慢 想 录
ISSUE 01 · 2026.08
ESSAYS ON WORK, AI & CHOICE

在 AI 时代,
成为一个有选择权的人

当 AI 同时放大个人生产力与组织控制力, 真正值得讨论的,也许不再是哪家公司更“先进”, 而是人在新的工作结构里,如何保有判断、成长与离开的自由。

EDITOR'S NOTE

有些问题,看起来是在谈一家企业、一套管理方式,甚至一代人对技术的不同理解。 但再往下想一层,它真正触及的,也许是整个时代共同的职业处境。

THE QUESTION · 从一个关于 Agent 的讨论开始
最近读到 数字生命卡兹克的文章《创业2年半后,想跟你分享关于AI组织的这7点心得》。 它让我想到一个有些矛盾的现象: 同样是 AI,一方面可以减少重复劳动、扩大个人能力和创造空间; 另一方面,也可以让任务分配、绩效衡量和过程管理变得前所未有地精细。

起初很容易把这种差异理解为某一家企业的管理风格, 或者某一类管理者对技术的不同理解。 但再往下想,也许这并不只是个别公司的选择, 而与行业竞争、增长压力、组织治理乃至整个时代的工作方式有关。 如果这是一个比“好公司与坏公司”更大的问题, 那么作为个人,我们应该怎样理解这种变化? 身处不同类型的组织时,可以做什么? 又该如何规划自己的长期事业方向?

AI进入工作之后,有一个问题越来越值得注意:效率提升以后,省下来的时间最终去了哪里?

它当然可能让人从重复劳动里脱身。过去需要一天完成的工作, 现在几个小时甚至几十分钟就能完成, 于是人可以把更多精力投入判断、创造、沟通和探索。

但另一种可能同样存在。

原来一天完成5件事,现在变成20件; 原来管理者只能看到最终结果,现在可以实时掌握过程; 原来很多工作因为管理成本太高而无法量化, 现在Agent可以自动拆解、追踪、评价。

AI并不会天然把人带向自由。
它既能放大个人,也能放大组织。

于是问题也就从“某家公司是否懂AI”,变成了一个更大的问题:

在AI让组织控制力和个人生产力同时暴涨的时代,
一个普通人,怎样避免自己只是变成一种更高效的生产资料?

这个问题,其实值得认真想。

CHAPTER 01这确实不只是某一家公司的问题

如果只把这种现象归因于组织惯性、管理风格, 甚至某些代际差异,可能仍然把问题看窄了。

因为换一个更年轻的团队,也完全可能出现另一种高度数字化、 高度量化、高度控制的组织。

AI本身并没有“赋能员工”或者“控制员工”的价值观。 它更像一个组织放大器

信任型组织得到的是更大的个人自主权;
控制型组织得到的是更低成本的控制。

以前一个管理者想知道员工每天做什么,要开会、日报、周报、盯过程。 现在 Agent 可以自动读取项目记录、分析工作量、评估产出、生成排名。

所以同一套技术:

一种组织会想:“既然两小时做完了,那剩下的时间就去做更有价值的事情。”

另一种组织会想:“既然两小时能做完,那一天是不是可以做四份?”

这不是想象中的风险。OECD 对算法管理的研究已经发现, AI和算法正在越来越多地参与工作分配、监控和评价; 而工作者参与程度低的时候,它往往伴随着自主性下降、工作量上升、 信任下降和压力增加。反过来,员工参与技术实施和决策时, 这些负面影响会明显减弱。1

所以,如果把它理解为“某个行业和社会的底色”,还可以再往前一步:

它未必是某个社会独有的底色,
而更像是低增长、强竞争、人才议价能力下降之后,
很多组织自然会滑向的一种管理倾向。

当增长很容易时,公司会说“创新、探索、试错”。

当增长困难时,管理者天然会开始关注“人效、成本、可量化、可追踪”。

而Agent恰恰让这些事情前所未有地便宜。

因此,未来真正的分水岭甚至不是:

AI公司 vs 非AI公司。

而是:

