The Remote Already Exists: What “Click” Got Right About Agentic AI

万能遥控器已成现实：《人生遥控器》如何精准预言了智能体 AI

I rewatched “Click” recently, the 2006 Adam Sandler movie that everyone remembers as a dumb comedy and almost nobody remembers as the quiet tragedy it actually is. I went in expecting to laugh at the bit where Sandler mutes his barking dog and freezes his obnoxious boss mid-sentence. I came out thinking about my email inbox, my calendar, and the AI agents I now have permission to run on my own laptop. 最近我重温了 2006 年亚当·桑德勒主演的电影《人生遥控器》（Click）。大家都记得它是一部无脑喜剧，却几乎没人记得它其实是一部沉静的悲剧。我本以为会对着桑德勒静音吠叫的狗、让讨厌的老板在说话时暂停的桥段大笑，但看完后，我却开始思考我的电子邮箱、日程表，以及那些我现在获准在笔记本电脑上运行的 AI 智能体。

The movie is twenty years old. It feels like it was written last week as a warning aimed squarely at 2026. A quick note before we go further. I am going to describe the film’s premise and one turn that happens partway through, because the whole argument of this piece depends on it. Think of it as slightly more than what the trailer gives away. I will not reveal the specific things that happen to Michael in the back half of the movie, and I will not touch the ending. If you have never seen it and want to go in completely cold, watch it first and come back. It holds up better than you remember. 这部电影已经上映二十年了，但它读起来就像是上周刚写好的，专门针对 2026 年的警示录。在深入探讨之前先说明一下：我将描述电影的前提设定以及中途的一个转折，因为本文的所有论点都建立在此基础上。这比预告片透露的内容稍微多一点。我不会剧透迈克尔在电影后半段的具体遭遇，也不会触及结局。如果你从未看过这部电影，且想保持完全的未知感，请先看完再回来。它比你记忆中要耐看得多。

If you are still here, this is the setup. Michael Newman is an overworked architect played by Adam Sandler. He loves his wife Donna and his two kids, but he is drowning at work and constantly choosing the office over the dinner table. One night, fed up with the pile of remote controls on his coffee table, he drives to Bed Bath & Beyond looking for a universal remote. He wanders into a back room marked “Beyond” and meets a strange clerk named Morty, played by Christopher Walken, who hands him a prototype remote for free. The catch, which Morty mentions almost casually, is that it can never be returned. 如果你还在看，那我就介绍一下背景。迈克尔·纽曼是一位由亚当·桑德勒饰演的过度劳累的建筑师。他深爱妻子唐娜和两个孩子，却深陷工作泥潭，总是为了加班而牺牲家庭晚餐。一天晚上，他厌倦了茶几上堆积如山的遥控器，开车去 Bed Bath & Beyond 寻找万能遥控器。他误入了一个标有“Beyond”（超越）的后室，遇见了一位名叫莫蒂的古怪店员（克里斯托弗·沃肯饰），对方免费给了他一个遥控器原型。莫蒂几乎随口提了一句：这东西一旦拿走，就永远无法退货。

The remote turns out to control not just the television but Michael’s entire life. He can pause reality, mute people, turn up the volume, change the language someone is speaking, and most importantly, fast forward through the parts he does not want to sit through. At first this is a gift. He skips traffic. He skips a cold. He skips arguments with his wife. He fast forwards through a boring family dinner and zooms straight to the good part. For about fifteen minutes of screen time, the remote looks like the best thing that ever happened to him. Then the movie turns, and it turns hard. 事实证明，这个遥控器不仅能控制电视，还能控制迈克尔的整个人生。他可以暂停现实、静音他人、调大音量、切换语言，最重要的是，他可以快进掉那些他不想经历的时刻。起初，这简直是天赐之物。他跳过了交通拥堵，跳过了感冒，跳过了与妻子的争吵。他快进掉无聊的家庭晚餐，直接跳转到精彩的部分。在电影大约十五分钟的篇幅里，这个遥控器看起来是他人生中遇到的最棒的东西。然后，电影画风突变，而且变得非常沉重。

The feature that becomes the bug

从“功能”变成“漏洞”

Here is the one mid-movie development I have to spoil, because it is the part that should make anyone working in AI sit up straight. The remote starts learning. Michael keeps using it to skip the same kinds of moments, the conflict, the sickness, the waiting, the discomfort. And the device, being a good piece of technology, begins to anticipate him. When Michael panics and asks Morty why the thing is now fast forwarding on its own, Morty delivers the line that the whole movie hinges on. “It’s not a malfunction, it’s a feature,” he says. “It’s using its memory to execute your preferences.” 这里有一个我必须剧透的中段情节，因为正是这一部分应该让所有从事 AI 工作的人警醒。遥控器开始学习了。迈克尔不断用它跳过同一类时刻——冲突、病痛、等待和不适。而这个设备作为一件优秀的科技产品，开始预判他的行为。当迈克尔惊慌失措地问莫蒂为什么遥控器开始自动快进时，莫蒂说出了整部电影的核心台词：“这不是故障，这是功能，”他说，“它在利用记忆来执行你的偏好。”

