个人 AI 智能体胜过工具堆砌
我在科技公司和 AI 公司担任了多年的 CEO 和董事会成员，这是我所看到的趋势：一个非常了解你的 AI 工具，远胜过十个不了解你的工具。 而且你使用它的时间越长，你的效率就会越高。因此在这期视频播客中，我将向你展示如何打造属于你自己的那一个个人 AI 智能体（无论你使用什么平台），我们将用这个个人 AI 智能体去击败那些抽干你生产力的四大阻力。前三种力量偷走你的小时，而最后一种力量偷走你的岁月。最好的 AI 工具，是那个让你不必再去打开另一个工具的工具。
英文原文
I have spent many years as a CEO and board member in tech [music] companies and AI companies and here's what I am seeing. One AI tool that knows you [music] really well is better than 10 that don't. And the longer you use it, the more productive you become. So, in this video podcast, I will show you how to build that one personal AI agent, whichever [music] platform you use, and we will use that personal AI agent to defeat the four forces that drain your productivity. The first three forces [music] steal your hours, but the last one steals your years. The best AI tool is the one that saves you from opening another one.无论你试图完成什么，你的时间和注意力在默认情况下都是有限的。你花在管理琐碎小事上的时间越多，留给重要事情的时间就越少。所以现在的挑战在于：我们没有拥有一个优秀的 AI 幕僚长，反倒雇佣了一群 AI 实习生；你最终把更多时间花在管理他们上，自己反而成了他们的幕僚长——这种情况必须改变。
英文原文
Whatever you're trying to accomplish, [music] your time and your attention are limited by default. The more you spend managing the little things, the less you'll have left for the [music] things that matter. So, here's the challenge. Instead of one great AI chief of staff, we end up hiring a crowd of AI interns, and you end up spending more time managing them, and you become their chief of staff, and that needs to change.个人智能体的 ACT 原则
那么，究竟是什么让这些个人 AI 智能体有别于你已经用过的聊天机器人呢？有三点特征。你该如何记住它们？最简单的思考方式就是我所谓的 ACT 原则：
- Autonomous（自主性）：你给它一个目标，它自己会找出完成任务的方法。它在云端自己的计算机上运行，可以与其他工具、其他智能体和其他应用程序协同工作。它可以浏览网页，也可以执行操作。
- Continuous（持续性）：当你关闭标签页、关闭聊天窗口或合上笔记本电脑时，它不会停止工作。它可以调度安排长达数小时的任务，并在完成后向你汇报。
- Tailored（量身定制）：它以个性化为核心进行设计和训练。因此它能记住你的目标、你的偏好以及你做事的方式。
顺便提一下，在这个视频的演示中我将使用 Cache PT's DOTS（这是他们的个人 AI 智能体），但这些原则适用于所有个人 AI 智能体，无论你使用的是 Claude Co-work、Google 的 Gemini 智能体、Meta 新推出的 Muse，还是 Grock 的 Grockbot。你可以自由选择喜欢的工具，方法都是相同的。
现在请记住一条准则：让他们去管理，让他们去执行，让他们与你协作，但最终决定权必须由你来把握。 否则，你的智能体最后可能会把你的银行对账单发到你的 Instagram 快拍上。所以一定要小心。
英文原文
So, what makes these personal AI agents different from the chatbots that you've already used? Well, there are three things. How do you remember them? The easiest way to think about it is what I call act.First, they're autonomous. So, you give it a goal and it figures out how to get the job done. It runs on its own computer in the cloud. It can work with other tools, other agents, other apps. It can browse the web. It can take actions. Second, it's continuous. So, it doesn't stop working when you close the tab or you close the chat or you close your laptop. It can orchestrate a task that can run for hours. It'll come back to you when it's done. And third, tailored. It is crafted and trained with personalization in mind. So, it remembers your goals, your preferences, and how you like things done. And by the way, I'm going to use Cache PT's DOTS, which is their personal AI agent for the demos in this video, but the principles apply to all personal AI agents, whether you're using Claude Co-work or Google's Gemini agent or Meta has Muse now or Grock has Grockbot. So you can pick your favorite tool. The approach is the same.
Now one rule to remember, let them manage, let them execute, let them partner with you, but it's you who makes the final call. Otherwise, your agent might end up posting your bank statement on your Instagram story. So be careful.