把AI用于增加人的能力的公司 vs 把AI用于增加组织控制力的公司。

现实中,绝大多数组织大概都会处在两者之间。

CHAPTER 02不要把事业规划建立在“找到一家好公司”上

年轻的时候很容易有一个潜意识: 找到一个价值观好的公司、一个有远见的管理者、 一个尊重人的团队,然后在那里发挥创造力。

当然,最好能遇到。

但如果把人生建立在这个前提上,其实非常脆弱。

管理者会换。

公司会遇到经营压力。

组织会变大。

行业会进入下行周期。

甚至一个原本很理想主义的创始人,到了现金流紧张的时候, 也可能开始疯狂抓人效。

所以更稳的事业策略,应该从:

“寻找一个不会压缩我的组织”

变成:

建立一种即使组织开始压缩我,
我仍然拥有选择权的能力结构。

这两个思路差别非常大。

前者依赖别人。

后者建立的是自己的议价权。

CHAPTER 03AI时代,个人应该刻意积累四种东西

01 · 不要只积累“AI能力”,要积累「AI杠杆」

会ChatGPT、会Codex、会做Agent,很快都会变成普通技能。

NBER研究客服Agent时发现,生成式AI对经验较少、技能较低员工的提升尤其明显: 平均生产率提高约14%,新手和低技能员工提升更大。 换句话说,AI有一个很重要的作用,就是快速复制优秀员工的一部分方法和经验。 2

过去:

“我特别会做PPT。”

“我特别会整理资料。”

“我Excel特别快。”

“我特别会写报告。”

这些曾经都能形成明显竞争力。以后很可能越来越难。

真正值钱的是:为什么做?做什么?结果可信吗? 客户真正的问题是什么?怎样让系统长期运行?

因此不要把自己培养成:

最会操作AI的人。

而要培养成:

最懂某个问题,
并且知道怎样让AI持续解决这个问题的人。

这两个职业价值差很多。

02 · 把“判断力”当成核心资产

AI越强,反而越值得看重一个词:judgment

判断哪些信息值得相信。

判断一个项目真正的问题。

判断客户嘴上说的需求和真正需求有什么区别。

判断什么时候应该自动化,什么时候必须人工介入。

判断一个分析结果虽然逻辑成立,但现实中根本不会发生。

这些东西非常难写成Prompt。

ILO对AI就业影响的判断也不是简单的“大规模职业消失”, 而是大量职业会被重新拆解和改变; 真正重要的问题逐渐变成工作如何被重组、自主性如何变化、哪些任务留给人。 3

AI负责把可能性快速铺开,
人负责承担判断。

而承担判断的人,天然比执行判断的人更靠近价值链上游。

03 · 积累“带得走的东西”

这是很多职场人最容易忽略的。

工作十年以后,最危险的一种状态不是能力不强。

而是:

你很强,
但所有强都只能在这家公司里成立。

公司系统你懂。

公司流程你懂。

管理者偏好你懂。

公司内部关系你懂。

一旦离开公司,重新归零。

所以可以用一个问题长期审视自己的工作:

今年做的这些事情,
有多少会变成三年以后即使离开公司仍然属于我的能力?

不是带走公司的数据或者商业机密,而是带走: 方法论、行业理解、问题框架、项目经验、作品、公开表达、 个人声誉、人脉、产品思维、技术能力。

这才是真正属于个人的“资产负债表”。

工资属于当期收入。可迁移能力才属于资产。

04 · 最重要的是 Optionality——选择权

这是未来职业规划中被严重低估的一件事。

真正的安全感不是:

“公司暂时不会裁我。”

而是:

即使明天离开这里,
我也知道自己还能去哪里。

选择权可能来自很多地方: 专业能力、行业声誉、客户网络、储蓄、作品、一个小产品、 一个长期研究领域、个人品牌,甚至几个真正认识你能力的人。

它们共同构成一种东西:离开的成本。

当离开的成本很高,人会被很多荒谬的组织规则绑架。

当离开的成本越来越低,在公司里的状态也会发生变化。

不一定会天天想着离职。恰恰相反,反而会更从容。

CHAPTER 04面对一家“AI越来越像控制工具”的公司,能做什么?

不必立刻把它上升成价值观冲突。

因为“教育公司”通常是一件收益非常低的事情。

更现实的方法,是观察三个东西:

AI带来的效率有没有换来自主权?

能力提升以后,有没有换来更复杂、更有价值的问题?

贡献增加以后,有没有换来资源、收入、影响力或者职业空间?

