Original | Odaily Planet Daily (@OdailyChina)
Author | Golem (@web 3_golem)
The first batch of employees laid off by AI have returned to their posts.
On February 27, Jack Dorsey (founder of Twitter)'s fintech company Block laid off over 4,000 employees in one go, reducing the total headcount from 10,000 to less than 6,000. Jack's reason for the layoffs was that "AI tools have changed everything." It has long been a societal consensus that AI will eventually eliminate some professions, but the fact that it is first replacing white-collar workers in mid-to-high-level jobs has intensified workplace anxiety. (Related reading:Jack Dorsey's Company: 4,000 White-Collar Workers Are Being Replaced by AI)
However, less than a month later, some of the laid-off employees have already received invitations to return...
According to Business Insider, these rehired employees come from various departments, including engineering and recruitment. A design engineer at Block posted on LinkedIn that leadership told him he was laid off by mistake, a "clerical error"; an HR employee (in a since-deleted post) stated that they were rehired after their manager persistently advocated upwards for them; and another person mentioned receiving a call from Block out of the blue a week after being laid off and being asked to come back.
Jack has not publicly responded to the rehirings. Proportionally, these rehired employees represent only a very small fraction of those originally laid off, but it perhaps already indicates the problem: for some positions and tasks, AI is not as effective as humans.
First, from a usage cost perspective, an enterprise-grade AI employee is certainly more expensive than a regular human resource.
Hiring people to work costs money, hiring AI to work costs tokens. The standard base price for Claude Opus 4.6 is $5 per million tokens for input and $25 per million tokens for output; domestic large models are cheaper, with Qwen3.5 plus's standard base price being 0.8 RMB per million tokens for input and 4.8 RMB per million tokens for output.
Taking the recently popular OpenClaw as an example, a senior "shrimp farmer" within Odaily Planet Daily mentioned that using OpenClaw merely as a life and research assistant burned through about $6,000 worth of tokens in just over a month (they used the Claude 4.5/4.6 model). $6,000 a month – what kind of highly educated intellectual couldn't you hire for that (outside of Europe and America)?
If personal use is like this, the cost of integrating AI into enterprise work is even higher. Taking the simple replacement of customer service as an example, in regions with degree inflation, you can hire a good-looking college graduate as a customer service representative for 3,000 RMB. But training an AI customer service agent that can truly replace a human, handle complex tickets, connect to multiple knowledge bases, engage in multi-turn conversations, and remain stably online – that cost is definitely not something 3,000 RMB per month can cover.
In 2024, the Swedish payments company Klarna proudly laid off over 1,000 people, claiming its AI customer service could already handle the workload of 700 customer service agents. But in May 2025, Bloomberg and other media reported that Klarna had started rehiring people for customer service, and its CEO admitted that the company had indeed "moved too fast" with AI.
Furthermore, AI replacing human labor also faces the "Jevons Paradox".
The Jevons Paradox is an economic concept stating that an increase in efficiency does not necessarily lead to a reduction in the use of a resource. Instead, because the cost of use decreases and demand expands, the total usage may actually rise. Applying this theory to the workplace in the AI era means that when AI technological progress improves employee efficiency, companies will not allow employees to rest; instead, they will demand that they complete more tasks within the same unit of time.
So-called efficiency improvement has become another, more hidden form of increased burden. AI liberating human labor is completely a scam.
Capitalists also believe that in the AI era, companies simply won't need as many employees, as Jack said, "smaller teams with more intelligent tools." But in reality? The current situation is that after layoffs, the original work is not entirely inherited by AI; rather, the remaining employees, aided by AI, have taken on increased workloads.
If it were just单纯的工作任务也就罢了, but one must remember that, ultimately, a company is a human organization. Where there is organization, there is a "jianghu" (complex social dynamics). AI can integrate into the formal structure of a company, but it can never understand, nor integrate into, the informal/invisible structure of a company.
Therefore, when AI-driven layoffs occur, they cut not just labor, but organizational muscle. The remaining employees not only shoulder a heavier work burden but also swallow the anxiety, risk, and responsibility that originally belonged to the eliminated positions. There are fewer people to collaborate with, fewer people to execute, and most importantly, fewer people to take the blame.
During Nvidia's GTC 2026, Jensen Huang criticized companies that use AI efficiency gains as a reason for layoffs in an interview: "Those leaders who resort to layoffs in response to AI are simply because they can't think of a better way. They have no new ideas left in their heads. Even with the strongest tools, they won't use them for expansion," were Huang's exact words.
What Jensen Huang meant is that AI is not here to eliminate employees but to help companies expand and develop new businesses. Don't lay people off; instead, increase hiring. If management doesn't realize this, they are fools. But joking aside, managers in companies are often the cream of the crop of shrewd people. They certainly know the current high cost of AI and the continued necessity of human labor.
Layoffs in tech companies – perhaps AI is just a pretext, cost reduction is the real goal.
AI has become a universal excuse for layoffs in tech companies. In truth, what AI is really淘汰 isn't individuals, but those enterprises and businesses still living in the old era. When companies fail to keep up with AI advancements, leading to stagnant business growth and shrinking profits, the AI revolution instead becomes a new means for companies to PUA employees: reduce headcount, pressure costs, cram more work onto those who remain, and then let each person reflect on why they couldn't become someone more adapted to the AI era.
If they unfortunately cut a critical artery, they can just quietly ask them to come back. This method of layoffs is also common in Silicon Valley. In October 2022, after Musk completed the acquisition of Twitter, he laid off about half of the employees (over 3,000 people) in early November. He subsequently rehired dozens of laid-off employees because they were let go by mistake or key positions were found to be indispensable.
Returning to the present, ultimately, AI will change many things, but it is not yet magical enough to help companies compensate for strategic迟钝, business衰老, or managerial偷懒. The matter of being laid off by AI and then rehired, whether the underlying reason is the company realizing that some work doesn't just disappear with a statement like "AI changed everything," or whether it's just an excuse for cost reduction, is neither热血 (inspiring/heroic) nor a reversal.
It just shows us that before the future has truly arrived, some people have already been hurt by it提前.
