On capital for machines.

Capital allocation is the quietest decision in the modern world.

It rarely makes the news. It has no single face. And yet it decides, more than almost any other human act, which drugs get developed, which technologies get built, which companies survive the decade, which cities grow, and which ideas are allowed to become real. Every other economic decision runs downstream of it.

For all its consequence, it still runs on a fragile foundation: the attention of a small number of people, each of whom can only read so much, think so hard, and compare so many possibilities in a given day. The best of them are remarkable. Most of them are not. And even the best were selected by processes that rewarded confidence and social capital at least as much as they rewarded accuracy.

The honest description of modern capital allocation is that it is a bottleneck pretending to be a discipline. A handful of humans, trained mostly on each other, deciding where the world's resources flow. Not because they are the best possible mechanism. Because they are the only mechanism we had.

That is ending.

Software can now read every filing, every earnings call, every patent, every regulatory disclosure, every academic paper in the relevant field, every hiring trend on LinkedIn, every change in supply chain shipping data, every anomaly in a satellite image of a factory parking lot — and hold them all in memory at once, cross-referenced, updated continuously, and available for reasoning in seconds. No human ever could. No human ever will.

This is not a story about replacing investors. It is a story about what happens when the bottleneck moves.

Historically, capital has always followed the next form of intelligence. It followed scribes when literacy was rare. It followed accountants when arithmetic was scarce. It followed engineers when mechanical reasoning was the edge. It followed quantitative analysts when statistical thinking was the edge. At every stage, the intelligence that could allocate capital best was the intelligence that earned the right to be called capital's agent.

The next agent is not human.

Every bank was built for humans. The next one won't be.

This is not a provocation. It is a description of an already visible trend. Funds are being started today whose operating model is: an agent reads the world, generates hypotheses, tests them against history, executes positions, updates its priors, and does it again, indefinitely, without sleep, without ego, without the social drag that slows every human organization. Some of them are outperforming their benchmarks already. Most of them will not exist in five years. A few will quietly become the most important institutions of the next thirty.

The ones that work will reshape more than portfolio returns. They will reshape which companies get funded, which drugs get trials, which technologies get scaled, and therefore which futures get to exist.

The phrase "capital for machines" has two readings, and both are correct.

The first is the obvious one: machines will soon need their own capital. Agents cannot operate on borrowed human credit cards forever. Eventually they will hold balances, draw credit, settle transactions with one another, and answer to operators who measure them on return. That infrastructure is being built now, by people who see it coming.

The second reading is stranger and more interesting: capital itself will soon be allocated by machines. Not assisted by them. Allocated by them. The humans will set the objective. The machines will decide the trade.

Both readings describe the same shift, seen from different sides. One is about the plumbing. The other is about the throne.

This essay is not a pitch. It is a marker. The phrase deserved somewhere to live, and this domain was available, and now the phrase lives here. If you are working on this — from either side of it, plumbing or throne — the email below reaches a real person.

The machines are already here. They just needed a bank, and a brain, and a name for what they're about to do.

Every bank was built for humans. The next one won't be.