OK here’s a sentence I’ve caught myself saying way too often this year:
“That’s not really me.”
Usually right before I don’t do something. I should write more, but honestly I’m not really a writer. I should be more outgoing at parties, but I’m just an introvert. I should run in the morning, but I’ve never been a morning person. You get the idea.
Each one feels reasonable. Even kind of self-aware. And every single one is doing the same trick: using “the real me” as a permission slip to skip the harder, newer, more exposing version of me.
Lately I’ve been reading an old book that finally cracked this open. Psycho-Cybernetics, by some plastic surgeon named Maxwell Maltz. He kept noticing something weird at work. He’d successfully reshape someone’s nose, and they’d look in the mirror and still feel ugly. The face was different. The self-image wasn’t. Other times, a tiny little correction would crack a person open into a totally different personality. The size of the surgery had basically nothing to do with it. The size of the internal shift was everything.
So Maltz borrowed an idea from cybernetics: your brain is a servo. You set a target — “the kind of person I am” — and the brain runs feedback loops to keep you locked on it. Behaviors that don’t match get filtered out as “weird, not me.” Behaviors that do match get amplified. The servo doesn’t actually care whether you’ve aimed it high or low. It just keeps you parked wherever you’ve parked it.
A few annoying things follow.
One. The guy on the couch saying “I’m just not a fitness person” is not stating a fact. He’s setting a target. The servo will do what it’s told.
Two. Willpower can override the servo for two or three weeks. Then the servo wins. This is why every diet you’ve ever tried evaporated. You weren’t fighting your willpower. You were fighting your self-image.
Three. The self-image is editable. Just not by reading about it. Or deciding really hard. It’s editable by behavior. Your brain is locked inside your skull. It can’t see the outside world. The only data it gets about who you are is what you do. So if you keep doing the thing you don’t think you are, eventually the brain shrugs and updates: huh, I guess we’re a writer now.
This is what’s actually under “fake it till you make it.” A phrase that sounds like bullshit until you remember the alternative is “wait until you feel like it” — i.e., never.
Trap 1: I want it really bad
Naval has a sentence that gets sharper every year I read it:
“Desire is a contract you sign with yourself to be unhappy until you get what you want.”
Read that twice.
When you say “I want X,” what you’re announcing to your brain is “I currently don’t have X, and I will be unhappy until I do.” You just made now a deficit. You set the servo target to “person who is missing X.” That target stays even if you somehow get X — because the servo will spawn a new “I want Y” within roughly 48 hours and lock you into the next contract.
Money is mine. I have spent years vaguely “wanting to be more financially set.” The feeling never fully arrives. Even when the screen numbers go up, the feeling doesn’t. Because the servo target was never the number. The servo target was the role — “person who needs more.” That role is doing what it was told to do.
Naval and Maltz point at the same fix. Don’t want it harder. Enter the state. Stop saying “I want to be wealthy.” Start asking, with embarrassing literalness: “what would someone who already has enough money do today?” And then do that. The servo target just changed from “person chasing” to “person who has.” Which means the action stream now produces evidence consistent with the new self.
It sounds woo. It is also roughly what every successful person you’ve ever read about did, more or less knowingly.
Trap 2: I just want to be authentic
This one is harder to dismantle because it sounds so good.
I want to be authentic. I want to live as my true self. You say it, people nod. You’re attacked from no quarter.
But the “true self” you’re protecting is mostly a stack of accidents. Childhood scripts your parents accidentally installed. Imprints from things that didn’t work out when you were twelve. Habits you fell into because the dorm room had bad lighting. Lazy patterns that compounded because you were tired one Tuesday in 2019 and never quite came back online.
That stack got the label “the real me” because it’s familiar. Not because it’s true.
Buddhism has the cleanest take I know on this. Anatta, no-self. There is no fixed permanent you waiting underneath all the performances to be honored. Whoever you’ve been performing for the last twenty years is also a performance. It’s just been running longer.
Once you accept that every version is a costume, the question stops being “which one is real?” and quietly becomes “which one would I rather wear?” Because there’s no fact-of-the-matter version waiting underneath. Just versions with different track records.
The reframe: if every “me” is performance, the only honest question is — which performance gets me richer, stronger, happier? And why am I not playing that one?
Uncomfortable question. Hard to unhear once you’ve heard it.
So how do you actually move the servo?
Three levers. Not equal weight. You’ll lean on different ones at different stages.