第一大阻力：碎片化（Fragments）
让我们从第一种力量谈起：碎片化（Fragments）。有一件事你每天要做 1,200 次，而且它根本不在你的岗位职责说明里。有时候所有这些 AI 工具反而让我觉得我自己成了一个工具。
2022 年，研究人员在《哈佛商业评论》上发表了一篇文章。他们追踪了《财富》500 强企业的员工整整一个月，记录并分析了每个人每次从一个应用或网站切换到另一个应用或网站的行为。平均每个人每天切换近 1,200 次。每次切换虽然只耗费约 2 秒钟，但这相当于每天约 40 分钟——已经等同于一顿午休的时间了。
但在隐形成本方面，还有更令人担忧的问题：碎片化破坏了你的注意力质量。为什么？因为我们所需的信息分散在各处。因此在真正开展工作之前，你不得不把时间花在搜寻工作所需的信息上。这就是你的个人 AI 智能体必须去抗击的第一大力量。那个 AI 智能体需要成为黏合剂，把你所有的应用和信息孤岛连接起来，让你能够专注于收获信息，而不是四处寻找信息。
英文原文
Let's start with the first force, fragments. There is a job you do 1,200 times a day and it's not in your job description. Sometimes all these AI tools make me feel like I am a tool. In 2022, researchers published an article in Harvard Business Review. They tracked employees of Fortune 500 companies over a month and they tracked and logged every time someone switched from one app or website to another. The average person switched nearly 1,200 times a day. Now each switch costs only about 2 seconds. So that's about 40 minutes a day. But that's as long as a lunch break. But there's something more alarming in terms of hidden cost. So fragmentation damages the the quality of your attention. Why? Because the information we need is scattered everywhere. So you spend time hunting for the information you need to do the work before you can actually do the work. That's the first force. Your personal AI agent has to fight. that force and that AI agent needs to become the glue that connects all your apps and information islands so you can focus on harvesting the info not hunting for it.来看一个例子。设想你获得了一场渴望已久的工作面试机会，这可能会改变你的职业生涯。但接下来怎么办？因为你的简历在 Google Drive 的某个角落；你以往面试的笔记放在另一个地方；职位描述在别处；你与招聘人员的交流记录散落在某些邮件里；公司背景资料分散在网络各处；而证明你最优秀工作的证据，实际上散布在整个网盘的演示文稿、文档和邮件之中。要将这些整合起来需要花费数天时间。你该如何为这场面试做好定位？为什么不尝试点不一样的做法呢？
你可以带着你的个人 AI 智能体去散个步。你可以给它提示，你可以说：“我拿到了一份我渴望多年的工作面试。请帮我准备，让我作为他们所见过准备最充分的候选人走进考场。请使用我的简历、过去的成果、往来邮件、公司背景，并调研这家公司的面试流程。”
现在，这可以是任何公司，但看看现在的公司都在做什么：例如麦肯锡正在测试让 AI 智能体在初级顾问面试中扮演客户。因此候选人必须现场解决客户的问题，而人类面试官则在一旁观察并给他们打分。这就是我们正在迈入的世界。
因此你的面试准备也必须提升一个层级，对吧？你可以让你的 AI 智能体模拟那场面试。提示词大致可以这样写：“现在根据你对我、对该岗位以及对公司的了解来模拟这场面试。向我提出真实的问题，挑战我的回答，并在我推理薄弱时进行反驳。不要放水。”
当你完成后，你可以让它评估你的表现和回答。你可以写下笔记，然后再来一次。最棒的地方在于：如果面试表现很差，嘿，谁在乎呢？那只是一场排练。一遍又一遍地练下去，至少你还达成了每天走 10,000 步的目标。
这样一来，你的智能体负责收集一切。它访问了七个不同的工具和应用，扮演你的面试官。而你只需专注于一件事：不断进步并拿下那份工作。请记住，只有在你将这些智能体连接到具体的应用和工具后，它们才能从其他地方获取信息。
英文原文
Let's go through an example. Imagine you have landed an interview for a job you have wanted for years. This could change your career. But now what? Because your resume is in Google Drive somewhere. Your notes from previous interviews sit somewhere else. The job description is somewhere else. Your conversations with the recruiter are in some emails. [music] The company background is out there somewhere. The evidence of your best work is actually scattered across the entire drive in presentations and documents and emails. [music] So, this would take days to consolidate. How do you position yourself for this interview? Why not try something different?You can go for a walk with your personal AI agent. You can prompt it. You can say, "I've got an interview for a job I have wanted for years. Help me walk in as the most prepared candidate they've ever seen. Use my resume, my past work, my emails, company background, and research the company's interview process." Now, this could be any company, but here's what companies are doing right now. McKenzie, for example, is testing AI agents that play a client in interviews for entry-level consultants. So, the candidate has to work through the client's problem live while the human interviewer is watching and scoring them. That's the world we're walking into. So, your interview prep has to move up a level, too, right? You tell your AI agent to simulate that interview. The prompt would be something like, "Now simulate the interview based on what you've learned about me and the role and the company. Ask me realistic questions, challenge my answers, and push back when my reasoning is weak. Don't make it easy." And then when you're finished, you can ask it to critique your performance, your answers. You can write up notes, and you can do it again. Now, here's the best part. If the interview goes badly, hey, who cares? That was just a rehearsal. Do it again and again. At least you got your 10,000 steps in. So, your agent gathers everything. It visits seven different tools and apps and acts like your interviewer. And you focus on just one thing, getting better and getting that job. Now remember that these agents can get information from other places only after you connect them to those specific apps and tools.
