In the second half of the nineteenth century, the cost of coal fell sharply. Engines got better; mines got deeper; rail moved tonnage that used to move on river barges. Economists at the time made a confident prediction. Cheaper coal would mean less coal, the same heat, the same light, the same iron, for fewer tons in the ground.
The opposite happened. Cheaper coal meant more coal. Cheaper coal made railroads, and railroads made cities, and cities made factories, and factories made everything else, and everything else needed coal. The ton of coal that fell in price in 1865 looked like a luxury saved. Twenty years later it looked like the precondition for a world that burned fifty times more of it. The economist William Stanley Jevons named the pattern. The price of a thing we use a lot of can fall, and consumption of that thing can rise faster than the price falls.
We think software is having its coal moment.
Cheaper code does not mean less software. It means more software, and more of the work around the software.
The cost of writing a working line of code has fallen for two years now at the speed of model improvement. The fall is real. We have watched it inside our own customers’ sessions. A tool that took a contract engineer six weeks now takes a marketing manager eleven days. A working prototype that used to clear a quarter’s engineering backlog now clears it on a Tuesday. The numbers are real; the productivity is real; the savings, in the narrow sense, are real.
And then the demand curve moves. The marketing manager who wrote the tool finds three more tools she would like to write. The COO who saw the first one ship asks her to ship four more. The lawyer who heard about it has six. None of these tools would have been built before; they were not in any backlog; they did not justify a developer-month; they would have lived as a spreadsheet, or a Notion page, or a fact someone had to keep in their head. They exist now because the cost of making them fell below a threshold.
Multiply that by every department in every company in every country. We are not in the early innings of a contraction in the demand for software. We are in the early innings of an expansion that will, in retrospect, look more like the rail boom than the cost-saving narrative ever predicted.
The interesting question is what happens to engineers in that world.
The naive prediction, the same prediction the nineteenth-century economists made about miners, is that engineers contract along with the price of code. They don’t. The price of code is falling, but the work that surrounds the code is not falling at the same rate. Code that runs in production has to be reviewed, integrated, deployed, secured, observed, owned. Code that handles money or health information or someone’s identity has to be defended. Code that breaks has to be fixed by somebody who can read it. The line of code is cheaper. The shipping of the line is not.
Engineers, in the new world, do less line-writing and more line-shipping. The work is denser than the work it replaced. Reading a stranger’s diff at 2pm and saying this will hold, ship it takes more skill, not less, than writing the diff would have taken. Reading a stranger’s diff at 2pm and saying this will not hold; here’s why; here’s what to do instead takes more skill again. The software engineer of 2030 will write fewer lines than the software engineer of 2020. They will be responsible for more of them.
We think this is the correct shape for the next decade. More software, written by more people, supported by more engineers per ton-of-code-shipped than the previous era ever needed. The bench, in our terms. The press. The relay. A software engineer one click away from the moment a person who isn’t an engineer is about to ship something to a customer.
The coal economists missed the boom because they were measuring the wrong thing. They were measuring tons-per-furnace, not furnaces-per-country. We think the people predicting the contraction of engineering are making the same mistake. They are measuring lines per engineer. They should measure systems per company, and people downstream of those systems, and the pace at which the next system gets built. By those measures, engineering is the largest growth sector of the next ten years.
We are building Relay because we believe that. The price of writing software fell. The volume of software is going to rise faster than the price fell. The work in the gap, the human shipping the machine’s output, is the work we are organizing the category around.
The founders, May 2026.