内容来源：https://www.wired.com/story/i-finally-found-an-ai-agent-worth-the-risk/

内容总结：

离开《连线》杂志的全职工作后，我做的第一件事之一，就是让Claude Cowork接入我的邮箱和日历。在报道人工智能多年之后，我对这项技术能为我做什么充满好奇。我曾把Cowork想象成一个能力超强的数字助手，结果却失望地发现，它更像是在和一个普通聊天机器人互动，只不过多了些工具而已。我想让它做什么？我毫无头绪。

后来，Instinct——一款仅限邀请、通过iMessage和WhatsApp与用户沟通的AI智能体——开始在湾区火爆起来。它能连接你的邮箱、日历和通讯应用。可以把它理解为面向普通人的OpenClaw。该公司于2月启动内测，据报道正在洽谈在已筹集的3.5亿美元基础上再融10亿美元，估值将达100亿美元。

Instinct备受瞩目的同时，Meta也因其本月早些时候推出的AI助手Muse而意外走红。截至发稿，据第三方估算，Muse是苹果App Store最受欢迎的自由应用，下载量超过90万次。X上的评论人士似乎对它 genuinely 兴奋，尽管它上线时存在一个严重的安全漏洞。我并不打算把大量个人信息交给Meta，但如果Muse能让我保持登录彭博社，或者帮我退订从2017年起每周给我发邮件的某瑜伽馆，我会重新考虑。

智能体挑衅者

使用AI智能体处理一切事务的人，与从未尝试过的人之间的鸿沟从未如此之大。这在一定程度上解释了为什么科技CEO们对数据中心的强烈反对声浪大多措手不及。如果你认为智能体能自动化人们大部分的行政琐事并大幅提升生产力，那么建设数据中心的成本和干扰或许看起来是可以接受的取舍。但如果你只是把聊天机器人当作高级谷歌来用，这笔交易恐怕就没那么有吸引力了。

正如科技记者Jasmine Sun所写，最根本的问题在于“大多数人的问题并非软件形态的，而且即使是的，大多数人也注意不到。”AI智能体和聊天机器人一样，仍然需要用户大量的输入和指导。如果你知道自己想要什么——作业辅导、虚拟女友——这些工具会热情地提供。但如果你不知道，那么，欢迎来到永久底层。

我迫切想避免这种命运。我可能不够“智能体化”，但我自认为是个行动派。我肯定能找出生活中一个软件形态的问题。我之前用Claude生成过几张自由职业发票，省了大约10分钟。不错，但算不上改变生活。我决定试试Instinct。

如果你和AI研究者聊Instinct，你会频繁听到“形态”这个词。Instinct在形态上做对了——没有开放式文本框，只有iMessage、WhatsApp和几条有用的提示。这一点的重要性怎么强调都不为过。你不必自己琢磨如何让聊天体验更有用，而是在和一个智能体发短信，它主动建议能为你做什么，然后真正去做。而且与这个领域的其他智能体不同，它不会发太多消息。我试过一款竞品智能体Lindy，一周后删掉了，因为它每天早上给我发消息，还未经我批准就闯入Zoom会议。

第一个项目，我把去意大利威尼斯出差的行程给了Instinct。我让它根据我的位置预订餐厅，包括通过WhatsApp联系那些隐蔽的小店。成功了！餐厅预订似乎是许多对智能体好奇的人的入门毒品，我也不例外。

接下来，我给了Instinct一个更复杂的任务。我已经和妹妹订好了11月去纽约的旅行，这时《连线》让我参加WIRED World Fair，活动在迈阿密，正好是前一天。不巧的是，我订的是不可改签的特价票，如果错过第一段行程，整个行程会自动取消。

Instinct能取消我的行程并拿到退款吗？出乎意料的是，它可以。智能体发现阿拉斯加航空把我的航班提前了90分钟——这一点我完全没注意到——这让我有资格获得全额退款。它取消了机票，为我省了大约550美元，并重新预订了回旧金山的单程票。不知道你们怎么想，但我开始感受到AGI了。

