Stanford’s 2026 AI Index dropped today, and one number stands out above everything else: a 50 percentage point gap between how AI experts and the general public view the technology’s impact on jobs. According to MIT Tech Review, 73% of U.S. experts feel positive about AI’s effect on employment, while only 23% of the public shares that optimism.
That’s not a disagreement. That’s two groups looking at entirely different things.
What’s Driving the Divide
MIT Tech Review’s Will Douglas Heaven points to a compelling theory: the gap tracks closely with how people actually use AI. A software developer posted on X that “the degree to which you are awed by AI is perfectly correlated with how much you use AI to code.” It sounds like a joke, but it captures something real.
Here’s why. The latest models from top labs are genuinely excellent at coding. Technical tasks have clear right-or-wrong answers, which makes them easier to train for. Models that code well are also proving profitable, so labs keep pouring resources into making them better. Experts, who skew technical, see this progress daily and extrapolate forward.
The general public? They’re more likely to encounter AI through chatbots that hallucinate, customer service tools that loop endlessly, or headlines about layoffs. Same technology, radically different experience.
The Inconsistency Problem
The report paints a picture of an industry that genuinely defies simple narratives. As MIT Tech Review highlights, Google DeepMind’s Gemini Deep Think scored a gold medal in the International Math Olympiad but fails to read analog clocks half the time. That single fact captures the weird moment we’re in: superhuman at math, stumped by a wall clock.
The Stanford authors frame it well. AI is simultaneously a gold rush and possibly a bubble. It’s taking jobs and creating them. It’s brilliant and brittle. These aren’t contradictions that will resolve neatly. They’re features of a technology that’s genuinely uneven in its capabilities.
The Hardware Reality Check
Beyond the perception gap, the report surfaces a structural risk that deserves more attention. The U.S. hosts 5,427 data centers, more than 10 times any other country. But here’s the vulnerability: “A single company, TSMC, fabricates almost every leading AI chip, making the global AI hardware supply chain dependent on one foundry in Taiwan.”
One foundry. Every leading AI chip. That’s a concentration risk that should keep strategists up at night, regardless of where they fall on the optimism spectrum.
What This Means for Practitioners
A few practical takeaways from the gap:
- If you’re building AI products, remember your users aren’t AI researchers. The public’s skepticism isn’t ignorance. It reflects real experiences with mediocre AI deployments. Meet people where they are, not where your benchmarks say they should be.
- If you’re making business decisions about AI adoption, don’t rely solely on expert enthusiasm or public fear. Both are skewed by their respective exposure. Run your own pilots with your actual workflows.
- If you’re investing in AI infrastructure, the TSMC dependency is a material risk factor that most market analyses underweight.
The 50-point gap isn’t going to close by telling the public they’re wrong. It’ll close when AI products consistently deliver value outside of coding and technical tasks. Until then, we’re living in two AI realities at once.
The full Stanford AI Index report and MIT Tech Review’s analysis offer much more detail for those tracking these trends closely.