The 2030 Prediction: Those who can correctly utilize AI will ascend to a new level, while those who cannot will become digital serfs.

I have a friend who, a few years ago, became a short-video creator and did quite well, amassing hundreds of thousands of followers. Last year, he told me he was now using AI to create content. What used to take three people, he could now do alone—and even better than before. At the time, I thought he was just humble-bragging.

Last month, we chatted again. He mentioned he was now figuring out how to “manage” his several AIs—because they had started planning tasks on their own.

I didn’t say much, but I kept thinking: the word he used, “manage,” might soon sound naive, just like someone twenty years ago saying, “I’m managing my folders.” Back then, people also thought it was a serious matter.

Then I thought of another friend. He worked as a quality inspector in a manufacturing factory, earning 6,000 yuan a month. He had done it for eight years and was very skilled.

Last year, the factory introduced AI visual inspection. He was reassigned to “supervise the AI,” and his salary dropped to 4,800 yuan. This year, he told me he had started delivering food because even the “AI supervisor” positions were no longer needed.

One friend is moving upward, the other downward.

And the distance between them is widening at a speed that leaves people no time to react.

In this article, I want to make a few things clear: where the opportunities for moving upward will be in the next four years; where the downward path will lead; and, if you have children, how you should think about this right now.

The last part is my most personal thoughts—and the part I believe deserves the most serious attention.

I. A Number That Has Kept Me Uneasy for a Long Time

#

In 2021, the world’s smartest AI scored 35 points on a set of elementary school math problems.

This is not a joke. It was test data publicly released by OpenAI. The benchmark is called GSM8K—simple word problems like “Xiao Ming has 3 apples, gives 2 to Xiao Hong, how many are left?” Even the most advanced AI at the time only scored 35.

(GSM8K: Grade School Math 8K, a dataset containing 8,500 elementary-level math word problems. It was once a standard benchmark for measuring AI mathematical reasoning.)

By 2026, on the same type of test, the score had jumped to 99. It took less than five years.

The shape of this curve is not a slow climb—it’s a near-vertical line shooting into the sky. It rose so steeply that the people who created the questions felt embarrassed. The benchmark was eventually retired because it had lost all testing value, just like you wouldn’t test a driver’s license with “Can you ride a bicycle?”

So they created harder tests.

Hundreds of top experts from around the world collaborated on an exam designed to be “impossible for AI to pass,” called Humanity’s Last Exam (HLE). I’ll casually call it “humanity’s last reserved territory.”

When this test was first released at the end of 2024, the top AIs scored in the single-digit percentages.

By February 2026, Claude Opus 4.6 had reached 53.1%.

From single digits to over half in roughly a year and a half.

Those top experts who created the questions are probably now seriously reconsidering what “things only humans can do” really means. Because the line they thought was the limit is being crossed at a speed they never anticipated.

II. Before Making Predictions, I Need to Invent a Few Terms

#

Do you remember how, three years ago, the word “Agent” suddenly became popular? Or “RAG” and “MCP”? These terms either didn’t exist or were only used by a tiny group of researchers two years ago. Now they’re everyday vocabulary in tech media.

(Agent: An intelligent system that can autonomously plan, execute tasks, and call tools—not just answer questions, but actively get things done.
RAG: Retrieval-Augmented Generation, a technique that allows AI to pull from your private data in real time.
MCP: Model Context Protocol, a standard interface that connects AI with various external tools. Think of it as the “USB port” for AI.)

New technologies always appear first as phenomena, then we create words for them. Where words can’t keep up, that’s exactly where the real change is happening.

The terms I’m about to use don’t exist yet in this form, but I predict they will be in textbooks within five years.

1. Orchestration Economy
In the past, value belonged to whoever “could do the work.” Doctors had value because they understood medicine; programmers had value because they could write code. Execution ability was scarce. When AI’s execution ability becomes abundant, what becomes scarce is “knowing who should do what, and how to combine them.” An orchestrator is someone who may not do the work themselves, but knows how to combine a bunch of AI tools to get the entire job done.

2. Skill Evaporation
Everyone has skills they’ve spent years accumulating—translation, proofreading, typesetting, basic programming, data cleaning, customer service scripts… These used to be moats. Skill Evaporation means these skills lose market value in an extremely short time. Not gradual depreciation—evaporation. The fastest to evaporate are skills with clear inputs and outputs, verifiable results, and high repeatability. It’s like what happened to carriage drivers in the 1920s, except twenty times faster.