不安的本能

说到这里，我必须停下来承认使用Instinct的风险。风险很多。有人报告说，当他们试图将智能体从邮箱断开时，发现它仍然保留着收件箱的副本。一位风投说，在机器人尝试预订时每小时 ping Resy 的API约200次后，他被Resy封禁了。另一位科技投资者说，在判断Instinct极易被钓鱼后，他删除了该应用。

Instinct的服务条款相当宽泛，允许公司至少使用部分对话来训练其AI模型。我还经历了一次烦人的（虽然轻微的）事故：有一天，Instinct取消了一个严重延迟的DoorDash订单，迫使我损失了64美元。（我明确告诉它只有能退款时才取消。）这让我意识到Instinct非常非常擅长另一件事：大量道歉，但不赔偿损失。

Instinct公司未回应《连线》的置评请求。

Instinct感觉像一个略显混乱的个人助理，既然我从未有过真人助理，我就继续用着。上周它告诉我“不要打开”一封朋友邀请我参加后院烧烤的邮件——结果那是钓鱼诈骗。智能体是安全噩梦。我懂！但我很忙。我是妈妈。现在什么都像安全噩梦。目前，我愿意冒这个险。

中文翻译：

我离开《连线》的全职工作后做的第一件事，就是让Claude Cowork接入我的邮箱和日历。报道了多年人工智能之后，我对这项技术能为我做什么充满好奇。我原本以为Cowork会是一个能力超强的数字助理，结果却发现它更像是在跟一个普通聊天机器人对话，只不过多了一些工具而已。我想让它做什么？我完全没有头绪。

后来，Instinct——一款仅限邀请使用的AI代理，通过iMessage和WhatsApp与用户沟通——开始在湾区火了起来。它能连接你的邮箱、日历和通讯应用。可以把它理解为普通人版的OpenClaw。据The Information报道，这家公司于二月份开启内测，据称正在洽谈在已筹集的3.5亿美元基础上再融资10亿美元，估值将达到100亿美元。

Instinct之所以备受瞩目，恰逢Meta凭借Muse迎来了意料之外的一波热度——Muse是这家科技巨头本月早些时候推出的AI助手。截至本文撰写时，据第三方估算，Muse是苹果App Store上最受欢迎的免费应用，下载量已超过90万。X上的评论圈似乎对它充满热情，尽管据Ars Technica报道，它上线时存在一个严重的安全漏洞，可能“让攻击者在受害者的Mac上为所欲为”。我不打算把大量个人信息交给Meta，但如果Muse能让我保持登录彭博社，或者帮我退订Hot Yoga São Paulo——这家瑜伽馆从2017年起每周给我发一封邮件——我会重新考虑。

代理者的挑衅

把AI代理用到极致的人和从未尝试过的人之间的鸿沟，从未如此之大。这在一定程度上解释了为什么科技CEO们对数据中心的反弹普遍措手不及。如果你认为代理能自动化人们大部分的行政琐事并大幅提升生产力，那么建设数据中心带来的成本和干扰或许看起来是可以接受的代价。但如果你只是把聊天机器人当作一种高级版的谷歌搜索，那这笔账可能就没那么划算了。

正如科技记者Jasmine Sun所写，最根本的问题在于“大多数人的问题并不是软件形态的，而且大多数时候即使问题是软件形态的，他们也不会注意到。”AI代理和聊天机器人一样，仍然需要用户大量的输入和指导。如果你知道自己想要什么——帮忙做作业、一个虚拟女友——这些工具会热情地满足你。但如果你不知道，那欢迎来到永久底层。

我急切地想避免这样的命运。我可能不够“代理化”，但我自认为是个行动派。我肯定能在生活中找到一个软件形态的问题。我之前用过Claude生成几张自由职业发票，省了我大约10分钟。挺好，但谈不上改变生活。我决定试试Instinct。

如果你跟AI研究人员聊Instinct，你会频繁听到“形态参数”这个词。Instinct在形态上做对了——没有开放式文本框，只有iMessage、WhatsApp和几个有用的提示。这一点的重要性怎么强调都不为过。你不需要自己去琢磨如何让聊天体验更有用，而是在和一个代理发短信，它会主动建议能为你做什么，然后真的去做。而且这个领域里的其他代理不同，它不会发太多消息。我试了一款竞品代理Lindy，用了一周就删了——它每天早上给我发消息，还未经我批准就闯入Zoom会议。

