The Flat Curve Society
Yegge's argument is a reversal of the usual AI-hype curve. Everyone else is drawing a line that shoots to the moon; he says that for you, personally, the line is going to look flat — and then he explains why that's actually good news.
The thesis
The exponential doesn't stop. What changes is your ability to perceive it. Yegge argues the frontier models will pull away and get locked down — leaving most people on a stable tier of very good, non-frontier models — while two cognitive limits cap what any given person can even notice about further improvement.
"In reality, behind the scenes, the curve is NOT flattening at all; the exponential growth will continue, and you will be able to see outwardly observable signs of it, e.g. in data center growth. But the curve will appear to flatten out for you, through two separate phenomena."— The Flat Curve Society
The two horizons
This is the essay's most durable idea — and it holds even if you have full frontier access, which is why it matters more than the model-restriction prediction it's bundled with.
The Demand Horizon caps how much a better model can help you. If the hardest thing you ask is easy, a smarter model has no room to show its edge.
"It's set by the hardest problem you bring. If all you have handy are easy problems, they don't give a smarter model any room to pull ahead — the outputs look the same because the problem never stretched either one."— The Flat Curve Society, on the Demand Horizon
The Discernment Horizon caps how much of the model's help you can trust. Past a point, the work is beyond your ability to check.
"Past this scary line, you can't tell whether the model is right, because checking the work is itself beyond you."— The Flat Curve Society, on the Discernment Horizon
He compresses the whole idea into three words:
"Superhuman means unverifiable."— The Flat Curve Society
What he tells you to do about it
A plateau, in Yegge's framing, is not a disappointment — it's stable ground. You can't build a craft on a cliff face. The move is to raise your own game: bring harder problems, sharpen the judgment that lets you verify, and get literate with the tools.
- Run an AI literacy audit and train everyone up at least into the single-agent cohort — teams of 5–10, using real work, on company time.
- Treat token efficiency as an advanced skill, not a billing detail.
- Expect engineering to remain essential — models are approximate helpers that need supervision, not replacements for judgment.
"At the highest levels, AI literacy turns into the art of achieving great outcomes with the least spend."— The Flat Curve Society
The lines worth keeping
"A plateau lets us set up a camp and start building."— The Flat Curve Society
"The only thing you get for free is AI Anxiety."— The Flat Curve Society
The one honest objection
The essay's floor is a prediction — that governments and corporations will lock frontier models away "like nuclear weapons." He asserts it more than he argues it, and it's the shakiest joint: model weights are copyable and there's enormous commercial pressure to release them, not restrict them. Notice, though, that the two horizons flatten your curve regardless of access — which is why they're the part of this essay most worth carrying.