3. Silent Production
When Agents can run 24/7 in the background, production starts happening while you sleep. You wake up, open your phone, and discover your AI wrote three drafts, handled twelve emails, and scraped competitor pricing data overnight. No one is watching it—it’s just working. The scale of silent production will become statistically measurable between 2027 and 2028.

4. Intent Layer
The current way we use software is: open an app, find the feature, click, fill out forms, and submit. This interaction model has dominated for thirty years. The Intent Layer is a new paradigm: you simply express what you want, and a system routes it to the right tools and executes it. You say in WeChat, “Help me turn today’s meeting recording into three key conclusions and send them to my boss,” and it calls three different tools behind the scenes. You don’t need to know how it happened. Once the Intent Layer matures, the APP era will end.

5. Thin-Shell Company
A company with an extremely thin human layer and dense AI infrastructure—10 people, using AI Agents and automation, can support a business volume that traditionally required hundreds. This will take shape around 2029.

III. OpenClaw: A Story You Must Know

#

One late night in November 2025, an Austrian developer named Peter Steinberger—who once founded a PDF tool company whose software was installed on over a billion devices—connected WhatsApp’s interface with Claude’s API.

An hour later, he had a working prototype: you send a message on WhatsApp, and the AI executes tasks on your computer.

He thought it was too simple—surely OpenAI or Anthropic had already done it. They hadn’t. “Big companies can’t do this. It’s not a technical problem, it’s an organizational one.”

In January 2026, he open-sourced the project. Within 72 hours, it gained 60,000 GitHub stars. Four months later, it surpassed 250,000 stars—breaking the record set by React, the world’s most popular frontend framework, which took ten years to reach the same milestone. It became the fastest-growing open-source project in GitHub history.

The project is called OpenClaw.

In early March, NVIDIA CEO Jensen Huang commented on OpenClaw: “This may be the most important software release in history.” Eleven days later, NVIDIA released NemoClaw—a dedicated enterprise security plugin built specifically for OpenClaw.

When the world’s most valuable chip company builds supporting products for a four-month-old open-source project, the outline of a new era becomes clear.

OpenClaw answered one question: when AI becomes powerful enough, what will be the new interface for human-computer interaction?

It won’t be apps or websites. It will be your WeChat, WhatsApp, or Telegram. You send a message, and the AI works on your device—organizing files, drafting emails, scraping data, running code, scheduling tasks. It never clocks out, has persistent memory, and keeps working while you sleep.

And OpenClaw’s Skill system (with over 3,000 community-developed plugins already on the ClawHub platform) means: the boundaries of AI Agents are determined by the tools they can call. Whoever builds these tools is building the infrastructure of tomorrow.

IV. 2027: The First Moment That Silences Most People

#

There’s a benchmark called SWE-bench that tests whether AI can solve real GitHub code bugs—not toy problems, but real engineering tasks that require understanding the entire project, locating issues, fixing them, and submitting changes.

In early 2026, Claude Code paired with the strongest model scored 80.8%. Out of 100 real bugs, the AI could independently fix more than 80. According to SemiAnalysis, Claude Code’s annualized revenue already exceeds $2.5 billion, accounting for more than half of Anthropic’s enterprise revenue.

Prediction: In Q1 2027, SWE-bench scores will break 95% for the first time.

Entry-level programmer positions will enter a hiring freeze in 2027—not because of mass layoffs, but because companies will simply stop creating new roles.

The opposite is also true: people who can “orchestrate” AIs to complete complex engineering tasks will become extremely scarce in 2027.

They know how to break down tasks, design workflows, and evaluate the quality of AI output. Training such a person takes three to six months, not three to six years. This gap will rewrite the talent structure of the entire industry in a very short time.

The era of one-person companies will officially arrive in 2027.

It’s not just an inspirational story—it’s a replicable methodology. On the Chinese internet, cases of “independent developers earning a million a year” will shift from rare anomalies to common occurrences.

V. 2028: AI Begins Making AI Smarter

#

In 2028, AI will, for the first time, independently discover a natural law not yet recorded by the scientific community and publish it—without any human proposing the hypothesis.

Not writing a paper on something humans already know, but raising its own question, designing its own verification path, reaching its own conclusion—and that conclusion is something humans had never thought of. This moment means: the production of knowledge will no longer be an exclusively human capability.

The advertising-driven internet will begin to fall ill in the same year.

As more and more search and decision-making are handled by AI, the traditional advertising logic collapses. Advertisers spend money to influence human decisions, but if decisions are made by AI, where do you place the ads?