第一个项目，我把去意大利威尼斯出差的行程给了Instinct。我让它根据我的位置预订餐厅，包括通过WhatsApp联系那些藏在巷子里的小馆子。成功了！餐厅预订似乎是许多对代理好奇的人的入门毒品，我也不例外。（值得一提的是，虽然Muse也能帮你打电话预订餐厅，但据404 Media报道，其中一些电话实际上是呼叫中心的人工打的。）

接下来，我给Instinct布置了一个更复杂的任务。我已经和姐姐订好了十一月去纽约的旅行，这时《连线》让我参加WIRED World Fair，活动在迈阿密，正好是前一天。不幸的是，我订的是特价票，意味着不能改签，而且如果我错过第一段行程，整个订单会自动取消。

Instinct能帮我取消行程并拿到退款吗？出乎意料的是，它可以。代理发现阿拉斯加航空把我的航班提前了90分钟——这个变化我完全没注意到——这让我有资格获得全额退款。它取消了机票，帮我省了大约550美元，并重新预订了我回旧金山的单程票。不知道你们怎么想，但我开始感受到AGI了。

不自然的直觉

说到这里，我必须停下来承认使用Instinct的风险。风险很多。有人反映，当他们试图将代理与邮箱断开连接时，发现它仍然保留着收件箱的副本。一位风险投资人说，他在Resy上被封号了，因为机器人在尝试预订时每小时 ping 其API大约200次。另一位科技投资者说，他在确定Instinct极容易被钓鱼攻击后删除了该应用。

Instinct的服务条款范围相当广，允许公司至少使用部分对话来训练其AI模型。我还遇到了一次令人恼火（虽然不算严重）的失误：有一天，Instinct取消了一个严重延迟的DoorDash订单，害我损失了64美元。（我明确告诉过它只有能退款时才取消。）这让我注意到Instinct另一件非常非常擅长的事：诚恳道歉，但绝不主动赔偿损失。

Instinct公司——与其旗舰AI代理同名——未回应《连线》的置评请求。

Instinct感觉像一个有点混乱的个人助理，既然我从来没有过真正的人类助理，我就继续用它了。上周它告诉我“不要打开”一封朋友邀请我参加后院烧烤的邮件——结果那是一个钓鱼骗局。代理是安全噩梦。我懂！但我很忙。我是妈妈。现在什么感觉都像安全噩梦。目前，我愿意冒这个险。

