agency is the last scarce resource

for all of human history, if you wanted more thinking done, you needed more humans. thinking was the one input you could never stockpile. it lived in heads, it took decades to grow, and it went home at six.

that constraint just died. quietly, in about three years.

the collapse

the price of a fixed level of intelligence is falling faster than any input cost in economic history. epoch ai tracked it across benchmarks: the cost of hitting a given capability level halves roughly every few months. answers that cost $20 per million tokens in late 2022 cost seven cents two years later. a 280x collapse. and the frontier itself got cheaper in absolute terms: the best model ever built costs less to run today than last year's second best.

make an input 100x cheaper and you get a different economy.

every era is defined by its scarcest input. agrarian wealth was land. industrial wealth was capital. the last seventy years ran on knowledge: the scarce thing was people who knew how, and we built our schools, firms, and status games around rationing them.

knowledge just became a utility. so the only question that matters: what's still scarce?

my answer: agency.

what i mean by agency

i mean something narrower and rarer than hustle: the will to decide. to want something specific, commit to it, move without being told, and keep moving through fog with nobody confirming you're right.

you can watch the constraint shifting in real time. the mit study everyone quotes found that roughly 95% of corporate ai pilots produced no measurable return. the same study found employees quietly automating their own work with personal ai accounts, and that worked. same company. same models. the pilots had budgets, committees, and steering groups. the employees had agency.

the difference was permission. agency is the thing that acts before permission arrives.

i see it in builders too. the ones winning right now aren't smarter than the ones stuck. they start uglier and earlier, ship into silence, and treat every closed door as routing information. the stuck ones are often more talented. they're waiting for something: a signal, a cofounder, a permission slip that isn't coming.

but wasn't it always?

the obvious pushback: agency was always the scarce thing. every generation's self-help shelf says so. nothing new.

here's what's new. agency was always rare, but it was never the binding constraint. an agentic person without capital, a team, or credentials still hit a wall: execution. you could want and decide all day; you couldn't build the thing alone. the wall was real, so the economy never actually priced agency. it priced the things that got you through the wall: degrees, capital, headcount.

the wall is gone. one person who decides can now command what used to take a department. for the first time, the gap between people is activation energy.

when every other constraint falls, the remaining one becomes the whole price.

the reprice

follow the logic and a lot of familiar things reprice:

• credentials: a degree is proof of stored knowledge, and stored knowledge is what just became free. proof of initiative has no institution yet. someone will build one.
• hiring: every interview process we have screens for knowing. almost none screen for moving. the miss rate on the thing that now matters most is enormous.
• education: we built schools to transfer knowledge and, honestly, to suppress agency: sit still, wait your turn, ask permission. that design is now exactly backwards, and nobody is in charge of fixing it.
• management: when execution was scarce, managers allocated it. when execution is abundant, the job flips to deciding what's worth executing. most orgs haven't noticed the flip, which is why they're drowning in ai output nobody asked for.

i might be missing something on how fast institutions can adapt to this. my honest guess is: slower than anyone inside them believes, and it won't be their choice in the end.

is agency trainable?

the uncomfortable open question. if agency is a fixed trait, the ai era hands a permanent aristocracy to the people born with it. if it's a habit, then the most valuable education on earth is whatever builds it.

i lean habit. i've watched quiet people become dangerous the first time something they shipped got used by a stranger. agency compounds off evidence that your actions change things, and for most of history that first evidence was expensive: you needed resources, permission, and luck just to find out that acting works. that first taste now costs a weekend and zero dollars.

so here's the most optimistic sentence i can write about this decade: we may be about to mint more agentic people than every previous era combined, simply because the cost of discovering your own agency collapsed.

but i can't prove it. the signal to watch: whether the number of first-time builders keeps climbing now that the tools are nearly free, or whether the same small crowd just ships faster. that number decides which world we're in.

the last scarce resource

intelligence used to be the engine and agency the spark. we spent seventy years treating the engine as the valuable part, because engines were expensive.

the engines are free now.