内容来源：https://www.wired.com/story/why-normal-people-arent-using-ai-agents/

内容总结：

硅谷热捧AI智能体，普通用户为何“不买账”？

尽管硅谷正全力押注AI智能体（AI Agent）为未来趋势，但最新数据显示，这一前沿技术在普通公众中的普及率依然极低，与业界的热度形成鲜明反差。

近日，The Browser Company首席执行官乔什·米勒在社交媒体上的一番“吐槽”引发广泛共鸣。他指出，理论上技术已成熟到可以让AI智能体彻底改变工作与生活方式，但“大众根本不在乎”。这一言论击中了许多硅谷内部人士的困惑：为何公众对业界最新的痴迷无动于衷？

米勒在接受采访时强调，业内从未听到圈外人在谈论使用某个具体的智能体产品。尽管行业对未来充满兴奋，但值得停下来反思：“嗯，这是怎么回事？”

数据也印证了这种冷热不均。OpenAI上月表示，其Codex和ChatGPT Work智能体合计周活跃用户约1000万；Anthropic的同类产品也大致相当。然而，对比ChatGPT和Gemini动辄近10亿的月活用户，智能体的使用量几乎可以忽略不计。

分析认为，这对AI实验室而言是个严峻问题。当前，多数消费者使用生成式AI仍停留在信息检索和闲聊等简单任务上。而硅谷投入巨资训练出能力远不止于此的模型，智能体本应是变现这笔投资的关键，前提是用户真正愿意使用。

米勒的核心观点更为犀利：AI智能体目前更像是一种“技术”而非“产品”。他认为，智能体是行业自创的框架，并非用户需要的“东西”。正确的做法应是打造让用户打开电脑就能感到平静、专注和流畅的产品，至于背后是否由智能体驱动，用户根本无需知晓。

他以自家公司为例，其AI浏览器Dia最受欢迎的功能是“个性化晨间简报”——用户开机即可看到问候语、根据日程和邮件生成的待办清单，以及能带来愉悦感的随机内容。这个功能技术上正是由AI智能体实现的，但用户对此并无感知。

业界观察指出，当前多数智能体产品在设计时并未以消费者为中心。科技公司更热衷于展示模型最“炫技”的能力，如自主浏览网页或编写代码，而非针对用户真实渴望打造产品。这些能力未来或许能成为优秀功能，但现在更像是在展示“演示品”。

米勒将问题部分归因于AI行业的“群体思维”。他透露，在寻求收购期间与各大AI实验室领导会面时，几乎所有实验室都用科幻电影《她》来描述其产品愿景。这种缺乏多元观点和信念的趋同，可能限制了产品的想象力。

目前，AI智能体的最新趋势是赋予用户构建个人自动化软件的能力，如制作演示文稿、整理数据等。虽然这些工具比前代更实用，但真正有兴趣构建此类生产力工具的人群有限。行业若想实现主流普及，仍需更大胆的构想。

米勒呼吁，不要仅仅接受“AI智能体”这一叙事，而应回归本源，思考什么是能带来愉悦、实用且平易近人的好产品。他反问：“如果它看起来不像AI智能体，有什么关系？没有人‘使用’AI智能体，人们使用的是能解决实际问题的产品。”

中文翻译：

硅谷已经深信，AI智能体就是未来。在这个科技产业的心脏地带，人们正在为智能体设计支付系统，用它们来实现工作自动化，而最近，还在试图阻止它们入侵其他组织。但在现实世界中，大多数人都从未接触过AI智能体。这可能是因为科技公司还没有给人们一个充分的理由去使用它们。

“说句冒犯的话……难道不觉得有点疯狂吗，居然没有人在真正使用AI智能体，”The Browser Company的首席执行官乔什·米勒本周在X平台上发布了一篇疯传的帖子。“从理论上讲，技术已经准备好让AI智能体彻底改变我们工作和生活方式了……但可惜，普通大众根本不在乎。”米勒对智能体的顿悟在硅谷引起了广泛共鸣，那里的人们普遍困惑于一个事实：普通大众似乎对他们最新的痴迷之物毫不关心。米勒在周三接受我采访时说，他在欧洲家庭旅行途中半睡半醒时发出了那条帖子，但他坚持自己的观点。他说，整个行业需要专注于打造人们真正想要的智能体产品。

“在科技圈之外，我还没有听到任何一个人说起他们正在使用的智能体，”米勒说。“尽管我们作为行业，对前沿技术和递归自我改进感到兴奋和乐观，但还是值得停下来问一句：‘咦，怎么回事？’”