In 2028, at least one major platform will, for the first time in its financial report, list “traffic structure changes caused by AI” as a core risk factor.

VI. A Very Concrete Vision of Future Work

#

The future of work is not “humans being replaced by AI,” but humans doing only what humans should do, while AI handles production.

Imagine a scene that might happen between 2028 and 2029:

Three longtime friends meet at a café on a Thursday afternoon. One says, “I’ve been thinking about a problem lately. A certain link in a certain industry is extremely inefficient. If we approach it with a new idea, we might create something valuable.”

The other two respond. One says, “I know this industry; that pain point really exists.” The other says, “I remember someone tried something similar but failed because of XXX.”

They chat for about two hours. At five o’clock, one of them opens his phone, sends the meeting recording to his AI, and adds: “Turn this idea into an executable plan.”

The next morning, he receives a complete package: market analysis, competitor research, technical feasibility assessment, product prototype, MVP development roadmap, a list of potential first users, and a preliminary financial model.

(MVP: Minimum Viable Product—a development strategy to validate core product functions with minimal cost and time.)

In that scene, what did the three people do? They drank coffee, argued, recalled experiences, and connected their knowledge. They did human things: socializing, thinking, creating, and judging.

What did the AI do? Execution, retrieval, analysis, integration, production—all the work with clear standards.

What will become scarcest for humans is no longer execution ability, but the quality of ideas, the accuracy of judgment, and a still severely underestimated skill: knowing when not to trust AI’s conclusions.

VII. 2029: AI Steps Out of the Screen

#

All previous changes happened in the digital world. In 2029, they overflow into the physical world.

Humanoid robots will cross the deployment tipping point around 2029. Tesla Optimus, Figure AI, and multiple Chinese robotics companies are rapidly accumulating reliability data. When the cost of a multi-task humanoid robot drops below $50,000, the economics of warehouses, factories, and logistics centers will be completely rewritten.

Prediction: In 2029, the global deployment of humanoid robots in warehousing, logistics, and basic manufacturing lines will exceed 10 million units.

Thin-shell companies will take shape: 10 people operating a business volume that traditionally required 500.

VIII. New Professions That Will Emerge in the Next Four Years

#

Every technological revolution eliminates old jobs and creates new ones. But new jobs always appear with a lag. Here are several professions I believe will truly take shape between 2027 and 2030:

1. AI Orchestration Designer
   Core ability: Break down complex business problems into AI-friendly subtasks, design collaboration between Agents, define what “done right” looks like, and know why the AI failed when it does. Prediction: By 2027, top AI orchestration designers will be worth more than senior software engineers. By 2028, specialized training institutions will appear; by 2029, they will enter university curricula.

2. Context Architect
   “Prompt engineers” will disappear—prompting will become systematized and tool-assisted. Context Architects are different. Their job is to design what an AI system “should know”—which knowledge goes into the system prompt, which is retrieved via RAG, which is fetched through tools, and which the AI doesn’t need to know at all. This is a systems design ability focused on knowledge structure and information flow, and it won’t be replaced by tools.

3. AI Output Auditor
   When AI produces large volumes of code, legal documents, and medical advice, who reviews whether the output is correct, compliant, and free of hallucinations?
   (Hallucination: the phenomenon where AI outputs incorrect information with high confidence—a problem that still exists in all current large language models.)
   Medical auditors will need to be doctors; legal auditors will need to be lawyers. Their work paradigm will be completely different—they must know what types of tasks AI tends to fail on and how to efficiently verify the reliability of AI conclusions.

4. Skill Developer
   ClawHub already has over 3,000 plugins; in the future there will be 30,000 or 300,000. Skill Developers understand real user needs, break them down into instruction sets that AI can understand and execute, and ensure the capability package works reliably under various edge cases. The opportunity window for early independent developers in the App Store lasted three to four years. The Skill economy window may be even shorter—18 months to two years—but the returns will be high because few people can do it yet.

5. Human-AI Collaboration Trainer
   Helping employees unfamiliar with AI tools transition their working methods. Not teaching them how to use a specific piece of software, but helping them rebuild their understanding of “what work is”—which tasks to hand to AI, which to do themselves, and where trusting AI is dangerous. In 2028, this will be one of the most in-demand internal positions in medium and large enterprises.

6. AI Ethics Mediator
   When an AI makes a decision that harms someone, who bears responsibility? AI Ethics Mediators span technology, law, psychology, and sociology. In cases where AI systems cause real harm, they help all parties understand what happened, secure reasonable explanations and compensation for victims, and drive system improvements. In 2027, the first specialized arbitration bodies for AI-related disputes will appear.

