Every economy in history has priced labour, land, capital, attention, and time. None of them ever priced a thought.
The Thing Never Priced
Not because thinking was worthless - because it had no unit. You could pay a person for an hour. You could not pay for one idea, or half a consideration, or the extra three seconds of doubt someone spent before signing. Thought was bundled inside a human and sold by the hour, and the internal allocation was invisible. Nobody, anywhere, ever asked "how much did that conclusion cost?" because the question had no denominator.
It has one now. It's a token. And it has a public, quoted, per-unit price that varies by roughly two orders of magnitude depending on how hard you'd like the thinking to be.
This is a bigger deal than the capability. Capabilities improve continuously and we adapt continuously. Pricing a previously unpriced thing is a discontinuity - it happens once, and it permanently changes what gets done. When land got titled, feudalism ended. When risk got priced, insurance and the modern corporation became possible. When attention got priced, we got the entire surveillance-advertising complex and, arguably, a mental health crisis.
Thought just got priced. We are maybe three years in.
What Always Happens to a Newly Priced Thing
Four things, in this order, every time.
It gets budgeted. Something with a price gets a line item, an owner, and a variance report.
It gets arbitraged. If the same output is available at $0.30 and $18.00, someone builds a routing layer, and the routing layer becomes the actual product.
It gets rationed. Budgets bind. When they bind, allocation becomes political, and political allocation follows status.
It gets hedged and speculated on. Anything with a volatile price and a business dependency ends up with forwards, prepaid capacity, and eventually derivatives.
Now apply each one to reasoning and watch how strange it gets.
Cognitive Budgeting, and the First CFO Question
Sometime soon a CFO looks at an inference bill, traces it to a decision, and asks: "why did that call cost us $4.10 of thinking?"
That question has never been asked in the history of commerce. The moment it is asked out loud in a company, the culture of that company changes permanently and irreversibly - because from then on, thinking harder is a thing you have to justify.
Note what makes this so dangerous. The cost of thought is precisely measurable. The value of thought is not measurable at all. Every management system in existence, when handed one precise number and one unmeasurable one, optimises the precise one. This is not cynicism, it's just what the machinery does.
So the default trajectory of every organisation with an inference bill is cognitive austerity: a slow, invisible, well-intentioned reduction in how hard anything is considered, driven by the only variable anyone can see.
The Arbitrage, and the Failure Mode Nobody Is Logging
Because price varies ~100x across model tiers and effort settings, every task now carries an implicit question: how much thought does this deserve?
That question was unanswerable for all of human history. Now it's a config value, and it's usually set by whoever built the pipeline, months ago, with no idea what would flow through it.
Here's the part I think is genuinely underexplored. Everyone is watching for hallucination. Hallucination is loud, embarrassing, and detectable. The real failure mode of a priced-cognition economy is silent.
"A decision that mattered got routed to the cheap tier, produced a fluent and plausible answer, and nobody ever found out it had been under-thought."
There is no error message for insufficient deliberation. The cheap model does not say "I would have caught this with more compute." It produces something confident and correct-looking, at 3% of the price, and the organisation logs a cost saving. The saving is real and immediate. The cost is diffuse and arrives eighteen months later as a strategy that was wrong from the start.
We built an economy-wide dial for how carefully to think, connected it to the accounting system, and pointed no instrument at the other end of it.
Rationing, and the Inequality Nobody Is Measuring
The optimistic story is that inference costs fall to zero and depth becomes universal. I think this is half wrong, and the wrong half matters enormously.
Costs are collapsing - but not uniformly. They collapse fastest for easy thought, because easy thought consumes few tokens and gets cheaper on every axis at once. Hard thought stays expensive almost by definition: hard problems consume more reasoning tokens, need the frontier tier, and require multiple passes. The gap between the cost of a shallow answer and a deep one is not narrowing. On current trends it is widening.
Which relocates inequality to a place we have no vocabulary for.
The twentieth century solved access to information. Libraries, then the internet, then search. That fight is over and information won. But nobody has ever built a public institution for access to deliberation. There is no library of thinking-things-through. And we are now in a world where a wealthy institution can afford to consider a question from forty angles at $2,000 of compute, and everyone else gets the $0.02 version - and both answers arrive in the same font, at the same speed, with the same confident tone.
Information inequality was visible, because you could see who had the books. Depth inequality is invisible, because shallow and deep output look identical on the page. That is the whole problem.
When Reasoning Quality Meets the Treasury Desk
If your product's quality is a function of how much inference you buy, and inference has a volatile price, then eventually someone prepays capacity, someone else writes a forward, and the quality of your company's thinking becomes partly a treasury function.
That sounds absurd for about five seconds, then you remember airlines hedge fuel and their entire competitive position swings on it. The logical endpoint is a firm whose products get measurably dumber in a quarter when its compute hedge rolls off badly. There will be an earnings call where an analyst asks about the impact of inference costs on product quality, and the CFO will have to answer it honestly, and that will be a genuinely new moment in business history.
The Instrument We're Missing
If the failure mode is invisible under-thinking, the fix is not "spend more." It's reasoning provenance - attaching to every consequential output a record of how it was produced: which model, at what effort setting; how many passes and how much verification; what it cost; and the one field that changes everything - what it would cost to be wrong.
That last one is the whole ballgame. Almost every organisation on earth is currently spending two cents of thought on two-million-dollar decisions, and there is no system anywhere that flags the mismatch. The ratio of decision cost to error cost is now computable for the first time ever. It is, I'd argue, the single most valuable unbuilt metric in business.
The personal version is simpler and more uncomfortable: you now have a thinking budget, and you are spending it unconsciously. Every time you take the fast answer on something that deserved the slow one, you made a purchasing decision. You just didn't see the price tag.
Predictions, So This Can Be Wrong
By 2027, "inference cost per employee" appears as a standard internal dashboard metric, and shortly after, in performance conversations.
By 2028, a public post-mortem from a serious company names the root cause as a cheap-tier routing decision on a case that needed the frontier tier. Not a hallucination - an under-provisioning.
By 2029, at least one company discloses compute-cost hedging as material to product quality in a public filing.
Ongoing, the price spread between the cheapest usable reasoning and the best available reasoning widens rather than narrows. If it narrows, my inequality argument is dead and I'll take the loss.
The One Line
We didn't automate thinking. We metered it. And the first thing every civilisation does with a newly metered resource is ration it - long before it works out which uses were the ones worth paying for.
Part two of three. Part one asked what humans are actually paid for. Part three asks where the next generation of experts is supposed to come from.