本文是《模型行为》通讯的一期。点击此处阅读往期通讯。

评论

返回顶部

英文来源：

One of the first things I did when I left my full-time job at WIRED was give Claude Cowork access to my email and calendar. After reporting on artificial intelligence for years, I was curious about what the technology could do for me. I had imagined Cowork as a hyper-capable digital assistant, but was let down to discover that it was more like interacting with a regular chatbot, albeit one with extra tools. What did I want it to do? I had no idea.
Then Instinct, an invite-only AI agent that communicates with users through iMessage and WhatsApp, started popping off in the Bay Area. It connects to your email, calendar, and messaging apps. Think of it as OpenClaw for normies. The company, which launched in private beta in February, is reportedly in talks to raise $1 billion on top of the $350 million it’s already raised, bringing its valuation to $10 billion, according to The Information.
Instinct’s moment in the spotlight comes as Meta is experiencing its own unexpected burst of popularity thanks to Muse, an AI assistant the tech giant launched earlier this month. As of writing, Muse is the most popular free app in Apple’s App Store, with more than 900,000 downloads, according to third-party estimates. The commentator class on X seems genuinely excited about it, despite the fact that it rolled out with a serious security vulnerability that would have “let attackers do ‘whatever’ they wanted on a victim’s Mac,” according to Ars Technica. I’m not about to give Meta a ton of my personal information, but if Muse can keep me logged into Bloomberg or unsubscribe me from Hot Yoga São Paulo, which has been emailing me once a week since 2017, I’ll reconsider.
Agent Provocateur
The divide between people who use AI agents for everything and those who have never tried one has never been wider. It explains, in part, why tech CEOs were largely caught off guard by the data center backlash. If you think agents can automate the majority of people’s administrative drudgery and turbocharge their productivity, the costs and disruptions associated with building data centers might look like an acceptable trade-off. But if you’re using chatbots as a fancy form of Google, that deal is probably much less appealing.
The overarching problem, as technology journalist Jasmine Sun writes, is that “most people’s problems are not software-shaped, and most won’t notice even when they are.” AI agents, like chatbots, still require a lot of input and direction from users. If you know what you want—help with your homework, a virtual girlfriend—the tools are eager to provide. But if you don’t, well, welcome to the permanent underclass.
I was eager to avoid such a fate. I may not be agentic, but I like to think I’m action-oriented. Surely I could identify a software-shaped problem in my life. I had used Claude previously to generate a few freelance invoices, saving me approximately 10 minutes. It was nice, but not life-changing. I decided to give Instinct a try.
If you talk to AI researchers about Instinct, you’ll hear the phrase “form factor” a lot. Instinct got the form factor right—no open text box, just iMessage, WhatsApp, and a few helpful prompts. It’s hard to overstate how important this is. Rather than trying to figure out for yourself how to make the chat experience more useful, you’re texting with an agent that’s suggesting things it can do for you, and then actually doing them. And unlike other agents in this space, it doesn’t text too much. I tried a competing agent called Lindy for a week and deleted it after it texted me each morning and started popping into Zoom meetings without my approval.
For its first project, I gave Instinct my itinerary for a work trip to Venice, Italy. I had it book reservations based on where I was going to be, including WhatsApping hole-in-the-wall spots. Success! Restaurant reservations appear to be a gateway drug for many agent-curious people, and I was no exception. (It’s worth noting that while Muse can also call restaurants to book reservations on your behalf, some of those calls are reportedly made by humans in call centers, according to reporting in 404 Media).
Next, I gave Instinct a more complicated task. I had already booked a trip to New York in November with my sister when WIRED asked me to attend the WIRED World Fair, which was taking place in Miami the day before. Unfortunately, I’d booked a saver fare, meaning I couldn’t move the ticket, and if I missed the first leg of the trip it would automatically cancel the whole thing.
Could Instinct cancel my trip and get a refund? To my surprise, it could. The agent detected that Alaska had moved my flight 90 minutes earlier, a fact I’d completely missed, which qualified me for a full refund. It canceled my ticket, saving me roughly $550, and rebooked my one-way return to San Francisco. I don’t know about you guys, but I’m starting to feel the AGI.
Unnatural Instincts
This is where I have to pause and acknowledge the risks of using Instinct. There are a lot. People have reported that when they tried to disconnect the agent from their email, they found out it was retaining a copy of their inboxes anyway. One venture capitalist said he was banned from Resy after the bot pinged its API roughly 200 times per hour while trying to make a reservation. Another tech investor said he deleted the app after determining that it would be trivially easy for Instinct to be phished.
Instinct’s Terms of Service are pretty far-reaching, allowing the company to use at least some of your conversations to train its AI models. I also experienced an annoying (if minor) snafu: One day, Instinct canceled a DoorDash order that was seriously delayed, forcing me to forfeit $64. (I’d explicitly told it to cancel only if I could get a refund.) This alerted me to another thing Instinct is very, very good at: profusely apologizing without offering to pay for its mistakes.
The company Instinct, which shares the same name as its flagship AI agent, did not respond to requests for comment from WIRED.
Instinct feels like a mildly chaotic personal assistant, and since I’ve never had a real human assistant, I’m sticking with it. Last week it told me “NOT TO OPEN” an email from a friend inviting me to a backyard BBQ—it turned out to be a phishing scam. Agents are a security nightmare. I get it! But I’m busy. I’m a mom. Everything feels like a security nightmare these days. For now, I’m willing to take the risk.
This is an edition of the Model Behavior newsletter. Read previous newsletters here.
Comments
Back to top