如果其中两项长期都是“没有”,就应该开始警惕。

例如:以前报告8小时,现在Agent两小时完成。

比较好的结果是:

剩下时间被用于研究新的问题、见客户、做产品、建立方法。

勉强可以接受的是:

承担更多项目,但职位、收入、能力范围同时扩大。

最值得警惕的是:

从一天一份报告,
变成一天四份报告。

本质上还是那个问题: AI省下来的时间,最后给了谁?

个人可以接受一段时间利用这样的组织训练自己,但不要无限期停留。

甚至可以把公司分成两种:“训练场”与“归宿”。

有些公司不一定值得长期待,却是很好的训练场: 资源多、项目多、数据多、问题多。

那就在那里快速积累。

不是等待一个组织变成理想的样子,
而是在其中完成自己的能力资本积累。

这个心理变化其实很重要。

CHAPTER 05事业方向:不要成为纯粹的“AI工具人”

AI时代有一个很容易掉进去的误区: 把职业升级理解成“再多学几个AI工具”。

真正更有价值的路线,往往是形成一种交叉能力:

行业业务理解 × 数据 × AI × 产品化能力

纯AI技术的竞争会越来越激烈,模型本身也会不断降低技术门槛。 但一个人如果既理解真实业务,又能把AI从Demo变成稳定工作流, 价值反而更难被替代。

特别是在酒店、咨询、制造、零售、文旅、专业服务等传统行业里, 真正稀缺的往往不是“知道一个模型名字的人”, 而是知道业务为什么这样运转、数据在哪里、决策如何发生, 并且能把AI嵌入流程的人。

如果把这条职业路线画成一条线,它可能是:

业务执行 / 数据分析 / 专业岗位 AI赋能具体业务 AI × 行业的产品与流程设计 AI Transformation / AI Product / 行业智能化负责人 企业内部负责人 / AI咨询 / 独立产品 / 创业

它最大的好处是:

不需要押注“哪一家公司的AI战略会成功”。

真正押注的是:

未来十年,传统行业一定需要一批既懂行业,
又真正知道怎么把AI嵌入组织的人。

纯懂行业的人很多。

纯懂AI的人也会越来越多。

懂行业、懂数据、懂Agent,还真正做过组织落地的人,会少很多。

这才是一种值得长期保护的能力组合。

CHAPTER 06把最初的问题,再往前推一步

AI化很容易被理解成一种“怕被时代淘汰”的动作: 大家都在做,所以公司也必须做。

但如果只停在这里,还是太浅。

AI不会天然让公司变得更先进,
它只会让一家公司更有能力成为它原本想成为的样子。

控制型组织会拥有更强的控制;

效率型组织会拥有更高的效率;

创造型组织会拥有更大的创造力;

尊重人的组织,会让普通人的能力第一次被大幅放大。

而对个人来说,答案不是等待最后一种公司出现。

而是:

让AI先放大自己。

放大自己的学习速度,

放大自己的行业理解,

放大自己的创造能力,

放大自己可以独立完成事情的边界,

最终放大自己的选择权。

这样即使外部世界的底色没有变,人与它之间的关系也会变。

未来五到十年职业发展的一个好目标, 或许不再只是传统意义上的:

“我要爬到什么职位。”

而是:

有多少事情,
即使没有这家公司,我仍然可以做到?

当这个答案越来越多的时候,职位反而没那么重要了。

ONE LAST LINE

不要把事业押在公司会不会尊重你;
把事业押在自己能不能逐渐成为一个拥有选择权的人。

这可能比“找到一家更年轻、更懂AI、更开放的公司”更长期主义。

注释与延伸阅读
  1. OECD:Algorithmic Management in the Workplace(2025),关于算法管理、员工自主性、工作量与参与机制。
  2. NBER:Generative AI at Work,关于生成式AI在客服工作中的生产率影响。
  3. ILO:Generative AI and Jobs: A 2025 Update,关于生成式AI对职业任务与工作结构的影响。
  4. 数字生命卡兹克: 《创业2年半后,想跟你分享关于AI组织的这7点心得》
SLOW NOTES
ISSUE 01 · 2026.08
ESSAYS ON WORK, AI & CHOICE

In the Age of AI,
Becoming Someone Who Has the Power to Choose

When AI amplifies both individual productivity and organizational control, the more meaningful question may no longer be which company is more “advanced,” but how one might preserve judgment, growth, and the freedom to leave within a new structure of work.