It remembers stuff about him. It noticed what he always skipped, and it started skipping on his behalf, without asking. This is the moment the convenience tool becomes an autopilot. And Morty explains the mechanism with a cruelty that only Christopher Walken could make sound gentle. During the skipped stretches, Michael’s body stays on what the movie calls “auto-pilot,” going through the motions of everyday life while his mind jumps ahead. He is physically present for everything. He is just never there. 它记住了关于他的一切。它注意到他总是跳过什么，于是开始替他跳过，甚至无需询问。就在这一刻，这个便利工具变成了“自动驾驶”。莫蒂用一种只有克里斯托弗·沃肯才能说得如此温柔的残忍方式解释了其机制：在被跳过的时段里，迈克尔的身体处于电影所说的“自动驾驶”状态，机械地完成日常生活，而他的意识却跳到了未来。他在物理上参与了一切，但他的灵魂从未真正存在过。

What the remote chooses to cut is the part that turns the comedy into something else. “The remote goes by your behavior,” Morty tells him. “Every time there was a conflict between work and home, work won.” And then the kicker: “Lie to your wife. Lie to yourself. But you cannot lie to the remote.” The machine did not invent Michael’s priorities. It read them off his actual choices and then optimized for them with brutal honesty. 遥控器选择剪掉的部分，正是将喜剧转变为其他东西的关键。“遥控器是根据你的行为来运作的，”莫蒂告诉他，“每次工作和家庭发生冲突时，你都选择了工作。”接着是致命一击：“你可以骗你妻子，也可以骗你自己。但你骗不了遥控器。”机器并没有发明迈克尔的优先级，它是从他实际的选择中读取出来的，并以一种残酷的诚实为这些选择进行了优化。

We built the remote

我们造出了遥控器

For most of the last twenty years, Morty’s remote was pure fantasy. You could not actually hand a device your life and have it run the boring parts for you. That is no longer true. As of 2026, the universal remote exists. We just call it an agent. 在过去二十年的大部分时间里，莫蒂的遥控器纯属幻想。你无法真正把生活交给一个设备，让它替你处理无聊的部分。但现在情况不同了。到了 2026 年，万能遥控器已经问世。我们只是把它称为“智能体”（Agent）。

Start with the tools that are unambiguously real and shipping right now. Anthropic’s Claude Code is an agentic coding system that, in the company’s own words, reads the full codebase, plans an approach across multiple files, executes changes, runs tests, and iterates on failures, with the developer setting the goal and reviewing the result rather than guiding each step. The handoff has gone further than a demo. Anthropic has said that the majority of its own production code is now written by Claude Code, with engineers shifting to architecture and orchestration rather than typing lines themselves. 先从那些明确真实且已投入使用的工具说起。Anthropic 的 Claude Code 是一套智能体编程系统，用该公司自己的话来说，它能读取整个代码库，跨多个文件规划方案，执行更改，运行测试，并在失败时进行迭代。开发者只需设定目标并审查结果，而无需指导每一步。这种交接已经超越了演示阶段。Anthropic 表示，其大部分生产代码现在都是由 Claude Code 编写的，工程师们已转向架构和编排工作，而不是亲自敲击代码行。

In early 2026, Anthropic extended this beyond code, letting users message Claude a task from their phone and have the agent complete it on their computer, opening apps, navigating a browser, and filling in spreadsheets. One demo showed a user running late asking Claude to export a pitch deck as a PDF and attach it to a meeting invite, and the agent doing it. OpenAI went down the same road. It launched Operator in January 2025, an agent that uses its own browser to look at a webpage and interact with it by typing, clicking, and scrolling, handling things like filling out forms and ordering groceries. By July 2025 OpenAI folded that into ChatGPT agent, which can summarize your inbox, find open slots on your calendar, plan and shop for a multi-course dinner, and build a slide deck. 2026 年初，Anthropic 将此功能扩展到了代码之外，允许用户通过手机向 Claude 发送任务，让智能体在电脑上完成——包括打开应用、浏览网页和填写电子表格。一个演示显示，一名快迟到的用户要求 Claude 将演示文稿导出为 PDF 并附在会议邀请中，智能体顺利完成了任务。OpenAI 也走上了同样的道路。它在 2025 年 1 月推出了 Operator，这是一个使用自带浏览器查看网页并通过打字、点击和滚动进行交互的智能体，能够处理填写表格和订购杂货等事务。到 2025 年 7 月，OpenAI 将其整合进 ChatGPT 智能体中，它可以总结你的收件箱、查找日程表中的空档、规划并购买多道菜的晚餐，以及制作演示文稿。