In 2026, the first real shock AI delivered to employment no longer lives in the realm of prediction. It has entered a colder phase: major companies are openly cutting jobs, capital markets are openly rewarding them, and even profitable companies are reducing headcount anyway.
In the first quarter of 2026, Oracle had already laid off a cumulative 25,000 employees.
In January 2026, Amazon confirmed another 16,000 job cuts; since October 2025, its total layoffs had reached roughly 30,000.
In March 2026, Reuters reported exclusively that Meta was preparing a large new round of layoffs, one that could affect at least 15,800 people.
In March 2026, Dell disclosed that its workforce had fallen from 108,000 to about 97,000—a reduction of roughly 11,000.
In February 2026, Block cut more than 4,000 jobs in one stroke, and the market’s reaction was not fear but a rally.
To stay precise: not every one of these layoffs proves that “AI has already fully replaced the jobs in question.” But together they say something very clearly: labor budgets are being systematically redirected toward compute, automation, and smaller teams.
Execution Capacity Is Detaching From Headcount
Pull the camera back a little farther, and the scene no longer feels new at all.
On December 15, 1811, The London Statesman, writing about Nottingham’s knitting industry, reported that 20,000 textile workers had lost their jobs, 900 lace machines had been smashed, and the government had sent six regiments into the city. The piece ended with a line that still lands today: “God only knows what will be the end of it; nothing but ruin.”
Two hundred years later, we are standing at the same intersection again.
But what actually matters in 2026 is not that some model broke another benchmark, or that some AI demo looked impressive. It is a quieter fact, a more practical one, and a more dangerous one:
For the first time, execution capacity is starting to detach itself from “people you have to hire” and become a resource ordinary individuals can also command.
That sounds abstract, but it changes the meaning of the word company.
We Accepted New Tools. We Still Live Inside an Old Idea of the Company
There is a strange phenomenon: for the past few years, almost no one has seriously updated the way they imagine a company.
Of course people know AI is smart. They know it can write emails, write code, create content, do research, run workflows. They also know that some people have already started using AI to manage websites, organize customers, handle replies, generate articles, and even replace part of an operations role.
But the moment the word company comes up, people still default to the same old picture: offices, org charts, layers of approval, cross-functional coordination, hiring, reporting, performance reviews, meetings.
In other words, we have accepted new tools while still living inside an old organizational imagination.
That is not a small problem. The way people use tools usually depends on the way they imagine organizations.
If, by default, you assume a company naturally has to be built by ten people, twenty people, a hundred people working together, then your use of AI will probably amount to nothing more than adding a cheaper plug-in to an old system. You will ask AI to polish copy, summarize meetings, fix a few lines of code, make a slide. It will function like a somewhat faster intern, inserted into an old workflow to save you a little time, but not to change anything fundamental.
But if you start to question something else—if many companies needed so many people not because the work itself inherently required that many, but simply because, until now, there had never been an execution layer cheap enough, callable enough, and replicable enough—then the way you look at AI changes completely.
You start to realize that what is truly scarce inside a company has never really been “the people who do the work” themselves. It has always been three things:
– Who defines the task.
– Who decides the priority.
– Who bears the outcome.
In the past, those three things were almost always bound up with “hiring lots of people.” Because if you wanted to scale what you were doing, you had to scale execution outward.
Pieter Levels: One Person Defining Problems, Systems Catching Execution
If I had to pick one archetype, I would start with Pieter Levels.
Not because he tells the best story. He did not get here by raising money, expanding teams, or mastering organizational management. He got here by continuously shipping products like Nomad List, Remote OK, and Photo AI—largely on his own—and by publicly posting his revenue for years.
What ordinary people can learn from him is not how to “copy his talent,” but how to pull themselves out of repetitive labor and concentrate their effort on direction, judgment, and iteration.
He is not proving that “one person can heroically grind through ten jobs.”
He is proving something much more uncomfortable:
A lot of companies look like they need a lot of people only because, until now, there was no way to compress execution.
Alex Finn: Individual Media Company Production Line
If Pieter Levels represents organizational compression through one person managing multiple products, then Alex Finn represents another path entirely. His Clawbot, “Henry,” can connect itself to Twilio, hook into voice APIs, make phone calls, and keep executing tasks.
What he represents is a different but equally important shift: the content business, for the first time, being compressed into a production line that can be orchestrated by very few people—or even by one.
What actually changed is this:
A content publishing system that once depended on multiple people collaborating has started to be cut apart and redesigned.
When a content chain is restructured like that for the first time, its meaning is no longer just “efficiency improved a little.” It means an ordinary person now has a continuous production capability that used to belong only to a small media team.
Why 2026 Will Be the Year of the Agent
A “year one” is never the year a technology first appears.
The Industrial Revolution did not begin on the day the steam engine was invented.
The internet did not change the world the moment two machines first connected.
The true dividing line is always the moment a technology enters production relations.
2026 deserves to be called the Year of the Agent not because models can write, reason, or call tools for the first time, but because several things became true at the same time for the first time:
1. Models have become strong enough to reliably catch a large number of low- to medium-complexity tasks.
2. Tool use across browsers, files, email, and back-end systems is starting to mature.
3. The dirty work no one used to want to patch—state, logs, recovery, constraints—is finally being taken seriously.
4. And more and more ordinary people are no longer satisfied with “give me an answer”; they are starting to ask, “just get it done.”
Once those four things stack on top of each other, organizational form starts to loosen.
In the past, if one person wanted company-level capability, that person had to actually build a company first.
Now one person can acquire something close to company-level execution capacity first, and only then decide whether a company even needs to be formed.
The Real Shift
The first thing Agents change is not imagination.
It is the cost of execution.
The first thing they rewrite is not what a human being is.
It is what a company is.
So what deserves to be remembered about 2026 is not some model’s benchmark score, and not some demo’s flashy moment.
It is a quieter fact, and a harsher one:
For the first time, an ordinary person can begin to possess the kind of execution capacity that used to belong only to something with company-level scale.
That is the real meaning of the Year of the Agent.
Not every kind of work can be handed over to Agents.
But Agents will become a new means of production—and a new form of asset.
发表回复