Cold start: replay the role in your head. Maltz’s headline finding is that the nervous system can’t fully tell the difference between a real experience and a vividly imagined one. So 30 minutes of specific mental rehearsal — not “imagine being successful” but “imagine exactly what I’d do at 9 AM tomorrow if I were already this person, what I’d say, what I’d push back on, what I’d ignore” — feeds the servo synthetic training data. The servo doesn’t audit the source.
I used to roll my eyes at visualization. Then before a difficult conversation last year I tried it for two weeks. The actual conversation was easier than every version I had run in my head. Whether that’s “training” or just “I thought it through more carefully,” I don’t know. Probably some of both. Either way it was free.
Live execution: don’t argue with your brain. Here’s a thing nobody tells you about doing new behavior: your brain will actively sabotage it. Not because the goal is wrong. Because new neural pathways are expensive, and your brain is a deeply lazy organ that would rather you keep doing the cheap thing.
So when you sit down to write and a wave of “ugh, not today, you’re tired” rises up — that isn’t a signal. That’s the cost of doing business. The technique I’ve found most useful is to step back from the brain. Don’t fight it, observe it. Like there’s a screaming toddler in the back seat and I’m the adult driving. The toddler doesn’t want to listen to reason. Fine. We’re still going to the dentist.
It’s the most boring superpower you can develop. Annoyingly, it also works.
Compounding: feedback the process, not the outcome. Praise a kid for being smart and they avoid challenge — because being smart is the asset, and challenge risks the asset. Praise a kid for the effort and they seek challenge, because effort is where the asset is built.
Same for you, today. Stop measuring “did I become a writer this month.” Start measuring “did I write today.” The first metric is fragile and depressing. The second one compounds. Every time you do the action of the new role, you cut a millimeter off the old neural groove and add a millimeter to the new one. Month one is brutal. Month twelve is meaningfully easier. Year one and the new role is the default — willpower can go back to other jobs.
This is the part that I wish someone had told me earlier: it does get easier. The servo really does install the new target. You just have to feed it real data for long enough.
One last thing — this isn’t only about you
I write code for a living, and I’ve been noticing: the same mechanism that makes humans hard to change makes AI agents hard to change too.
If you’ve used Claude Code or any of the agent harnesses, you know there’s a config file (CLAUDE.md, or whatever your harness calls it) where you describe how the agent should behave. People assume the secret to a better agent is a smarter prompt. It isn’t. The secret is the file — the persona, the standing rules, the durable instructions. That’s the agent’s self-image. Change the file, change the behavior. Add a vague “be more careful” instruction without changing the file, and the agent reverts within a few turns.
Same servo loop. Persona is the target. Completions are the feedback. The agent drifts toward whoever it thinks it is.
Kids too, and harder. Ages 0 to 10 is when the servo target gets installed for the next 70 years. A six-year-old who’s been told “you’re a reader” is not a kid who reads more. They’re a kid whose servo is locked onto reader. Which means decade over decade they will, on every margin, drift toward more reading. Not because reading is fun. Because that’s who they are.
This is also why “praise the process, not the outcome” matters so much for parents. Outcome praise installs a fragile target (“I’m smart, must protect smart”). Process praise installs an antifragile one (“I’m someone who tries hard, more challenge = more me”). One self-image compounds. The other one cracks under the first real test.
OK I’ll stop. The whole thing collapses to a question that sounds dumb until you actually sit with it:
You always act consistently with who you think you are. So — who do you think you are?
If your answer is something like “a person who’s been meaning to write more for like four years” — the brain has been listening, and it’s been giving you exactly that. The servo is doing its job. The job’s just installed wrong.
The good news is, you can install something different. The bad news is, not by deciding. By doing.
Anyway. Going to go drag my tail in the mud now, I guess.
