The Hiring Paradox: AI Labs Can't Stop Hiring Humans
The companies most primed for AI automation can’t stop hiring humans
If AI is about to eliminate millions of jobs, you’d see the evidence at AI labs first. They have the best models, the deepest expertise, and the strongest financial incentive to replace human labor with algorithms. They should be the canary.
Instead, they’re hiring like it’s 1999.
Anthropic went from 240 employees to over 1,000 in eighteen months. OpenAI went from 770 to 3,500. Google DeepMind absorbed the Brain team and now employs 5,600. Meta laid off 600 workers in late 2025 and simultaneously offered $100M signing bonuses to top AI researchers.
These aren’t companies desperately automating their way to profitability. They’re companies throwing money at humans as fast as they can find qualified ones.
The high-profile departures (Ilya Sutskever, Andrej Karpathy, Mira Murati) weren’t layoffs. They were poaches or new labs. The story isn’t “AI labs replace humans with AI.” The story is “AI labs can’t hire humans fast enough to keep up with demand for AI.”
If automation were going to collapse employment, it would start at the companies building the automation. They have every structural advantage. They use Claude Sonnet 4.5, GPT-4o, o3, and Gemini 2.0 before anyone else. They understand the capabilities better than any other organization on Earth. They have the capital. And they keep hiring.
The Pattern Repeats
This isn’t new. When ATMs launched in the 1980s, everyone predicted bank tellers would disappear. In 1985 the US had 60,000 ATMs and 485,000 tellers. By 2002: 352,000 ATMs and 527,000 tellers. More ATMs, more tellers.
Banks opened more branches because ATMs made banking cheaper. New branches needed tellers for the complex transactions ATMs couldn’t process.
McKinsey studied this across the labor market from 1980 to 2015. Computers eliminated about 3.5 million jobs and created 19 million. Net: +15.7 million. The jobs changed (fewer typists, more software engineers), but total employment went up.
Amazon runs over a million warehouse robots. It also employs 1.56 million humans. The robots didn’t replace the humans. They made the operation large enough to require more of them. Automation doesn’t delete jobs. It restructures them.
The OECD looked at task-level automation across developed economies. Only 9% of jobs are fully automatable when you decompose them into tasks. Most jobs are bundles. AI handles the automatable parts. Humans focus on the parts that need judgment, creativity, or social interaction. The job changes; it doesn’t vanish.
The World Economic Forum projects a net 2% job loss through 2027 from AI and automation. Not 20%. Not 50%. Two percent. That’s a reshuffle, not a collapse.
The Bureau of Labor Statistics projects 5.2 million new US jobs over the next decade. Computer occupations grow at 11.7%, three times faster than overall employment.
Why AI Labs Keep Hiring
AI labs hire because building and improving AI systems takes humans at every level. Researchers push the frontier. Engineers build infrastructure. Safety teams red-team the models. Product managers figure out what users want. Sales explains the capabilities to enterprises. Support helps customers implement. Operations manages the chaos.
The better the models get, the more humans you need. Not fewer.
Better models unlock new use cases. New use cases mean new products, new teams, new customers, more support load, more integration work. Claude Sonnet 4.5 didn’t reduce Anthropic’s headcount. It justified hiring another few hundred people to handle the demand it created.
The automation apocalypse narrative assumes a zero-sum game. Every task AI handles is a job lost. But markets aren’t zero-sum. When AI makes something cheaper or faster, demand goes up. When demand goes up, adjacent jobs emerge. McKinsey found exactly this pattern in the computer revolution.
The question isn’t “Will AI replace jobs?” It’s “Will the new jobs emerge fast enough to replace the old ones?” History suggests yes.
I’m not saying no one will lose their job. Specific roles in specific industries will disappear, the way elevator operators disappeared when automatic elevators arrived. The macro story isn’t “AI eliminates employment.” It’s “AI restructures employment, and the companies building AI are hiring aggressively because they understand that better than anyone.”
Until Anthropic, OpenAI, DeepMind, and Meta start shrinking their workforces instead of fighting over the same talent pool, the automation apocalypse narrative is premature. They’re the canary. The canary is hiring.
Common Questions
“This time is different. We’re automating cognitive work, not physical work.” People said the same thing about computers in the 1980s. Cognitive work restructures differently than physical work, but it still restructures rather than vanishing. The OECD task-level data shows this.
“Specific industries will still get decimated.” Yes. Travel agents by Expedia. Typists by word processors. But decimation of one industry isn’t collapse of total employment. It’s transition. Some jobs end. New jobs emerge. The question is speed of adjustment, not direction.
The companies building the automation can’t stop hiring. That’s the signal.