The Age of Abundant Cognition

Cognition is becoming abundant. Agency isn’t.

You are not crazy. You are early.

If you’ve felt the ground moving these last few years, a hum under every headline about AI, that feeling is reading something real. Here is what it is.

For all of history, thinking was expensive.

Every hard thing you ever wanted done needed a mind to do it. An expert who took a decade to train. Labor, by the hour. Time, which nobody gets back. An education that cost a fortune and years of your life. Organizations, to get enough minds pointed at one problem.

A trained mind was scarce, slow to produce and costly to rent. So much was priced on that scarcity: careers, credentials, wages, every fee anyone ever paid for advice.

Then the price fell through the floor.

In late 2021, writing at the level of GPT-3 cost about sixty dollars per million tokens. Three years later the same level cost six cents. A thousand times cheaper in three years. The people who track this find the price of any fixed level of machine thinking falling somewhere between nine and nine hundred times a year, depending on the task.

Every other shift you’ve lived through was a change in degree: faster computers, cheaper calls, a new platform every few years. This one is a change in kind. When the price of the thing so much was built on falls toward zero, you don’t get an adjustment. You get a discontinuity.

The mistake is reading it as a trend you can plan around. You aren’t standing on a slope. You’re standing at a wall.

What breaks first.

A job was never just an income. It held up four things at once: the shape of your days, the answer to “so what do you do?”, the feeling of being needed, and the path from who you were to who you could become.

The machines reach the bottom rung first, because it’s the most cognitive and the least embodied: the entry-level role, where you used to become somebody by being paid to learn. The catastrophe isn’t a number going up. It’s a severing of the path.

What stays scarce.

When one thing gets cheap, value doesn’t vanish. It moves to whatever the cheap thing can’t supply.

Cheap cognition can draft, code, research, summarize, plan and explain. It can’t want anything for you. It can’t give you permission. It can’t be in the room. It can’t carry what you learned into next year, or tell you whether any of it actually happened.

So the bottleneck moves. To meaning: knowing what matters. To judgment: choosing a future worth wanting. To authority: being allowed, and trusted, to act. To coordination: getting people and agents to move together. To embodiment: bodies, presence, the physical world. To context: holding what you know across time. To verification: knowing when something actually happened. And to the one that holds the rest together, agency: owning the whole loop, from what you want to what actually changed.

That’s the whole diagnosis in one line. Cognition is becoming abundant. Agency isn’t.

This isn’t hype, and it isn’t doom.

Hype says the machines will do everything, so relax. Doom says the machines will do everything, so despair. Both make the same mistake: they count the part that got cheap and forget the part that didn’t. The machines will do more thinking than anyone can picture. They can’t want, own, answer for, show up, or live with the result. Those were always the point. We paid for thinking because it was the bottleneck.

Why now: the concrete is wet.

A few times in history the medium a civilization is built from goes soft. The printing press did it to information. The industrial firm did it to work. For a short window the material is soft enough to press, and then it sets, and the shapes pressed in while it was soft are the shapes the whole age inherits.

We’re in one of those windows, and the medium going soft is intelligence itself. Being early isn’t a little better. It’s a different category: a shape pressed while the concrete is wet rides the whole hardening. A shape attempted after it sets has to be chiseled against the grain.

The ground moving is also the ground opening.

The deal that’s breaking was never that good. Trade your one life for a wage, become your function, climb a ladder someone else built, hope it holds. What replaces it, for the people who cross in time, is stranger and freer: stop pricing yourself on the part that got cheap, and move onto the part of you no machine can touch.

Alexander had Aristotle. Now everyone does. The question is no longer whether you can get a great mind to think with you. It’s what you’ll cause with it.

So here’s the move for tomorrow morning. Write down everything you spent half an hour on today. Mark each one cheap or scarce. Hand one cheap thing to a machine, and give that hour to the scarce thing you’ve been starving.

You felt it before anyone said a word. That feeling isn’t your weakness. It’s your instrument, and it’s reading the terrain correctly.

---

Sources: a16z, Welcome to LLMflation · Epoch AI, LLM inference prices have fallen rapidly but unequally across tasks