今年我老挂在嘴边一句话，最近才注意到——
“我不是那种人。”
基本上每次出现，都是我没去做某件事之前：
“应该多写点东西吧，可我又不是会写的人。” “应该饭局上活跃一点嘛，但我就是 i 人。” “应该早起跑步——我从小就起不来。”
每一句听起来都挺合理。甚至有点自省的味道。
但说白了，每句都在干同一件事——拿”真实的我”当借口，绕开那个更难、更陌生、更暴露的我。
最近重读了 Maxwell Maltz 那本《Psycho-Cybernetics》（中文译名《心理控制术》）。里面一个反直觉的观察，终于把我这毛病撬开了。
Maltz 是个整形外科医生。他在工作里注意到一件挺怪的事。
手术做得很成功，鼻子改漂亮了，病人照镜子还是觉得自己丑。脸变了，心里那个”我”没变。 反过来也有：一个非常小的修整，能把一个人整个性格炸开。这一刀切到的不是脸，是脑子里”我长这样”的那张图。
Maltz 顺手从控制论里借了一个词，叫 servo（伺服）。
你的大脑就是个伺服。你给它设一个目标——”我是个怎样的人”——它就老老实实跑反馈循环，把你的行为往那个目标上拽。
跟目标不符的，被它过滤掉，标签是”这不像我”。 跟目标一致的，被它放大。
伺服不挑剔。你设个高的，它给你拉高；设个低的，它就死死按住你。它真的不在乎你瞄的是哪儿，只负责让你停在那个点上。
——其实这事金庸早写过。郭靖学降龙十八掌之前，洪七公看他一眼说”我教你也是教，不教你也是教”。郭靖资质鲁钝，可他认了”我就是个老实学武的人”这个目标。然后降龙十八掌就给他练成了。反观令狐冲——天资极高，独孤九剑一上手就开窍——可他早期身体里塞了七八门内功路数，互相打架，差点死掉。他的伺服锁的目标不是”练成什么”，是”我是什么样的人”，而那个人是”风流浪子兼少林大力金刚”——身体兜不住。换句话说，练什么招式不重要，先得想清楚自己锁的是哪个目标。郭靖锁对了，所以降龙十八掌从他手里出来重若千钧。这就是伺服。
这件事再往下推，有几条挺烦人的结论。
一，那个躺在沙发上说”我天生不适合健身”的哥们，不是在陈述事实。他是在给伺服设目标。然后伺服很尽职地照办了。
二，意志力可以短期把伺服压住，撑两三周。然后伺服赢。你试过的每一次减肥最后都不了了之，原因在这里——意志力没掉链子。是自我形象在拖你。
三，自我形象是可以改的。但改它只能靠做。看书、立 flag、下决心都没用。你的大脑被锁在头骨里，看不到外面世界。它判断”你是谁”的唯一证据，就是你做的事。所以你哪怕硬着头皮做”那个你以为不是你”的事——做久了，大脑会自己更新：”行吧，看来咱们是写作的人。”
这才是 “fake it till you make it” 在神经科学层面真正的意思。这话听上去一直挺鸡汤。直到你想想替代方案是什么——”等我有感觉了再开始”。基本就是永远不开始。
陷阱一：我特别想要
Naval 那句话每年读都更狠一点：
“欲望是你和自己签的一份合同，规定在拿到 X 之前你都不许开心。”
读两遍。
你说”我想要 X”的时候，你其实是在告诉你的大脑：我现在没有 X，并且拿到之前我都不能开心。这一下子把”现在”标记成了亏空。伺服立刻把你的身份认成”那个还没拿到 X 的人”。
更糟糕的是这个状态会自我延续。就算你真的拿到了 X，伺服会立刻给你生出一个 Y，让你继续不开心。
我自己最典型的例子是钱。”想要财务更宽裕”我想了好多年。那种”到了”的感觉从来没真正来过。 屏幕上的数字明明在涨，那种感觉就是不来。因为伺服锁定的目标从来不是数字，而是”那个总觉得自己缺一点的人”这个角色。这角色干得很尽职。
Naval 跟 Maltz 都指向同一个解法：别更努力地”想要”。直接进入状态。
把”我想变富”翻译得直白一点——一个已经不缺钱的人今天会怎么过？ 然后照那样过。一旦你开始这么干，伺服锁定的目标就从”在追”切换成了”已是”。你的行为流，开始产生跟”已是”一致的证据。
听起来挺玄。但你想想，你读过的那些”成功的人”，多多少少都在干这件事——只是有些人自觉，有些人不自觉。
陷阱二：我只是想做真实的自己
这个更难拆，因为它听上去太无懈可击了。
“我想活得真实啊。”“我想做最真实的自己嘛。”——你说出来，谁敢反对？
可你要是认真扒一扒，所谓”真实的我”，其实是一摞意外堆起来的东西：
- 童年时候爸妈无意装在你身上的剧本。
- 12 岁那年某件事失败留下的阴影。
- 大学宿舍灯光不好养出的习惯。