第二大阻力：信息泛滥（Flood）
现在让我们转向对抗拖慢生产力的第二大力量：信息泛滥（Flood）。每天 117 封电子邮件，153 条聊天消息。你的 AI 智能体能做的最棒的事，或许恰恰是它永远不展现给你的内容。
2025 年，微软对约 31,000 人进行了调查。普通员工平均每天收到 117 封电子邮件和 153 条 Teams 消息。诚然，大多数邮件得到的关注不足一分钟，但这就是信息的洪流；这甚至还没算上从四面八方向你袭来的社交媒体内容。而在那堆信息里的某处，正藏着真正关键的那一条。
而且，重要的事情通常都相当复杂。卡耐基梅隆大学的研究人员曾针对拥有私人保险的美国人测试了四个基本概念（如免赔额 deductible 和挂号费/自付额 copay），只有 14% 的人把全部四个定义都答对了。所以你实际上在与两重泛滥抗争：一是随时涌向你的信息过多，二是信息有时过于晦涩密集以至于难以据此采取行动。
这正是你的个人 AI 智能体大显身手的地方。它可以过滤噪音，决定什么值得你关注；并化繁为简，提炼出什么值得你进行判断。这恰恰是一位幕僚长会做的事，对吧？当然这其中也存在风险：无论谁或者什么充当守门人，它都在决定你能看到什么。因此不要仅仅停留在设定目标上，还要检查它的工作并持续微调。
英文原文
Now let's move to fighting the second force that slows down productivity flood. 117 emails, 153 chat messages every single day. The best thing your AI agent can do might be what it never shows you. In 2025, Microsoft surveyed about 31,000 people. The average worker [music] received 117 emails and 153 Teams messages a day. Most of those emails got less than a minute of attention, sure, but that's the flood of information. And that's not even including the social media stuff that comes at you from every direction. And somewhere in there is that one message that actually mattered. [music] And also the things that matter are usually pretty complex. So researchers at Carnegie Melon tested Americans with private insurance on four basic terms [music] like deductible and co-ay. And only 14% [music] got all four definitions right. So you're fighting two floods really. Too much reaches you all the time and at times it is too dense [music] to act on. And this is where your personal AI agent comes in handy. [music] It can filter the noise and decide what deserves your attention and simplify what deserves your judgment. And that's exactly what a chief of staff would do. Right? Of course that's also the risk. Whoever or whatever is the gatekeeper decides what you get to see. So don't stop at setting the goals. Check the work and keep refining.那么我们来构建这个过滤器。你可以给你的个人 AI 智能体写类似这样的提示词：“我想保护我的专注力，只在真正需要我介入的事情上被分心。因此，在上午 10:00、下午 1:00 和下午 4:00 检查我的邮件、消息和日历，只标记正在等待我做决定的事项、变更的截止日期或需要我知晓的变动；把其他所有内容都暂存到下午 5:00 的汇总中。” 这是一个极佳的提示词，我自己一直在用，因为你会惊讶地发现，你以为紧急的工作中有那么多完全可以等到下午 5:00，甚至有些事情等到你退休都没问题。
好的，那么面对复杂性该怎么办呢？这就是你解开复杂性的方式。假设你正在审查新公寓的租约。把它交给你的个人智能体，它会提供帮助。你可以这样说：“我马上要签这份租约了，我想确切知道我同意了什么内容。” 你看，在所有这些提示词中，你给出的都是目标而不仅仅是指示，是目的而不仅仅是指导。“标记出任何可能让我花钱或限制我能做的事情的条款，并向我展示确切的原文条款。” 极好的提示词。你现在在复杂与简单之间拥有了一个翻译层。你可以快速从被动吸收信息转变为主动做出决策。
英文原文