上个月，OpenAI表示，其Codex和ChatGPT Work智能体合计每周约有1000万用户。据与Anthropic关系密切的人士告诉我，其Claude Code和Cowork智能体的采用率也大致相当。相比之下，ChatGPT和Gemini等聊天机器人平均每月活跃用户都在10亿左右，智能体的用户量基本上可以忽略不计。

这对AI实验室来说是个问题。如今，大多数消费者使用生成式AI只是为了完成相当简单的任务，比如查信息和聊天。但硅谷已经花费了数十亿美元训练能做更多事情的模型，而智能体代表了从这笔投资中获利的一种方式——前提是人们真的会用它们。

这与我一直在AI业内人士中听到的更广泛的担忧相吻合。他们指出，尽管已经投入了大量资源来构建AI智能体，但到目前为止还没有人创造出杀手级的消费产品。智能体什么时候——如果有那么一天的话——才能迎来属于自己的ChatGPT时刻？

尽管米勒为人自嘲——他经常提到自己“只是”一个社会学专业出身的人——但我认为他的论点值得关注。他的最新创业公司开发了Arc，一款拥有忠实粉丝群体的网络浏览器，去年被Atlassian以6.1亿美元收购。他的第一家公司Branch在2014年被Facebook收购，米勒此前还曾在奥巴马总统任内担任白宫首位产品总监。

米勒更重要的观点不只是AI智能体尚未流行起来，而是它们更像是一种技术，而非独立的产品。

“没有人想要AI智能体，因为AI智能体本身并不是一个东西。那只是我们行业凭空发明的一个框架，用来统称某种东西，”米勒说。“让我们做一款产品，让你一打开笔记本电脑就感到平静、专注、进入心流状态。而我们能做到这一点是因为有一种叫‘调度器’的东西在调用各种工具——谁在乎呢？没有人需要知道这些。”

米勒有亲身经历来支持他的说法。他说，The Browser Company有史以来最受欢迎的功能，是其AI驱动的浏览器Dia中为用户提供的个性化晨间简报。当人们打开笔记本电脑时，会看到一个主页，上面有问候语、从日历和邮件中整理出来的每日待办清单，以及一些旨在带来愉悦感的随机小惊喜——比如一件艺术品。米勒说，从技术上讲，这个功能只有借助AI智能体才能实现，但用户不需要知道这一点。

米勒的这些论点出自一个卖AI驱动浏览器的人之口，听起来有些顺理成章。但我认为他说得对，如今的智能体产品在很大程度上并没有考虑到消费者。科技公司推出的是其模型能做的最令人印象深刻的事情——比如浏览网站或编写代码——而不是创造迎合客户需求的产品。这些能力也许有一天会变成很棒的功能，但现在，它们感觉更像是演示。

米勒将这个问题归咎于AI行业的群体思维。很多正在构建这些技术的人对它们完全痴迷，而且他们通常对产品的形态和感觉有一种特定的科幻愿景。去年夏天，当The Browser Company在探索被收购时，米勒说他会见了顶级AI实验室的许多领导者。

“这让我觉得很不可思议。我想除了一个实验室之外，其他每个实验室都用电影《她》来阐述他们的愿景，”米勒说，他指的是2013年斯派克·琼斯执导的那部电影。“我非常尊重任何说‘嘿，这是我眼中世界应该去的方向，我就要往那里去’的人。但我认为，在观点和信念上，确实缺乏多样性。”

AI智能体的最新趋势是赋予人们构建个人软件来自动化生活中部分事务的能力，比如制作演示文稿、整理数据集以及其他任务。我个人一直在大量使用ChatGPT Work和Claude Cowork，发现它们比前代产品更有用。但对构建自己的生产力工具感兴趣的人群体量有限，如果行业想要实现主流普及的目标，就需要一些更大的想法。

米勒说，他希望自己的信息能传达到OpenAI和Anthropic之外，更广泛地引起创始人和产品建设者的共鸣。他的目标是推动人们跳出固有思维，思考AI智能体产品可能是什么样子，然后将这些愿景变为现实。

“让我们不要只是接受AI智能体这套叙事，而是真正去质疑：什么是好的产品，什么是让人愉悦、有用且平易近人的工具？”米勒说。“谁在乎它看起来像不像AI智能体？没有人使用AI智能体。”