- 2019 年某个礼拜二你太累没爬起来，然后这件事自动复利了。
这一堆东西被贴上”真实的我”四个字，不是因为它真，是因为它熟。
佛家有句话是最准的——无我。底下没有一个固定不变的”你”在等你回去尊重。 你过去二十年扮演的那个角色，也是表演。它只是演得久。
一旦你接受”每个版本都是表演”，问题就不再是”哪个最真”。问题会很安静地变成——
既然每个版本都是演的，那我演哪一个，能让我变富、变强、变开心？
第一次听到这个问题，是会扎心的。然后你会发现，它就立在那儿，没法反驳。
那怎么真的把伺服推过去？
三个杠杆。不是均等使用——不同阶段你会偏重不同的那个。
冷启动：在脑子里把那个角色排练一遍。
Maltz 最被引用的一个发现是：神经系统分不清真实经验和生动想象。所以”每天 30 分钟具体的心理排练”，不是你以为的那种泛泛”想象成功”。它要非常具体——
如果我已经是那个人了，明天早上 9 点我会在做什么？说什么？推掉哪个会？忽略哪条消息？
用细节去喂伺服。伺服懒得查你给它的数据是真的还是合成的。
我以前对 visualization 一直翻白眼。直到去年一个我特别怕的对话之前，连续两周这么排练了一遍。等真的对话发生的时候，比我脑子里跑过的任何一个版本都简单。是”训练”起作用了，还是我只是多想了几遍？说不清。但反正——免费的事，不做白不做。
真上场：别跟你的大脑讲道理。
这件事真的没人事先告诉你——你尝试新行为的时候，大脑会主动给你制造阻力。焦虑、犯困、各种合理化（”明天再开始”、”这真不像我”）。
别把这些读成”目标错了”。它们是开张成本。 新神经回路烧能量，大脑非常懒，它当然想让你回到便宜的旧轨。
我用过最有效的技巧是：别跟它吵，绕到它后面看着它。 想象你后座坐了个三岁小孩在闹脾气，你是开车的大人。小孩不听道理。没关系。我们还是去看牙医。
这是世界上最无聊的一种超能力。烦人的是，它真管用。
复利：反馈过程，不反馈结果。
夸一个孩子聪明，他会回避挑战。因为”聪明”是他的资产，挑战会威胁这个资产。 夸一个孩子用功，他会主动找挑战。因为努力才是产出资产的地方。
你现在也一样。别量”我这个月成为作家了吗”，量”我今天写了吗”。第一个指标脆弱又压抑。第二个会复利。
每完成一次新角色的动作，旧回路砍一刀，新回路加一砖。第一个月很苦。第十二个月明显轻松一点。第一年之后，新角色变成默认配置——意志力可以挪去别的地方了。
这是很少有人提的一段：它真的会变容易。伺服真的会把新目标装上。只是你得喂它真的数据，喂够长时间。
最后一个延伸——这事儿不光是你自己的事
我吃饭的家伙是工程，最近一直在想——让人难以改变的同一个机制，让 AI agent 也一样难改。
用过 Claude Code 这类 agent 的人应该熟悉：它有个配置文件（CLAUDE.md，或者你 harness 里叫别的名字），你在里面写这个 agent 应该怎么行动。
大家都以为 agent 表现好坏的关键是”prompt 写得好不好”——不是。关键是这个文件。Persona、长期规则、稳定指令。这就是 agent 的”自我形象”。
改文件，行为就变。不改文件，光在对话里临时命令一句”你要更仔细一点”——几轮之后它就漂回去了。
是同一个伺服回路。Persona 是目标。Completion 是反馈。Agent 朝它”以为自己是谁”漂移。
孩子也一样，而且更狠。0 到 10 岁是自我形象安装的关键窗口，装好了跟你 70 年。
一个 6 岁就被告知”你是个会读书的孩子”的小孩——他不只是”读得更多”。他的伺服锁定在’读书的人’这个目标上。几十年下来，他真的就变成了那个人。
这跟读书快不快乐没关系。这就是他。
这也是为什么”夸过程不夸结果”对家长这么承重。
夸结果，装的是脆弱目标——”我聪明，我得保住聪明”。 夸过程，装的是反脆弱目标——”我是个肯下功夫的人，挑战越多越像我”。
一个会复利。一个会在第一次真考验来的时候碎掉。
行了，打住。
整件事最后落到一个听起来很傻、但你真的坐下来想就坐不住的问题——
你的行为永远和”你以为自己是谁”一致。 所以——你以为自己是谁？
如果你的答案是”一个想多写东西想了四年的人”——大脑听到了，而且非常忠实地兑现了它。伺服在尽职工作。只是任务装错了。
好消息：你可以装一个不一样的进去。 坏消息：不是靠决心，是靠去做。
也就这么回事。我去泥里曳尾巴了。