Stanford dropped their 400-page AI Index this week and everyone latched onto the same line. GPT-5.4 scored 83% on GDPval. Matches or beats human experts on 44 economically valuable jobs. Good tweet. Cool hoodie.

That's not the story.
While you were watching the capability race, a second race started. Nobody mentioned it. It has no benchmark. No leaderboard. And it's the one that's going to eat most of the people reading this.
The capability numbers are nuts. GPT-5.4 crossed 83% on GDPval, a benchmark OpenAI built because the old ones got boring. 44 occupations graded blind by real professionals. Models sit at or above human expert level on the stuff you'd pay $300/hr for. Scaling laws are holding. 10x compute, 2x intelligence, rinse, repeat.
And the "US pulling ahead" narrative your LinkedIn uncle keeps posting is cope. American models led Chinese models by 9.26% in January 2024. By last month, the lead was 2.7%. They've traded the top spot multiple times since early 2025. Meanwhile AI researchers immigrating to America dropped 89% since 2017. 80% in the last year alone.
The capability edge is compressing fast. Everyone has smart models now. Everyone is going to have smarter ones.
That's why nobody told you about Race #2.
PwC's report hit the same news cycle like a rock through a window. Twenty percent of companies are capturing seventy-four percent of the value from AI. The gap keeps widening. The 20% generate 7.2x more revenue gains than the average. They aren't on a curve that's going to flatten. They are running a different race with a different finish line.
The leaders point AI at growth. The laggards point it at cost reduction. Firing people and calling it a strategy is not a strategy. The companies cutting headcount to prove they're "AI-forward" are the same companies that will underperform hard through 2028.
Here's the stat I had to sit with. Employment for software developers aged 22 to 25 dropped nearly 20% since 2024. Twenty percent of a whole cohort, in under two years. The kids who spent four years learning to code are walking into a market that doesn't need them because the entry-level work they would've done is the exact work the 83% GDPval model is best at. Nobody wrote the obituary because the companies didn't issue a press release titled "We Stopped Hiring Your Graduate Because We Bought a Subscription."
So where does this leave you.
You are not going to benefit from AI by using it. That was true in 2023. It stopped being true about six months ago. Using ChatGPT to write your emails faster is the 2024 equivalent of using Excel to organize your recipe collection in 1998. Cool trick. Doesn't make you Goldman Sachs.
The question in 2026 is whether you're in one of three positions. You build with it. You work at a company in the 20%. You are the customer of the 20%. If you are not in one of those three buckets, the data says you are being squeezed out of the economy. Quietly. Your wages don't move. Your boss starts talking about efficiency.
Do more with less is what you say right before you fire someone.
Three numbers from the Stanford report are the only ones that matter. 83% on GDPval. 20% capturing 74%. Software jobs for 22-25 year olds down 20%. The machines became capable of doing the work, companies concentrated the value faster than anyone planned, and the people who took the longest to train are losing their seats first.
We are past the hype phase. The gap between "I'm in" and "I'm not" opened up fast. Everyone can see the edge. Some people are walking away, some are building bridges, and some are taking one more step forward because nobody told them the ground was moving.
The canyon doesn't care which group you're in. Neither does the model.