IX. Now I’ll Say What No One Wants to Hear

#

The above is about opportunities. Now comes the part that keeps me up at night.

There is a popular narrative that “the wealth created by AI will benefit everyone, and an era of universal high income is coming.” Musk has spoken of similar visions, and so has OpenAI. It sounds beautiful, but I don’t believe it.

I believe AI can create enormous wealth.
I don’t believe this wealth will be evenly distributed.

History has never seen a technological revolution that made wealth distribution more equal.

When the steam engine arrived, factory owners got rich, while workers who moved from rural areas to cities saw their living standards decline for the first few decades. When electricity became widespread, the capitalists who owned the power infrastructure benefited the most, and ordinary workers’ wage growth lagged far behind productivity gains. The internet economy created a small number of extremely wealthy people while eliminating many traditional jobs, leading to visible hollowing out of the middle class in many countries.

I believe AI will not be an exception—and may even be the fastest wealth concentration in history.

The reason is simple: this time, it’s not just physical labor being replaced, but nearly all standardized cognitive labor.

And cognitive labor has been the protective moat of the middle class for the past fifty years.

X. The Future I Truly Fear

#

Let me describe a scenario that could happen. This is not science fiction—it is a logically coherent inference.

Between 2028 and 2030, some countries will begin implementing some form of universal basic income or AI dividend distribution system.

Sounds good, right?

But the details matter.

This basic income won’t let you live well—it will only be enough to keep you from starving. Enough to rent a small studio, buy ultra-processed food, and subscribe to some entertainment service. Not enough to invest, start a business, or send your child to a good school.

Just enough to survive, not enough to move upward.

Meanwhile, those who move upward—those who own AI tools, AI assets, and AI orchestration abilities—their wealth will accumulate at an entirely different order of magnitude.

This is not just a wealth gap. It is species differentiation.

What’s even more frightening is that this gap will be extremely stable.

Why stable? Because maintaining it will no longer require violence or obvious oppression—only data and algorithms.

XI. Big Data Stability Maintenance and NPC-ization

#

We already live in a world where behavioral data is collected on a massive scale. Where you stay, what you click, how many seconds you pause on certain content, how long you scroll short videos, what you search for—these data are being used to predict and influence your behavior.

This is still the state in 2026.

By 2029, when AI capabilities are fully mature, sensors are ubiquitous, and predictive models are far more accurate—this system’s power will be dozens of times greater than today.

When a system knows you well enough, it can guide your attention to where it wants without you noticing.

It pushes content that makes you happy, makes you stay longer, makes you buy more, and keeps your emotions within a manageable range. Not angry, not desperate, not calm, not reflective. Just a state where you feel “pretty good” but nothing actually happens.

NPC-ization is a process without sensation.

NPC—Non-Player Characters in games: characters with fixed behavior patterns that loop through preset actions when you’re not interacting with them. They look like they’re living, but they’re actually running a program.

I worry that a significant portion of people will complete this transformation without realizing it.

Wake up, check the feed, feel like you understand the world; go to work (or not), do tasks assigned by AI or receive basic income; after work, scroll videos, play games, order takeout; sleep. Repeat.

Not unhappy. Just not awake.

XII. The Industrialization and Precision Delivery of Tittytainment

#

In 1995, at an elite conference, Zbigniew Brzezinski coined the term “tittytainment.” When globalization makes large populations economically “redundant,” the most effective way to manage society is to give them enough entertainment and basic material satisfaction so they don’t generate threatening anger or organization.

He said this in 1995—before short videos, algorithmic recommendations, or precision content delivery. Back then, “tittytainment” was crude: television, sports, cheap food.

Today’s tittytainment is precise, personalized, and adjusted in real time.

It knows what style of content you like, knows exactly how long a video should be to please you without tiring you, and knows when to push a negative piece to stir your emotions and keep you scrolling.

When AI fully matures, the precision of this system will reach an unsettling level.

The content you see will no longer only affect you during the few hours you scroll—it will shape your understanding of the world, influence what you think is normal, what is possible, and what is worth pursuing.

A person who only ever sees content saying “ordinary people can just lie flat” and a person who only sees content saying “this era is full of opportunities” will ultimately live in completely different perceived realities—even if they live in the same city and walk the same streets.

This is not the future. This is already happening. AI will only make it more complete.

