The Uncomfortable Truth About AI and the American Worker
Workers hate AI. They see it as a threat: a silent layoff notice, a skill-eraser, a paycheck terminator. That fear isn't irrational. But new data from Morgan Stanley reveals a twist no one's talking about:
AI is turbocharging productivity. And workers don't even know they're the engine.
The Numbers That Should Change Everything
Industries most exposed to AI drove **1.7 of the 2.4 percentage points** in U.S. productivity growth over the last four quarters of 2025. A year earlier? Just 0.7 points. That's not growth. That's acceleration.
Here's the kicker: **headcounts didn't drop.** Employment trends were flat across high-, medium-, and low-AI sectors. The difference? Output. High-AI industries produced *more* with the *same* people. Low-AI industries? Their output slowed.
Economically, this is the "best-case" scenario: augmentation, not replacement. Psychologically? It's a trap. Workers are more valuable than ever—and feel more disposable than ever. That disconnect isn't a literacy problem. It's a warning.
The 90% Problem
Aggregate data hides the real story: **AI isn't replacing workers. It's letting the top 10% replace the other 90%.**
As AI strategist Daniel Miessler puts it: *"AI can't replace top performers. But companies would rather fire the bottom 75% and let the best wield AI to become 10x—or 100x—more productive."*
The productivity boom isn't lifting all boats. It's concentrating power. And even the top 10% aren't safe—they might just be buying time before the tools they rely on learn to work without them.
You're Not the Customer. You're the Training Data.
Here's the subsidy nobody talks about:
- Real cost of a serious AI user: **$80–$150/month** in compute
- What you pay: **$20/month**
- Who covers the gap? **You do—with your data.**
Every prompt, edit, and follow-up you feed a frontier model trains the next version. As technologist Shaun Warman notes: *"Synthetic data has crossed the quality threshold."* Models can now generate, filter, and grade their own training data. The marginal value of your human touch? Falling fast.
When the Free Ride Ends
Warman sees three forces slamming the "apprenticeship window" shut in 3–5 years:
1. **Quality**: AI-generated training data now rivals human input
2. **Autonomy**: Models can self-evaluate and improve in verifiable tasks
3. **Scale**: More human feedback yields diminishing returns
When they converge, the subsidy vanishes. Expect:
- The $20 tier to disappear or become ad-supported junk
- Top capabilities locked behind five-figure enterprise contracts
- AI labs stopping tool licensing altogether—and becoming the *operators*: law firms, consultancies, hedge funds capturing the value themselves
The productivity gains we see today? They're a snapshot of a temporary truce. Workers and labs both want better output—so both win, cheaply, for now.
The Repricing No One's Pricing In
Morgan Stanley's economists hedge carefully: rising output and stable employment *"may change as AI adoption picks up."* That understatement is the story.
When repricing hits, it won't be equal:
- **Large enterprises** will amortize six-figure contracts across thousands of employees. Barely a blip.
- **Smaller firms, public sector workers, teachers, healthcare aides**? They'll face a brutal choice: pay enterprise rates they can't afford, or lose the tools that made them productive.
The gains are broad today. The access won't be tomorrow.
Workers Aren't Wrong. They're Early.
The fear isn't about the job loss being measured *now*. It's about the bill that hasn't arrived *yet*. Workers sense the threat isn't displacement by machines—it's **repricing by markets**.
AI isn't the villain in this story. The business model is.
And when the subsidy ends, productivity won't save you. Only leverage will.
Are you building yours?