EDITOR'S NOTE

Some questions seem, at first glance, to be about one company, one style of management, or one generation’s particular understanding of technology. Yet with another turn of thought, they reveal something larger: a shared professional condition of our time.

The Question · Beginning with a discussion of AI Agents
I recently read “Two and a Half Years into Building a Company: Seven Things I Want to Share About AI Organizations,” by Digital Life Kha’Zix. It brought to mind a striking paradox: the same AI that can reduce repetitive work and expand an individual’s capabilities and creative room can also make task allocation, performance measurement, and process management more granular than ever before.

At first, it is easy to read this difference as a matter of one company’s management style, or of how a certain kind of manager understands technology. But on further reflection, perhaps it is not merely a choice made by a handful of individual firms. It may have just as much to do with industry competition, pressure for growth, organizational governance, and the wider way our era structures work. If this is a question larger than the distinction between a “good company” and a “bad one,” then how should individuals make sense of this change? What can we do within different kinds of organizations? And how ought we to plan the long arc of our professional lives?

AIhas brought one question increasingly into focus as it enters the workplace: once efficiency rises, where does the saved time actually go?

It may, of course, free people from repetitive labor. Work that once took a full day can now be completed in a few hours—or even a few minutes— leaving more room for judgment, creativity, communication, and exploration.

But another possibility exists as well.

Five tasks a day become twenty; what managers once saw only at the level of outcomes can now be followed in real time; work that used to resist quantification because management costs were too high can now be broken down, tracked, and evaluated by agents.

AI does not, by nature, lead people toward freedom.
It can amplify the individual, but it can also amplify the organization.

And so the question shifts from whether a certain company “understands AI” to something larger:

In an age when AI is causing both organizational control and individual productivity to surge,
how can an ordinary person avoid becoming nothing more than a more efficient unit of production?

That, I think, is a question worth sitting with.

Chapter 01This Is Not Merely About One Particular Company

If we explain this phenomenon only in terms of organizational inertia, management style, or even differences in how generations approach technology, we may still be looking too narrowly.

A younger team, after all, can just as easily become a different kind of highly digitized, highly quantified, highly controlling organization.

AI itself carries no moral preference for “empowering employees” or “controlling employees.” It behaves more like an organizational amplifier:

In a trust-based organization, it expands personal autonomy;
in a control-oriented one, it lowers the cost of control.

In the past, a manager who wanted to know what employees were doing each day needed meetings, daily reports, weekly reports, and endless process-tracking. Now, agents can automatically read project records, analyze workloads, assess output, and generate rankings.

So with the very same technology:

One organization thinks: “If the work is done in two hours, the rest of the time can go toward something more valuable.”

Another thinks: “If it can be done in two hours, shouldn’t one day now hold four times as much work?”

This is not an imaginary risk. OECD research on algorithmic management has shown that AI and algorithms are increasingly involved in task allocation, monitoring, and evaluation. Where workers have little say in how these systems are implemented, the result often includes less autonomy, heavier workloads, lower trust, and higher stress. Where employees do participate in implementation and decision-making, those negative effects are significantly reduced.1

So if we want to call this the “background tone” of a certain industry or society, we might go one step further:

It may not be the distinctive tone of one society at all,
but rather a managerial tendency that many organizations drift toward
when growth slows, competition intensifies, and workers lose bargaining power.

When growth comes easily, companies speak of innovation, experimentation, and exploration.

When growth becomes difficult, managers instinctively begin to focus on efficiency, cost, measurability, and traceability.

And agents make all of those things cheaper than ever before.

That is why the real dividing line ahead may not even be:

AI companies vs. non-AI companies.

It may be this instead:

companies that use AI to enlarge human capability vs. companies that use AI to enlarge organizational control.

In reality, of course, most organizations will fall somewhere in between.

Chapter 02Do Not Build Your Career Around the Hope of Finding a “Good Company”

When we are young, many of us quietly carry the same assumption: find a company with sound values, a far-sighted leader, and a team that respects people— and then do meaningful, creative work there.

And of course, if such a place can be found, all the better.