So let's create that filter. Prompt your personal AI agent something like I want to protect my focus and I want to only get pulled away for what truly needs me. So, at 10:00 a.m., 100 p.m., and 400 p.m., check my emails, my messages, my calendar, and flag only decisions waiting for me. Deadlines that moved or changes that I need to know about, park everything else for a 5:00 p.m. summary. And this is a great prompt. I use it all the time because you'll be surprised how much of your urgent work can wait until 5:00 p.m. and some of it can even wait until you retire. And okay, what about the complexity? Well, this is how you untangle it. Let's say you're reviewing the lease for your new apartment. Okay, get it and give it to your personal agent and it will help. You can say something like, "I'm about to sign this lease and I want to know exactly what I'm agreeing to." You see, in all of these prompts, you're giving objectives, not just instructions, goals, not just guidance. Flag anything that could cost me money or limit what I can do, and show me the exact clauses. Great prompt. You now have a translation layer between complexity and simplicity. You can quickly move from consuming information to making decisions.第三大阻力：低频事务（Frequency）
这把我们引向了第三大阻力：低频事务（Frequency）——即无论你准备好与否，总会周期性出现的工作。你每天晚上都会记得吃晚饭，这是个习惯；你会记得三周后要和你的一生挚爱飞往巴黎，你迫不及待地开始收拾行李并倒数日子，因为这是一件头等大事。
但为什么你总是记不住每周吃一次那个维生素？我也会忘记，原因有二：首先，习惯是通过重复建立的，而每周一次的任务重复频率不足以养成习惯；其次，我们都需要一个触发器来提醒我们应对那些发生频率低但具有周期性的任务。没有习惯，不是头等大事，就根本无法持续追踪——除非你有你那位可靠友好的随身 AI 智能体。
让我给你举一个真实的例子。几周前，我买了我的 iPhone 18。你知道，我通常每四年左右才换一次 iPhone，平时根本没有升级的理由。但在两天前，我的智能体提醒我：“嘿，你的折抵换购套件大约是在 14 天前送达的，所以今天是寄回旧 iPhone 以完成折抵换购的最后一天。” 我当时惊呼：“哇，太感谢了，Capes。” 顺便说一下，那是她的名字。在我的心目中，她是一个能同时应付一百件事的超级英雄，而我却连手机落在哪里都记不住。
英文原文
And that brings us to the third force, frequency. The work that keeps coming back whether you're ready or not. You remember to eat dinner every night. It's a habit. You'll remember you're flying to Paris with the love of your life 3 weeks from now and you can't wait to start packing. You're counting days. It's a headline event. But why do you have trouble remembering to take that vitamin once a week? I do for two reasons. First, a habit is built by repetition. And a weekly task doesn't get enough repetition to become a habit. Second, we all need a trigger to remind us about those infrequent but recurring tasks. No habit, no headline, no hope of tracking it. Unless, of course, you have your friendly neighborhood AI agent. Let me give you a a real example.A few weeks ago, I bought my iPhone 18. You know, I usually buy my iPhone every four years or so. There's no reason to upgrade it otherwise. But 2 days ago, my agent reminded me, "Hey, your trade-in kit arrived about 14 days ago. So today is the last day for you to return your old iPhone for your trade in." I was like, "Oh, wow. Thank you, Capes." That's her name, by the way. In my mind, she's a superhero who can juggle a hundred things at once. and I can't remember where I left my phone.