这是马克斯韦尔·泽夫《模型行为》新闻通讯的一期。在此阅读往期通讯。

评论
返回顶部

英文来源：

Silicon Valley is already convinced that AI agents are the future. In the heart of the tech industry, people are designing payment systems for agents, using them to automate their jobs, and most recently, trying to stop them from hacking into other organizations. But out in the real world, most people have never touched an AI agent. That might be because tech companies haven’t given them a good reason to.
“Hot take … isn’t it kinda crazy that nobody is really using AI Agents,” Josh Miller, CEO of The Browser Company, wrote in a viral X post this week. “Theoretically, the tech is ready for AI agents to totally transform how we work and live our lives … but alas the general public dgaf” (doesn’t give a fuck).
Miller’s agentic epiphany resonated with many in Silicon Valley, where people are broadly bewildered by the fact that the general public doesn’t seem to care about its latest obsession. Miller told me in an interview on Wednesday that he fired off his post while half asleep during a family trip to Europe, but he stands by it. He says the industry needs to focus on building agent products that people actually want.
“I just have not heard a single person outside of the tech community talk about an agent that they use,” Miller says. “As excited and optimistic as we as an industry may be about the frontier and recursive self-improvement, it's worth pausing and just saying, ‘Huh, what?’”
Last month, OpenAI said that its Codex and ChatGPT Work agents collectively have about 10 million weekly users. People close to Anthropic tell me that its Claude Code and Cowork agents are seeing similar levels of adoption. Compared to chatbots like ChatGPT and Gemini, which both have around a billion monthly active users on average, agents amount to basically a rounding error.
That’s a problem for the AI labs. Most consumers are using generative AI today for fairly simple tasks, such as looking up information and chatting. But Silicon Valley has spent billions training models that can do far more, and agents represent a way to capitalize on that investment—if people actually use them.
This tracks with broader concerns I’ve been hearing among AI insiders, who note that although they have poured considerable resources into building AI agents, no one has created a killer consumer product so far. When, if ever, will the ChatGPT moment happen for agents?
Despite Miller’s self-deprecating personality—he frequently noted he is “just” a sociology major—I think his arguments are worth paying attention to. His latest startup developed Arc, a web browser with a cult following, and it was acquired by Atlassian last year for $610 million. His first company, Branch, was bought by Facebook in 2014, and Miller also previously served as the White House’s first director of product under President Obama.
Miller’s bigger point is not just that AI agents haven’t caught on yet. It’s that they are more of a technology than a stand-alone product.
“No one wants AI agents, because AI agents aren't a thing. It is an invented frame made up by our industry to collectively refer to something,” says Miller. “Let's make a product that makes you calm, focused, and in flow right when you open your laptop. And the fact that we're able to do that because there's a thing called a ‘harness’ that calls up tools—who cares? Like, no one needs to know that.”
Miller has firsthand experience to back up his claims. He says The Browser Company’s most popular feature ever is a personalized morning briefing for users in its AI-powered browser, Dia. When people open their laptop, they receive a homepage with a greeting, a daily to-do list populated from their calendar and email, and some random tidbits designed to spark joy—such as a piece of art. Technically, Miller says, the feature was only made possible by an AI agent, but the user doesn’t need to know that.
Miller’s arguments are convenient coming from someone who sells an AI-powered web browser. But I think he’s correct that today’s agentic products are largely not designed with consumers in mind. Tech companies are shipping the most impressive things their models can do—such as navigating a website or writing code—rather than creating products tailored to their customers' desires. Those capabilities may become great features some day, but right now, they feel more like demos.
Miller blames the problem on groupthink in the AI industry. Many people who are building these technologies are totally obsessed with them, and they often have a particular sci-fi vision for how the products should look and feel. Last summer, when The Browser Company was exploring being acquired, Miller says he met with many of the leaders at top AI labs.
“It was wild to me. I think every single lab but one mentioned the movie Her as the way they articulated their vision,” says Miller, referencing the 2013 Spike Jonze film. “I have a lot of respect for anyone that says, ‘Hey, here's where I think the world should go, and I'm going to go there.’ But I think there is sort of a lack of diversity of opinions and convictions.”
The latest trend with AI agents is giving people the power to build personal software to automate parts of their life, such as building pitch decks, organizing datasets, and other tasks. I’ve personally been using ChatGPT Work and Claude Cowork a lot, and I’ve found them to be more useful than their predecessors. But there’s a limited pool of people interested in building their own productivity tools, and the industry needs some bigger ideas if it wants to reach the goal of mainstream adoption.
Miller says he hopes his message reaches beyond OpenAI and Anthropic and resonates with founders and product builders broadly. His goal is to push people to think outside the box about what AI agent products could look like, and then bring those visions to life.
“Let's not just accept this narrative of AI agents and actually question what are good products and tools that are joyful, useful, and approachable,” says Miller. “Who gives a shit if it looks like an AI agent? No one uses AI agents.”
This is an edition of Maxwell Zeff’s Model Behavior newsletter. Read previous newsletters here.
Comments
Back to top