But to build a life around that expectation is more fragile than it appears.

Managers change.

Companies run into pressure.

Organizations expand.

Industries enter downturns.

Even a founder who begins as an idealist may, under cash-flow pressure, turn obsessively toward managing efficiency.

So a steadier career strategy must shift from:

“finding an organization that will not compress me”

to:

building a structure of capability that still leaves me with choices
even when the organization begins to compress me.

The difference between those two ways of thinking is immense.

The former depends on others.

The latter builds one’s own bargaining power.

Chapter 03In the Age of AI, There Are Four Things Individuals Should Deliberately Accumulate

01 · Do Not Merely Accumulate “AI Skills”; Accumulate AI Leverage

Knowing how to use ChatGPT, Codex, or build agents will soon become ordinary competence.

NBER research on customer-service agents found that generative AI especially boosts less experienced and lower-skilled workers: average productivity rose by about 14 percent, with even larger gains for newcomers. In other words, one of AI’s most important functions is to rapidly replicate part of the method and experience that once belonged only to high performers.2

In the past, people could say:

“I’m exceptionally good at making presentations.”

“I’m very good at organizing information.”

“I’m incredibly fast in Excel.”

“I’m excellent at writing reports.”

Those things once created clear advantages. They will become harder and harder to rely on.

What becomes truly valuable instead are different kinds of questions: Why is this being done? What exactly should be done? Can the result be trusted? What is the client’s real problem? How can a system continue to operate over time?

So do not train yourself to become:

the person who is best at operating AI.

Train yourself instead to become:

the person who understands a problem most deeply,
and who knows how to use AI to keep solving that problem.

The difference in professional value is substantial.

02 · Treat Judgment as a Core Asset

The stronger AI becomes, the more one quality stands out: judgment.

Judging which information deserves trust.

Judging what the real problem in a project actually is.

Judging the gap between what a client says they want and what they actually need.

Judging when automation is appropriate and when human intervention is indispensable.

Judging when an analysis is logically sound but impossible in the real world.

These things are exceedingly difficult to write into a prompt.

The ILO, too, does not frame AI’s impact on employment as a simple story of mass job disappearance. Its argument is that a great many jobs will be reconfigured and decomposed; what matters is how work is reorganized, how autonomy changes, and which tasks remain human.3

AI can spread possibilities across the table with extraordinary speed;
human beings remain the ones who must bear judgment.

And the people who bear judgment inevitably sit closer to the upper end of the value chain than those who simply execute it.

03 · Accumulate What You Can Carry With You

This is one of the easiest things for professionals to overlook.

Ten years into a career, the most dangerous situation is not weakness.

It is this:

you may be very strong,
and yet everything that makes you strong only works inside this one company.

You understand the company’s systems.

You understand its processes.

You understand managerial preferences.

You understand its internal relationships.

And once you leave, everything seems to reset to zero.

So it is worth asking yourself, over and over:

Of the things I am doing this year,
how much will still belong to me as capability three years from now, even if I leave this company?

This is not about taking company data or trade secrets with you. It is about carrying away methods, industry understanding, problem frames, project experience, published work, public expression, personal reputation, relationships, product thinking, and technical ability.

That is what really belongs on a personal balance sheet.

Salary is current income. Transferable ability is an asset.

04 · Most Important of All: Optionality

This may be one of the most underrated ideas in career planning.

Real security is not:

“The company probably won’t let me go for now.”

Real security is:

Even if I leave tomorrow,
I know where else I can go.

That power to choose can come from many places: professional expertise, industry reputation, client networks, savings, a body of work, a small product, a long-term area of study, a personal brand, or simply a handful of people who truly know what you can do.

Together, these shape one thing: the cost of leaving.

When the cost of leaving is high, people become captive to absurd organizational rules.

When the cost of leaving grows lower, one’s entire state of mind inside a company begins to change.

You may not feel a constant urge to resign. Quite the opposite—you may feel more at ease.

Chapter 04What Can One Do Inside a Company Where AI Is Increasingly Becoming a Tool of Control?

There is no need to elevate it immediately into a full-scale conflict of values.

Trying to “educate the company” is usually a very low-return activity.

A more realistic approach is to observe three things:

Has the efficiency brought by AI translated into more autonomy?

