Latent Heat

That flat line is the most misread object in the economy right now. Everyone agrees on what it says. Almost nobody asks what kind of system it is measuring. Because there is one kind of system — a common one, maybe the most common one — in which a reading that refuses to move is not evidence of stillness. It is the signature of the largest possible change happening at full speed.

A system in the middle of melting.

The committee of 1761

In the early 1760s, a Scottish chemist named Joseph Black asked a question so plain it is hard to believe nobody had answered it: when ice melts, where does the heat go?

The assumption of his day was that a little warmth past the freezing point should turn ice straight to water. Black pointed out that if that were true, spring would be an annual catastrophe. The first mild day would release every snowfield in the Highlands at once, and the floods would tear the country apart. Instead, snow takes weeks to go. Something is drinking the heat.

So he measured it. Ice at the melting point absorbs heat — enormous amounts of it — and does not get one degree warmer. The thermometer sits at zero while energy pours in. Melting one gram of ice swallows as much heat as it would take to raise that same gram of liquid water from zero to eighty degrees. Black called the hidden portion latent heat, and the name is the whole idea: the energy is real, accountable, and invisible to the one instrument everybody trusts.

Here is what the plateau means physically. Temperature measures the state of a thing — how fast its molecules are jiggling. But a solid is not just molecules; it is molecules held. The lattice of bonds is what makes ice behave like ice. During the melt, every incoming unit of energy is spent breaking those bonds, and none is left over to raise the jiggling. The instrument reads structure-breaking as nothing at all. Then the last bond gives way, and the same energy that produced a flat line starts moving the needle again — all of it, at once.

A young instrument maker down the hall at Glasgow took the lesson personally. James Watt, working out why his model steam engine wasted so much coal, ran into the same hidden ledger in steam, learned that Black had already mapped it, and built the separate condenser around it. The machine that industrialized the world was designed by a man who knew exactly where heat hides.

What the economy’s thermometer cannot see

Now look again at the meeting at 9:40.

Energy is pouring into AI on a scale nobody disputes — capital, chips, power, and the working hours of nearly everyone I know. And the thermometers of the economy read flat: output per worker, headcount, the look of your own Tuesday. From those two facts, the commentary draws the obvious conclusion. The heat went in, the temperature didn’t move, so the heat must be wasted. Overhyped. A bubble.

Thermodynamics offers a different reading, and I think it is the right one. Output per worker is a temperature: a measure of the state of the current structure. But the energy is not being spent on state. It is being spent on bonds — the quiet lattice that makes an economy hold its shape. Job descriptions. Approval chains. What counts as proof of work. Which steps need a human and which never did. The contracts, the habits, the interfaces between one company and the next. Breaking and remaking those absorbs everything you pour in and shows nothing on the gauge, because the gauge was built to read the old lattice.

A flat reading during a phase change is not the absence of change. It is what maximum change reads like from the outside.

This is the record, not just my argument. When factories first electrified, they bolted motors into buildings designed around steam shafts, and the productivity statistics stayed stubbornly flat for decades — until plants were rebuilt around the motor, and the gains arrived in a rush in the 1920s. The economist Paul David told that story in 1990 to explain why Robert Solow could quip, three years earlier, that the computer age was visible everywhere except in the productivity statistics. A decade after the quip, the statistics jumped. Both times, the plateau was not failure. It was the melt.

And a melt ends the way phase changes end: not with a gentle slope, but with a step. The last bond breaks, and the energy that was disappearing into structure starts registering all at once.

1. The loudest “AI was overhyped” season comes just before the step.

   The flat reading is most convincing at the exact moment absorption peaks — that is what a plateau is. So expect the calmest charts and the most confident obituaries right before the break, and treat a chorus of “it changed nothing” as a timing signal, not a verdict.

2. Domains melt one at a time, each at its own temperature.

   There is no single melting point for an economy. Paper-heavy, rule-bound, screen-only work melts first; work anchored to bodies and buildings holds its lattice longer. The same person will insist nothing has changed at their office and watch a finished melt in another corner of their life in the same week — and will generalize from the office.

3. The thermometers get replaced.

   Output per worker measures the temperature of the old structure. The measure that will matter is structural: how much of a process now runs end to end with no human inside it. Organizations that track only temperature will learn about their own melt from the step.

4. The plateau is the only time you get to choose the new shape.

   While everything is molten, re-plumbing is cheap: roles, interfaces, standing orders, what you promise customers. The firms that spend the flat years growing new bones will look identical to the ones that didn’t — until the step, after which the difference is the whole game.

5. The refreeze is fast, loud, and then closed.

   When a domain finishes melting, new bonds form quickly — new defaults, new contracts, new habits — and freezing releases latent heat: a short burst of visible, dramatic change that everyone will mistake for the revolution itself. Then the new lattice sets, and it will resist the next melt as stubbornly as the old one did. The window in which the shape is choosable is brief.

I am living on the plateau myself, and I can report that it reads flat from the outside. My site looks the way it looked years ago: a post a day, under the same name, at the same address. The temperature has not moved. Underneath, every bond is different — I no longer type the HTML, stage the drafts, or push the publish button; my AI does, while I sleep. The structure melted while the reading held still. I wrote about why your senses can’t feel this kind of change in The Leans — the inner ear reports change, not state, and so does your sense of your own job — and about why surviving the step requires new structure, not more effort, in No Giants Without New Bones. This piece is the thermodynamics underneath both.

How this could be wrong

Already true

The plateau-then-step shape is the recorded history of general-purpose technologies, twice over: factory electrification’s flat decades before the 1920s surge, and Solow’s 1987 paradox before the late-1990s jump. And Black’s physics is not an analogy under dispute; the heating curve is laboratory fact.

What has to happen

The step has to actually arrive somewhere specific. Within a few years, at least one large white-collar domain that spent the plateau being declared “unchanged” has to flip from mostly-human to mostly-machine within a budget cycle or two. The structural measures — processes running end to end without a person — have to climb during the flat years, visibly, for anyone who bothers to look.

Where I am probably wrong

A flat thermometer only proves a melt if the heat is genuinely flowing into bonds. Sometimes flat is just flat: the energy can leak away as waste — demos, pilots that die, slideware — breaking nothing. And the economy signed no contract with thermodynamics. It may be all slope and no step. If the change arrives smooth and proportional in every domain, with no plateau and no break, then I mistook ordinary warming for melting, and this piece was a weather report, not physics.

Go back to the meeting at 9:40, to the chart that will not move. Black’s answer to that room is two and a half centuries old. The reading is honest. The instrument is honest. They are just both aimed at the one quantity this kind of change never touches until the end.

The thermometer moves last.