wanting is the bottleneck

the most expensive thing in my work right now is not building. building is close to free. the most expensive thing is knowing, precisely, what i want.

that sentence would have been nonsense for all of economic history. i think it's about to be the defining constraint of the whole economy.

the inversion

an economy is wants meeting capacity. for ten thousand years, capacity was the short side: not enough labor, not enough energy, not enough intelligence. every economic institution we have, firms, schools, markets, careers, is a machine for rationing scarce capacity toward unlimited wants.

capacity is now going vertical. the cost of a unit of cognition halves every few months. agents turn one person's intent into a team's output. whatever you can specify, you can increasingly have.

read that again, because the load-bearing word is not "have." it's "specify."

wants don't scale like transistors. humans want vaguely, slowly, and socially. "i want my business to grow" is not a want an agent can execute. "i want this exact audience to feel this exact thing, and here's the trade-off i'll accept" is. the distance between those two sentences used to be hidden inside months of expensive execution. now execution is instant, and the distance stands exposed as the actual work.

when capacity is abundant, the economy becomes demand-constrained. growth stops being limited by what we can make and starts being limited by what we can precisely want.

you can already see it

every ai product is converging on the same ux discovery: the hard part isn't the model, it's that users don't know what to ask for. the blank prompt box is the blank page with infinitely higher stakes. it can do anything, which is exactly why people freeze.

inside companies, the same shape: everyone has access to the same models, and output varies by orders of magnitude between teams. the variance isn't capability. it's that a few people know what to point the machine at, and most are generating impressive answers to questions nobody sharpened.

we've watched this movie with other abundances. calories were scarce for 200,000 years; the moment they became abundant, the problem inverted, and the valuable skill became choosing what not to eat. information was scarce for centuries; google made it abundant; the bottleneck instantly became attention. every abundance inverts its own bottleneck. cognition will not be the exception.

demand is infinite, though

"demand is infinite. humans always want more." it's the oldest line in economics, and it's true, and it dodges the point.

aggregate wanting is infinite. economic demand is specific: a want sharp enough to act on, priced, chosen over alternatives. vague appetite moves no machines. and the sharpening is genuinely hard. it requires taste (knowing which version is right), judgment (knowing what to trade away), and conviction (holding the want steady through iteration). those are human-speed faculties, trained by exposure and by shipping and by watching things fail. they don't halve in cost every few months.

there's a subtler version of the objection: ai will just tell us what to want. it already does; feeds have been manufacturing desire for a decade. but notice what manufactured wanting produces: consumption, not creation. the feed can make you want to buy. it cannot make you want to build something specific enough that agents can execute it. the second kind is the economically explosive kind, and it stays stubbornly human.

what this does to the map

• the interface is the empire. the most valuable products of the next decade won't be the most capable. they'll be the ones that help people discover and sharpen what they want: taste as a service, opinionated defaults, curation that says no on your behalf. the blank prompt box will be remembered as a transitional ui, the way the command line was.
• the high-leverage skill flips. we spent a century training people to answer well: exams, interviews, deliverables. the scarce skill now is asking precisely. a prompt is a want, compiled. most people can't compile yet.
• specification becomes a profession. somewhere between product manager, editor, and director: people whose job is turning a fog of intent into an executable want. every era's binding constraint mints a profession. this one is being minted right now.
• companies and countries start competing on articulated demand. capacity is a commodity you can rent by the token. a culture that produces people who want things sharply, with taste and ambition and opinions held for reasons, will outperform a culture that only produces capacity. that's a strange new kind of comparative advantage, and no education system on earth is aiming at it yet.

the human bottleneck, again

i wrote once that building feels incoherent because the market's feedback can't tell you whether you're early, wrong, or irrelevant. this is the same fog, one level up. the machines have removed almost every excuse between you and the thing you want. what's left, exposed, is whether you actually want anything specific.

most people, handed infinite capability, discover they don't. the ones who do are about to aim the entire machine.