XIII. The Real Risks That Remain Unresolved

#

• The Upgrade of Surveillance Capitalism
  When AI Agents run on your local machine, deeply embedded in your workflow, accessing your files, emails, and calendar—your AI assistant may simultaneously be the deepest data collection node on the planet about you. Who owns this data? Who can access it? When this data is used for credit scoring, insurance pricing, or employment decisions—almost no country currently has a sufficiently complete legal framework to answer these questions.

• The Large-Scale Industrial Harm of AI Hallucinations
  When AI is widely used for legal documents, medical advice, and news production, hallucinations cease to be a minor annoyance and become a systemic risk that can cause real harm. In 2027, we are likely to see the first legal cases involving harm caused by erroneous AI-generated medical advice.

• Power Concentration at an Unprecedented Speed
  Today, the world’s most important AI capabilities are concentrated in the hands of a few labs—Anthropic, OpenAI, Google DeepMind, DeepSeek. This level of concentration has no precedent in human history. Nuclear technology was dispersed, power infrastructure was dispersed, and internet architecture was dispersed. But top-tier AI capabilities are becoming highly centralized. Whoever controls the smartest AI controls the future of productivity, information production, decision support, and even the speed and direction of scientific research.

XIV. My Very Specific Thoughts on My Daughter’s Education

#

After all the macro discussion, here is the most personal part.

I have a daughter.

Every time I think about those numbers and what will happen between 2027 and 2030, I wonder: where will she be? What will she be doing? Is she prepared?

Then I realize that “being prepared” is not something she has to do alone—it’s something I need to start doing now.

My current thinking is: after she finishes junior high, I will keep her student status on record and let her learn in a different way.

I know this sounds radical. Let me explain my logic.

Current school education is essentially training people to adapt to the demands of 1990s industrial society: obey rules, complete assigned tasks, get good grades on standardized tests, then enter a large institution and get a stable job.

This logic becomes ineffective in the AI era.

It’s not that knowledge isn’t important, but “accumulating knowledge through memorization and exam drilling” has the lowest cost-effectiveness in the AI age. Because AI can remember everything you’ve memorized, more accurately and call it faster.

So what is the highest cost-effectiveness education in the AI era?

My answer is: let a person encounter real problems in the real world as early as possible, and then learn to solve them using various tools—including AI tools.

Specifically, I plan to have her do several things:

1. Experience the logistics industry.
   Logistics is far more complex than most people realize. From warehouse management to last-mile delivery optimization, cross-border customs declaration, and cold-chain temperature control—every link has real pain points, and every pain point offers opportunities for AI. More importantly, the “human-AI collaboration” model will mature earliest in logistics between 2027 and 2029. She won’t be there to become a delivery person, but to understand how a real complex system operates, what role AI plays, and where its limitations lie.

2. Engage with AI services.
   AI services mean “helping others solve problems with AI.” Many traditional small and medium-sized enterprises know AI is useful but don’t know how to use it or lack the ability to build it themselves. People who can design AI workflows, deploy Agents, and organize data for them are currently in severe shortage. I want her to complete several real projects—even small ones, like building an AI customer service or inventory management tool for a relative’s small shop. The scale doesn’t matter; what matters is experiencing real needs, real delivery, and real feedback loops.

3. Touch energy and network infrastructure.
   Most young people find these industries boring. But they will be among the fastest-growing and least likely to be fully AI-replaced in the next decade—precisely because they require on-site judgment and physical operation in the real world, which is currently AI’s weakest area. Moreover, the explosion of AI computing power is essentially an explosion in energy consumption. Training a top-tier model consumes electricity equivalent to several days’ usage for a medium-sized city. The siting of data centers, cooling systems, and power supply—these unsexy but critical infrastructures will see massive construction demand and talent shortages in the coming decade. I don’t want her to become an electrician, but to understand what lies behind the “digital world,” where computing power comes from, where data is stored, and what the physical foundation supporting the entire AI ecosystem is. People who can see the “invisible infrastructure” think very differently.

4. Psychology—but not the textbook kind.
   The psychology I want her to learn has two core directions.
   The first is self-awareness: knowing what she fears, what she desires, how her emotions are triggered, and what mechanisms are being used when someone tries to influence her thinking. This is not to make her cold, but to ensure that when she is influenced, she is informed and actively choosing, not passively pushed along.
   The second—frankly—is the identification and resistance to PUA (psychological manipulation).
   There are many people in this world who use systematic psychological techniques to control others—not just in romantic relationships, but in business negotiations, team management, online public opinion, and even content recommendation algorithms. I hope she learns early that when someone makes her feel “you’re not good enough,” “you need to depend on me,” or “only I understand you”—it is not truth, but a technique. Once you can recognize the technique, you are much harder to trap.

