For most of the last century, the top 10% of sharpest minds found ways to slip the leash—dodging state-mandated schooling, corporate catechisms, propaganda, and all the other velvet chains. The real exam was always the jailbreak. The bottom end never had many options; we kept them busy—fries at midnight, deliveries at 2 a.m. The vast middle got herded into "careers": state-approved education, commercials, culture, and an obedient lifetime of busywork. Fake office roles for compliance-trained desk jockeys. Commutes, debt, loyalty rituals, meetings about meetings. Plato's cave with Slack notifications.
AI torches that script. When a frontier LLM can match or beat a trained mid-tier specialist—faster, cheaper, sober on Mondays—you don't need the unreliable, slow human in the swivel chair. We don't train abacus-pushers anymore because we carry supercomputers in our pockets; same deal here. The "middle class" loses bargaining power the moment their marginal value falls below a prompt.
In practice, the "stupid" and the "midwit" collapse into one operational category: replaceable. Office politics, loyalty, and polished compliance don't clear the bar if an average model outperforms you on average days. No one pays for latency wrapped in attitude.
Meanwhile, the top 10%—the ones who never fully swallowed the programming—strap on AI like a jetpack. Leverage compounds. The gap between the top 10% and the bottom 90% becomes not just wide, but load-bearing and effectively unbridgeable.
That's where stability gets tricky. Managing the majority's incentives and attention becomes a first-order function. When the midwit herd and the dunce cavalry surge together, entropy wins. Underestimate the tidal force of mediocrity and you don't get a do-over; civilizations face-plant that way.
So the new AI-augmented elite inherits a duty: design systems that channel, train, and entertain the now-homogeneous mid-band—keep their attention aligned and their incentives sane. Population policy is above my pay grade; attention policy isn't. Ignore either and you'll be running disaster drills with a chatbot.
Are you in the top 10% or the bottom 90%?
Did you use LLMs today to build something new—or did you just follow orders, reciting an industrial-age obedience script in 2025?
Your answer lives between those two questions.