So here's something that's been bothering me about the whole AI layoff trend. Companies are laying off thousands of people claiming AI will replace them, then quietly rehiring those same people weeks later. It's not a glitch in the system—it's revealing something much deeper about how AI actually works in the real world.

Let me walk you through what happened. Jack Dorsey's Block announced in late February that they were cutting over 4,000 employees—dropping from 10,000 down to less than 6,000. The official story: AI changes everything, so we need fewer people. Sounds clean, right? Except by mid-March, those laid-off workers started getting calls back. Engineers, recruiters, designers—all returning. Some were told it was a clerical error. Others had managers fighting to get them rehired. A few just got a call a week after being let go asking them to come back with zero explanation.

This isn't new either. Remember Klarna? The Swedish payment company laid off over 1,000 people in 2024, claiming their AI customer service could handle the workload of 700 agents. By May 2025, Bloomberg and other outlets were reporting that Klarna was rehiring customer service staff. Their CEO basically admitted they moved too fast. Too fast with what, exactly?

Here's where the math gets interesting. Enterprise-level AI isn't cheap. Claude Opus costs 5 dollars per million input tokens and 25 per million output tokens. Domestic models are cheaper—Qwen 3.5 Plus runs 0.8 yuan per million input tokens—but we're still talking real money. Someone I know who uses these tools for research burned through 6,000 dollars in tokens over a month. Think about that. What kind of senior professional can you hire for 6,000 dollars a month outside of expensive Western markets? Now scale that up for a company trying to replace customer service, engineering, or recruitment with AI. Training an AI system that actually handles complex tickets, accesses multiple knowledge bases, manages multi-turn conversations, and stays stable? That's not a 3,000 yuan monthly hire situation. That's infrastructure costs that add up fast.

But there's something else happening here that people miss. It's called the Jevons Paradox. Basically, when efficiency improves, we don't use less of a resource—we use more of it because it's cheaper. In the workplace, this means as AI makes employees more efficient, companies don't let them rest. They just pile on more work. Efficiency becomes a hidden tax on the remaining staff. The narrative about AI liberating human labor? That's marketing fiction.

What I think is actually going on is this: companies aren't smart enough to integrate AI into real workflows without breaking things. They're using AI as cover for cost-cutting. Lay off people, claim it's progress, then rehire when you realize half the work didn't get done. The remaining employees? They're drowning in extra tasks, fewer collaborators, and way more stress. And this isn't just about workload. Companies are organizations made of people and relationships—informal networks, trust, institutional knowledge. You can't replace that with tokens. When you lay off people, you're cutting organizational muscle, not just labor.

Jensen Huang actually called this out at NVIDIA's GTC event in 2026. He criticized leaders who use AI as an excuse for layoffs, saying they're just out of ideas. Real leaders should be using AI to expand and hire more people, not shrink teams. But let's be honest—most tech executives understand the actual cost structure of AI. They know it's not a magic replacement for human labor. So why the layoffs?

Because the real story isn't about AI efficiency. It's about cost reduction. AI became the universal excuse for trimming headcount. When a company's growth stalls and profits shrink, suddenly AI becomes the reason to PUA your employees—cut people, increase workload, make everyone feel like they're not innovative enough for the new era. Then if you accidentally laid off someone critical, you quietly rehire them.

Musk did something similar at Twitter. After the October 2022 acquisition, he laid off roughly half the staff—over 3,000 people—in November. Then he realized he'd cut too deep, so he brought back dozens of people who turned out to be essential. Same pattern.

Look, AI will change things. That's real. But it's not magic. It can't fix a company's strategic problems or replace good management. What we're seeing now is companies using AI as a cover story for the oldest trick in the book: cutting costs and hoping the remaining people can somehow do it all. The rehiring that follows just proves the point—some jobs never actually disappeared. They were just convenient casualties in a cost-cutting exercise that needed a good narrative.

The people who got laid off and rehired? They're not seeing a reversal or a win. They're seeing proof that they were hurt by something that didn't even need to happen. That's the real story here.