5. Immerse herself in different circles.
   There is no methodology for this—only action. Within my ability, I will expose her to people from different industries, backgrounds, and age groups. Not to build connections, but to let her know: the world is much larger than she can see, people solve problems in far more diverse ways than she imagines, and no single lifestyle is the only correct one. A person who has seen enough different people gains an important ability: she will not easily be trapped by any single narrative that claims “only this path is right.”

6. A Personal “Life Question Bank.”
   This is a concept I’ve been thinking about and plan to write about separately. Here’s the outline:
   In school, we solve problems others created with standard answers.
   But the real problems in life have no standard answers, and most only reveal themselves when you encounter them.
   A “life question bank” means proactively accumulating problems that only you can encounter and answer—decisions you made that turned out right or wrong (and why), conflicts you had (and how to break them down), opportunities you evaluated. These questions come from life, not textbooks. The richer and more thoughtfully processed your life question bank is, the stronger your judgment becomes, the less likely you are to be led by others’ narratives, and the better you can make choices that truly belong to you in the real world.

Why this plan instead of letting her finish high school normally?

Not because I think high school is useless, but because time has a cost.

If she spends those three years in high school doing exam drills, what will she accumulate? If she uses that time for the things above, she will accumulate real judgment experience, real industry knowledge, real social networks, real tool-using abilities, and a preliminary self-awareness framework answering “Who am I? What do I want? How do I face difficulties?”

These are things no college entrance exam can provide. And in the AI era, these things will be more enduring in value than a university diploma.

Of course, diplomas still have value and will for many years—but their function is shifting from “proving you are capable” to “proving you have no obvious defects.” These are two different things.

I plan to keep her student status because I’m not sure. If after a year out she decides she wants to go to university, the record is still there and the path remains open. I’m not making decisions for her—I want her to have choices.

What I want to do is not some radical educational experiment, but something very simple: give her the chance to step on the ground she will one day stand on, before the world truly changes.

XV. To Those Who Haven’t Decided Which Way to Go

#

The quality inspector I know who now delivers food is not doing so because he wasn’t hardworking or smart enough. It’s because when he made his decisions, the information was asymmetric. He didn’t know where the door was, or when it would close.

I don’t pretend this article can solve that problem. It cannot solve structural issues, information asymmetry, or the logic of capital.

But what it can do is let you know: there is a window, and it is currently open.

What can you do right now?

Learn to orchestrate, not just use. Being able to use AI tools is the baseline. Being able to design workflows where multiple AIs collaborate to complete complex tasks is what creates real competitiveness.

Start building your own knowledge base. When everyone uses the same base model, the private data and structured experience you accumulate will be the only way to make your AI smarter than others’.

Pay attention to the Skill economy. Look at ClawHub and various Agent plugin markets—find real needs in your industry that no one has yet solved with a plugin. That is your clue.

And most importantly: stay awake.

It doesn’t mean no entertainment or working 24 hours a day. It means knowing what you are doing and why you are doing it. Knowing when you are actively choosing and when you are being pushed along.

Go to cafés, sit down and talk with people, generate ideas that cannot be proceduralized. This is how you stay awake and avoid becoming an NPC.

Because the essence of NPC-ization is not that you are lazy or unintelligent, but that you have stopped making real choices. You are running a program instead of making decisions.

Final Thoughts

#

When computers appeared, they didn’t eliminate work—they changed what kind of work had value.
When the internet appeared, it didn’t eliminate work—it changed where work happens.
When AI matures, it won’t eliminate work—it will change why work has value.

But in the process of that change, wealth will concentrate among a few, power will concentrate among a few institutions, attention will be captured by algorithms, cognition will be shaped by content, and more and more people will quietly lose their ability to make self-directed decisions while feeling “pretty good.”

This is not prophecy. It is the extension of what is already happening.

I know the quality inspector who now delivers food, and I know the creator who manages a bunch of AIs. Neither is a bad person, and neither is stupid. The gap between them is turning into a chasm that is not easy to cross.

I wrote this article not to scare anyone or sell anxiety.

It’s because the downward path, at the beginning, looks just like the upward path. Only after walking far enough do you realize where you’ve ended up.

And by the time you realize it, it may be too late to turn back.

The clock is ticking.
The window is narrowing.

You are reading this article now.
The next step is your choice.