What happens when intelligence becomes almost free

Every general-purpose technology has one move: it takes something that was expensive and makes it cheap. Steam did it to mechanical power. The transistor did it to calculation. Bandwidth did it to distance.

The claim worth examining is that AI is doing it to applied reasoning — the work of reading a situation, holding several constraints at once, and producing a competent answer. Not genius. Competence. The thing most professional work actually consists of.

Suppose that continues. What follows?

Cheap inputs don't reduce demand — they relocate it

The intuitive prediction is that if reasoning gets cheap, people who reason for a living lose their income. History suggests something more slippery.

When an input becomes cheap, demand for it usually rises rather than falls, and the value migrates to whatever is now the binding constraint. Cheap computation didn't end the demand for software — it created far more of it, and moved the scarcity to people who could decide what to build. Cheap distance didn't end the value of location — it concentrated it.

Four things that look genuinely scarce

If competent reasoning is abundant, my read is that these do not become abundant with it.

Judgment under responsibility. Someone must decide, and be accountable for having decided. A model can produce the recommendation. It cannot absorb the consequence — legally, reputationally, or emotionally. Accountability is not a capability problem; it's a structural one.

Taste — knowing which problem deserves the effort. When the cost of producing an option approaches zero, the number of options explodes, and the constraint becomes selection. This is already visible: the hard part of using these tools well is almost never generating candidates.

Trust and relationship. A large share of professional work is not the output; it's someone believing the output. That belief accrues to people and institutions over time and does not transfer to a cheaper producer just because the producer is competent.

Access to the physical and the private. Reality is still expensive. So is proprietary context — what actually happened in the meeting, what the customer really meant, what the data doesn't record.

The honest counterargument

I don't think that counterargument wins outright. I do think anyone making career decisions on the strength of "AI can't do judgment" is relying on a claim they haven't actually tested.

The transition matters more than the destination

Most of the discussion is about the equilibrium — what the world looks like when this has fully played out. That is the least useful part to reason about, because it's the furthest away and the least constrained by evidence.

The transition is where the real effects live, and transitions have a characteristic shape: the cost falls unevenly, the benefits concentrate in whoever adapts first, and the people hurt are rarely the people who were warned.

Abundance arrives on a schedule. The dislocation arrives on its own.

What I'd actually do about it

Not a prediction — a posture. If reasoning is getting cheap:

• Get good at specifying problems, because specification is the input that abundance still needs.
• Accumulate the kind of context that isn't in any training set: real customers, real systems, real consequences.
• Treat verification as a skill. In a world of cheap plausible answers, the ability to tell a good one from a confident one is worth more every year.
• Be suspicious of any plan whose main assumption is that the current cost curve stops.

The question I keep returning to

If intelligence becomes abundant, who captures the wealth it creates?

Abundance does not distribute itself. Cheap computation made a handful of firms extraordinarily valuable and made many jobs slightly easier. There's no law that says this one lands differently — and that, not capability, is the part I think we're underweighting.