Vernacular Software

It's 11:40 on a Tuesday night. A program director at a small nonprofit sits at her kitchen table with three spreadsheets open, trying to figure out which families still owe money for summer camp. She could explain exactly what her organization needs in two minutes. What she can't do is build it.

She isn't unusual. She's the nurse working around a records system built for someone else's hospital, the dispatcher retyping the same load into three different tools, the engineer whose inspection notes live in a shared drive nobody can search. Millions of people who know their work better than anyone alive, stuck inside software someone else decided they should have. We've lived this way so long it's hard to see how strange it is.

Two hundred years ago, a farm family that needed a barn didn't hire a builder. They called a raising. The neighbors came, dozens of them, lifted the frames upright with long poles, and by evening a barn stood where there had been a field. Nobody there had trained as an architect. Most of the buildings people have ever lived and worked in were made this way, by the people who needed them, from what they knew. Architects have a word for it: vernacular.

Every kind of human building has a vernacular. Software never did.

Software has always been written in a language of the few. Today it's called Python and JavaScript and SQL, and the arrangement is ancient: a small class speaks the language, and everyone else describes what they need and hopes it survives the translation. The person who understood the problem was almost never the person who built the solution. That's why so much software is good enough for everyone and right for almost no one, and why the most important system in so many organizations is a spreadsheet only one person understands.

We've seen this arrangement break before. For centuries in South Asia, scripture, law, and medicine lived in Sanskrit and Persian, languages most farmers and weavers couldn't read. Then in sixteenth century Punjab, ordinary speech was given a script of its own, Gurmukhi, simple enough for a child and meant for everyone. In Europe, Dante wrote in everyday Italian instead of Latin. Each time knowledge left the specialists' language, it stopped belonging to institutions and started belonging to people.

Software's turn has come.

Most people think AI's gift to software is speed: engineers coding ten times faster, anyone typing a sentence and getting an app. But an app made by prompting is still code, and code drags everything behind it. The hosting, the security, the backups, the day something breaks and nobody knows how to fix it. Prompting your way to an app doesn't free you from engineering. It quietly makes you an engineer.

The real change is bigger. For the first time, a machine can reliably understand a person describing their work in plain language, turn that description into a real, working system, and keep it correct as the description changes. The description stops being a request to someone who speaks the specialists' language. It becomes the system.

That is vernacular software. A camp director's system is made of families, sessions, and scholarships. Its rules are sentences anyone can read: Medical notes are only visible to the camp nurse. Three days before each installment is due, remind the family. Nobody using it ever has to know that code exists. And it's built to last, like the barns still standing two centuries later: a real database, a record of every change, a system that grows with the organization and belongs to the people who built it.

The old question was who can build this for us? The new one is how does our work actually work? Answer the second, and the system follows.

This is the start of a revolution in who gets to build. In 2010, the most valuable skill in the world was writing code, and the people who had it shaped nearly every tool you use. In the decade ahead, the scarce skill is understanding a piece of the world deeply and describing it clearly. That skill has always belonged to millions of people: social workers, teachers, nurses, farm managers, the person at every small organization quietly doing five jobs. It just never had a way to build.

Not all of it will be good. Work built on translating between those who know and those who build will shrink, and there will be a flood of careless tools that break the first time they're trusted with something real. But look at which way the power moves. For fifty years it has flowed toward people far from the work. Now it flows toward the people closest to it.

Go back to the kitchen table. A few years from now, she isn't there at midnight. Earlier that day, between meetings, she spent twenty minutes describing out loud how camp registration should work. She read it back, changed two sentences, and approved it. By afternoon, families were registering on a page that looked like her organization made it, because it did.

The person who understood the problem built the solution. That has almost never happened in the history of software. It's about to happen everywhere.

Farm families raised barns because nobody was coming to do it for them, and because they knew better than anyone what a barn had to hold. That has always been enough to build what people need. Now it's enough to build software.