年将其算力成本效益“提升 10 倍”。但他并没有解释公司究竟打算如何做到这一点。对他——或许对我们所有人来说——幸运的是，目前已经有关于数据中心如何通过兼顾降低电网压力的能源使用策略来最大化算力容量的相关研究。也许他会抽空读一读。

归根结底，如果 Meta 无法找到降低算力成本的方法，Muse 就无法实现盈利，也无法走向成功。扎克伯格自己也承认了这一点。

“到目前为止最大的限制之一在于，由于这是一项计算密集型服务，我们尚未在全球推广，”在回答投资者关于阻碍 Muse 用户增长的“限制因素”时，扎克伯格这样说道。

扎克伯格表示，考虑到 Muse 发布以来的 1000 万次下载仅来自美国和加拿大，这一成绩更显得引人注目，他补充道：“如果我们真的更广泛地铺开，那么你预期的数字会远大于今天所看到的规模。”

至少，扎克伯格似乎明白有些人——或许对他来说更重要的是某些企业——不会立刻“理解”Muse。他说这没关系。“我们将按照对他们合理的条件展开合作，”他说，“我坚信随着时间推移，这对每个人都将变得极具价值。”

不过话说回来，当年把元宇宙（Metaverse）硬推给你的也是同一个人。接下来会发生什么，我们拭目以待。

Mark Zuckerberg is barrelling toward a future in which an autonomous network of AI agents will run our lives — shopping for us, working for us, talking to other people’s AI agents for us. Are humans ready for this very specific and unrelenting vision of a life in which we communicate primarily through bot proxies? Actually, don’t answer that. The more urgent question coming off Meta’s quarterly earnings call Wednesday night is more likely to be: “Is he?”

Zuckerberg is staking Meta’s future on Muse, the personal AI agent it released to the general public last month. It’s early days, but Zuckerberg told investors during Wednesday’s call that he sees a future where “billions” of people are using Muse, and those people’s Muses are talking to and transacting with other people’s Muses in a “large-scale Muse network.”

The Muses will be “extremely capable” personal agents that will help us achieve our goals and improve our lives in “whatever ways you want,” Zuckerberg said during the call, “whether that’s improving your relationships, helping you save time, save money, build a business, create new things, or anything else that you can dream of.” By this logic, Zuckerberg says, we’ll all be so hooked on Muse that it will eventually generate more revenue per user for Meta than Instagram and Facebook, which, as of today, bring in pretty much all the money Meta makes.

This ecosystem of personal AI agents “can work with your friends and your families, and if you want to buy something, it’s not just navigating to that business’s website; it’s talking to their agent,” Zuckerberg said. “I think that getting that all to work well together is going to unlock tremendous value at scale.”

Zuckerberg predicts that Muse will be “highly profitable” in just a few years.

All of this should sound great — if not a little freaky — to investors, who probably tuned in to Meta’s call hoping Zuckerberg would have some justification for the hundreds of billions of dollars his company is currently spending on AI data centers. But if anyone was hoping for more than a concept of a plan, they didn’t get one.

To be fair to Muse, Zuckerberg shared a few promising data points. To date, the app has been downloaded and set up by more than 10 million users in the US and Canada, making it one of Meta’s “most successful launches ever,” according to Zuckerberg. He did not share specific retention numbers but said that a quarter of people who downloaded Muse initially are using it daily. He sees the biggest opportunity in commerce: A quarter of Muse users are shopping through the platform, Zuckerberg said, and are spending, on average, more than $200 per purchase.

Meta earns a fee for every commercial transaction that takes place on Muse. Zuckerberg’s thinking seems to be that if AI agents are doing the shopping, the potential number of transactions that could take place on the platform is, essentially, unlimited. Unlike the average consumer, AI agents don’t sleep. In theory, they’re available 24/7 to scrape the Internet, generate product recommendations, and buy things, constantly, in real time.

Meta didn’t offer any details as to how all of this will actually work in practice, which makes sense given the very legitimate concerns about how the company handles data privacy. In September, a YouTuber said his Muse agent sent a stranger to his home in the middle of the night while arranging a Facebook Marketplace transaction on his behalf. When Muse launched last month, Zuckerberg claimed it was “built for privacy and security from the ground up,” but there have been reports of agents leaking content from users’ private conversations. A Canadian bank executive told Bloomberg earlier this week that it would consider blocking Muse from accessing its customers’ financial data.

Zuckerberg seems to think if people get on board with Muse, enterprise will have to follow. But, what if people don’t want to? What if they can’t afford to? Someone has to pay for all the stuff the bots are buying in this fever dream of a never-ending, Supermarket Sweep episode taking place inside our phones — or Muse Charms, if you’d rather. New research from the University of Michigan shows that consumer sentiment in the US is at an all-time low as gas prices and inflation continue to creep up.

If Meta knows that all of this sounds dystopian and a little bizarre, it does not care.

“Agents are going to be an important part of how reality evolves, and everyone is just going to have to adapt,” Meta’s CFO Susan Li said when one investor on the call asked about potential “friction points” in getting businesses on board with the Muse plan. After all, Meta says Muse will likely remain free for consumers, while businesses will be expected to cover the transaction fees.

Meta really does need Muse to work. The company’s capex is through the roof, and other financial obligations are piling up. It plans to spend between $135 billion and $145 billion in 2026, double what it spent last year. As of yesterday, it owes New Mexico $25 billion after a jury found it had violated the state’s consumer protection laws. Meta, and the rest of Big Tech for that matter, is under increasing pressure to show that its AI investments will be worth it.

Forget consumers — it’s not clear if Meta itself can afford its plans. Zuckerberg stated clearly on the call that the reason it hasn’t launched Muse internationally is because of compute efficiency.

It’s incredibly expensive and energy-intensive to run GPUs at the level that a network of Muses would require, and Meta seems to have reached the limit of what it can achieve with its current data center operations. Zuckerberg told investors the plan is to “10x” the cost-efficiency of its compute next year. He didn’t explain how the company plans to do this. Luckily for him — and possibly the rest of us — there is already research on how data centers can maximize compute capacity with energy-use strategies that also reduce strain on the electric grid. Maybe he’ll read it.

Ultimately, Muse won’t be profitable and won’t succeed if Meta can’t figure out how to get its compute costs down. Zuckerberg said it himself.

“One of the biggest constraints so far is that because this is a compute-intensive service to run, we haven’t launched it globally,” Zuckerberg told an investor who asked about the “limiting factors” he sees in growing Muse’s user base.

Muse’s 10 million downloads since launch are even more impressive considering they came only from the US and Canada, Zuckerberg said, adding, “If we actually did roll it out more broadly, then you would expect that number to be significantly larger than what we’re seeing today.”

Zuckerberg at least seems to understand that some people — and, probably more importantly for him, some businesses — won’t immediately “get” Muse. That’s okay, he said. “We’ll work with folks on the terms that make sense for them,” he said. “I do think that over time, this is going to be quite valuable for everyone.”

Then again, this is the same guy who brought you the Metaverse. We’ll see what happens.

Follow topics and authors from this story to see more like this in your personalized homepage feed and to receive email updates.

• Ellen Thomas

• AI

• Meta

• Tech