这里还有另一个例子。以前我睡觉时下颌会阻塞气道，所以我每小时会醒来 34 次，因为每次停止呼吸时大脑都会把我惊醒。现在这个问题已经解决了。但前几天我给 Capes 发了条短消息：“我上次去看肺科医生（那位睡眠障碍专科医生）是什么时候？” 她很快答复说：“已经有一年了。” 于是我让她帮我预约门诊。然后，最关键的事情来了，我说：“请确保我每 3 到 4 个月去见一次医生。” 这就是你让智能体成为你行动触发器的方式。
这里还有一点需要牢记：我的 AI 智能体只负责预约，它并没有获取我的健康档案或任何财务信息。目前我还不放心将这些信息与 AI 共享，这是我现阶段的选择，你也可以做出你自己的判断。
所以你的行动项是：在你的生活中挑选出一件既无日常习惯依托、又非头等大事的事情，本周就把它交给你的智能体。顺便说一句，如果你喜欢这些理念，你也会喜欢我的新闻通讯。它每周二发出，包含一个洞察、一个工具、一种实践，而且是免费的。我很喜欢写这个，链接就在下方。
英文原文
So, here's another example. My jaw used to block my air waves when I used to sleep. So, I would wake up 34 times every hour because my brain would jolt me awake every time I stopped breathing. So, it's it's fixed now. Uh, but the other day I sent Capes a quick line saying, "When was the last time I saw my pulmonologist, the sleep disorder specialist?" And of course, she came back and said, "It's been a year." So, I asked her to set up an appointment. And then, this is the most important thing. I said, "Make sure I see my doctor every 3 to four months." And that's how you make your agent your trigger. And here's another thing to keep in mind. My AI agent only set the appointment. It did not have my health records or any financial information. Now, I'm not comfortable sharing any of that with AI. That's my choice for now. You can make your own judgment.So, here's your action item. Pick one thing in your life with no habit and no headline and hand it to your agent this week. And by the way, if you like these ideas, you'll enjoy my newsletter as well. It goes out every Tuesday. One inside, one tool, one practice, and it's free. I love writing it. The link is right below.
第四大阻力：未来与真正重要的事（Future）
到目前为止，这些智能体正在帮助你赢回注意力、夜晚以及内心的宁静。但这也引出了一个更重大的问题：你打算用赢回的这些做什么？这就是第四大阻力：未来（Future）。
前三种力量偷走你的小时，而这一种力量偷走你的岁月。“改天再做”正是生命悄然溜走的方式。YouTube 上有一个著名的视频叫《最后一课》（The Last Lecture），你可能已经看过了。它由计算机科学教授兰迪·波许（Randy Pausch）主讲。他在得知自己只剩几个月生命时（他患有胰腺癌）在很多年前发表了那次演讲。在去世前他还做过另一场关于时间管理的演讲。在那场演讲中他提到，在文化层面上美国人对金钱极为感兴趣——一个人赚多少钱、怎么花钱、净资产是多少等等；但我们对时间却没有那么感兴趣。我们为花钱做预算，却不对如何花时间做预算。
兰迪的洞见在于：我们全搞反了。在做完那场演讲不到一年后他就辞世了，年仅 47 岁。他深知时间的价值，因此他把时间花在了最重要的事情上。
英文原文
So far, these agents are helping you earn back your attention, your evenings, your peace of mind. But that leads us to a bigger question. What are you going to do with it? That's the fourth force, future. The first three forces steal your hours. This one [music] steals your years. I'll do it later. Is how life slips [music] away. There is a famous video on YouTube called the last [music] lecture. You may have seen it already. It's by computer science professor Randy Pouch and he gave it many years ago when he knew that he only had a few months to live. [music] He had pancreatic cancer and um he also gave another lecture before passing away and that was on time [music] management. In that one, he talks about how Americans as a culture [music] are very interested in money, how much someone earns, how they spend it, their net worth, and so on and on. But we're not that interested in time. We make a budget on how we spend money, but we don't budget how we spend time. And Ry's insight was we have it backwards. He passed away less than a year after that lecture. He was 47 and [clears throat] he knew the value of time. So he spent it on what mattered the most.所以真正的核心问题是：AI 智能体将把时间交还给你，你打算用它来做什么？生产力只有在帮助你达成人生目标时才有价值。
因此，你的行动项是：写下你一直在留给未来的三件事。然后告诉你的智能体：“这些是我的优先事项。每周日晚上，审查我本周的日历和工作。告诉我为了实现这些优先目标我都做了什么，以及下周我应该做出哪些调整。”
现在，你的智能体将推动你走向对你重要的事情，引向你的未来。关于时间最奇异的一点是：只有当它所剩无几时，它才显得尤为珍贵。
所以，看一眼你的手表。认认真真地看一眼。
你知道现在是什么时间吗？这是属于你自己的时间。
谢谢你们，我爱你们。
英文原文
So here's the real question. AI agents will give you back time. What are you going to do with it? Productivity is great only when it helps you achieve your purpose. So, here's the action item. Write down three things you keep saving for the future. And then tell your agent, "These are the priorities. Every Sunday evening, review my calendar and work from this week. Tell me what I did to achieve these priorities and what I should change next week." Now your agent moves you to what matters to you and toward your future.The strange thing about time is that it feels precious only when there is less of it left.
So take a look at your watch. Really look.
You know what time it is? It's your time.
Thank you. And I love you.