The people teaching AI how to think are about to be replaced by the AI they trained.
The Summary
- More than 700 workers at Meta contractor Covalen in Ireland face layoffs as AI data annotation becomes automated.
- A former Meta AI exec watched agents outperform her top human workers and launched a nonprofit to help Gen Z prepare for the agent economy.
- The irony: workers who label data to train AI are now redundant because the AI got good enough to label itself.
- Entry-level jobs that used to be stepping stones are vanishing before Gen Z can step on them.
The Signal
Over 700 people working for Covalen, a Meta contractor in Ireland, are staring down layoffs according to internal documents. These workers do data annotation, the tedious work of labeling images, text, and video so AI models can learn what a dog is versus a cat, what hate speech looks like versus sarcasm. It's the foundation of every AI model that feels smart today. And now that foundation is crumbling because the models don't need humans to do it anymore.
The timing tells you everything. Meta didn't gut this team because business is slow. They did it because the AI they were training got good enough to train itself. This is the flywheel nobody talks about in the AI hype cycle: the moment when your AI workforce replaces the human workforce that created it. Not in theory. In Ireland. In 2026.
"The people teaching AI how to think are the first to be replaced by it."
Meanwhile, Clara Shih, who ran AI product teams at Meta and Salesforce, watched this coming. She saw AI agents outperform her best human workers and realized Gen Z is walking into a job market where the entry-level rungs have been sawed off. So she left Big Tech and started a nonprofit to teach young workers how to work alongside agents instead of competing with them.
Here's what Shih understands that most people don't: this isn't about AI taking all the jobs. It's about AI taking the jobs that used to be how you got skills to do other jobs. Data annotation was never glamorous, but it was stable contract work that paid. It was a way in. Now it's a way out.
Key shifts happening now:
- AI model training is becoming self-supervised, reducing need for human labelers
- Entry-level roles in tech are shrinking faster than senior roles
- The skill gap isn't technical anymore, it's knowing how to direct AI agents
The Covalen workers aren't getting laid off because they're bad at their jobs. They're getting laid off because they were too good at teaching the AI to do their jobs. That's the undignified part. You spend months making an algorithm smarter, and your reward is a severance package.
Shih's nonprofit won't save those 700 jobs. But it's trying to stop the next wave of young workers from ending up in the same trap. The pitch: learn to use agents as leverage, not as competition. Learn to be the person who directs 10 AI workers instead of being the one AI worker who gets replaced.
The Implication
If you're in your twenties and looking for your first real job, the old playbook is dead. You won't grind your way up from data entry or junior analyst roles because those roles are being automated first. The new playbook is learning to manage AI agents before you have a title that says "manager." That means prompt engineering, workflow design, and understanding where humans still add irreplaceable judgment.
For companies, this is a warning shot. Every time you automate an entry-level role, you're cutting the pipeline that feeds your senior roles five years from now. If nobody learns the basics because the basics are done by bots, where do your future experts come from? Shih's bet is that Gen Z figures this out faster than their employers do. Watch for a wave of young workers who never held a "traditional" job but know how to orchestrate AI systems better than people twice their age.