Has the improvement in capability led to more complex or more valuable work?

Has increased contribution brought more resources, income, influence, or room for growth?

If two of those answers are consistently “no,” it is time to become cautious.

Suppose, for example, a report that once took eight hours can now be finished in two with the help of an agent.

A good outcome would be this:

The saved time is used to research new problems, meet clients, build products, and develop methods.

A tolerable outcome would be this:

You take on more projects, but your position, compensation, and scope of ability expand at the same time.

The outcome that should raise concern is this:

one report a day becomes four.

At bottom, it is still the same question: who ultimately receives the time that AI has saved?

You may use such an organization as a training ground for a while. But do not remain there indefinitely.

It may help to think of companies in two categories: training grounds and destinations.

Some companies may not be worth staying in for the long term, yet they are excellent places to train: rich in resources, projects, data, and difficult problems.

Then use them precisely for that purpose: accumulate quickly.

The point is not to wait for an organization to become your ideal one,
but to complete your own accumulation of capability capital within it.

That shift in mindset matters more than it first appears.

Chapter 05Career Direction: Do Not Become Merely an “AI Operator”

In the age of AI, one easy trap is to imagine career growth as nothing more than learning a few more tools.

What is often more valuable is to form an intersection of capabilities:

industry understanding × data × AI × the ability to turn things into products and processes

Competition in pure AI technology will only intensify, while the models themselves continue to lower technical barriers. But a person who understands real business and can turn AI from a demo into a stable workflow becomes harder to replace, not easier.

Especially in traditional sectors—hospitality, consulting, manufacturing, retail, culture and tourism, professional services— what is truly scarce is often not “someone who knows the name of a model,” but someone who understands how the business actually works, where the data sits, how decisions are made, and how AI can be embedded into the flow of work.

If one were to sketch that career path as a line, it might look something like this:

business execution / data analysis / specialized professional roles using AI to enhance concrete business functions AI × industry product and process design AI transformation / AI product / head of industry intelligence an internal leader / AI consultant / independent product builder / founder

Its greatest advantage is this:

You do not need to place your bet on which company’s AI strategy will win.

What you are really betting on is:

over the next ten years, traditional industries will inevitably need a group of people
who both understand the industry and truly know how to embed AI into organizations.

People who understand only the industry are abundant.

People who understand only AI will become more abundant too.

People who understand the industry, understand data, understand agents, and have actually carried out organizational implementation will be much rarer.

That is the kind of capability mix worth protecting over the long term.

Chapter 06Push the Original Question One Step Further

It is easy to understand “AI-ization” as a reactive move—something companies do because they fear being left behind by the times.

But if we stop there, the thought remains too shallow.

AI does not automatically make a company more advanced.
It simply gives that company greater power to become more fully what it already tends to be.

Control-oriented organizations gain stronger control.

Efficiency-oriented organizations gain greater efficiency.

Creativity-oriented organizations gain greater creative force.

Organizations that genuinely respect people may, for the first time, dramatically amplify ordinary human capability.

For individuals, then, the answer is not to wait for that last kind of company to appear.

The answer is this:

let AI amplify you first.

Let it accelerate how quickly you learn,

deepen your understanding of the industry,

expand your creative power,

extend the boundary of what you can accomplish on your own,

and ultimately enlarge your power to choose.

Then even if the background tone of the world does not change, your relationship to it will.

Over the next five or ten years, perhaps a better career goal is no longer the traditional one:

“What position do I want to climb to?”

But rather:

How many things can I still do,
even without this company?

As that answer grows, titles begin to matter less.

One Last Line

Do not stake your career on whether a company will respect you;
stake it on whether you can, little by little, become someone who truly has the power to choose.

That may be a more long-term way of thinking than simply trying to find a company that is younger, more AI-savvy, or more open.

Notes & Further Reading
  1. OECD: Algorithmic Management in the Workplace (2025), on algorithmic management, worker autonomy, workload, and participation mechanisms.
  2. NBER: Generative AI at Work, on the productivity effects of generative AI in customer-service work.
  3. ILO: Generative AI and Jobs: A 2025 Update, on the impact of generative AI on work tasks and employment structures.
  4. Digital Life Kha’Zix: “Two and a Half Years into Building a Company: Seven Things I Want to Share About AI Organizations”