The Next Hundred Million

There is a version of the no-code story that has been told and refuted three times already. This is not that story. The thing that changed is not the builder tools. It is that the input method stopped being a keyboard and stopped being English, at roughly the same moment, and almost nobody has repriced what that does to the market.

The structural shift

Software has always had an entrance exam. Not the logic, the interface. To build anything you had to type, in English, in a syntax, on a device shaped for typing. Every abstraction layer we shipped for thirty years, from spreadsheets to Airtable to no-code, lowered the difficulty of that exam without ever removing it.

Two things removed it, close together. Models became good enough to hold a specification that a person describes rather than writes. And speech became a reliable input across languages that never had a software ecosystem to begin with.

The exam is gone. And the people who were failing it were never the marginal builders. They were the ones sitting closest to the problems worth solving: operators who know exactly what is broken in their business and have never had a route from knowing to having.

The wedge

Not “AI writes code.” That market is enormous, crowded, and it is a market for people who already passed the exam. It makes existing developers faster.

The wedge is the population on the other side of the line, and the specific observation that their problems are small, high-frequency, deeply local, and economically invisible — which is precisely why no software company has ever served them. A distributor in a tier-two city with a fourteen-step reconciliation process is not a market of one. There are two hundred thousand of them, each needing something slightly different, which is unbuildable under the old cost structure and trivial under the new one.

The unlock is not that the software gets built. It is that it gets built by the person who understands the problem, which removes the translation step where almost all software value historically leaked out.

The framework

Three conditions have to hold together, and they only recently do:

1. Description beats specification. The person can say what they want in their own words and the gap gets closed by the system, not by them learning ours.
2. The language is theirs. Not translated into English and back. The idiom, the domain words, the way the business is actually discussed.
3. The artifact is real. It connects to their data, their payments, their customers. A demo that cannot touch the business is a toy, and this population has excellent instincts about toys.

Miss any one and you have a novelty. Hit all three and you have relocated who gets to make software.

My position

I am long the vernacular, voice-first end of this and sceptical of the English-language prompt-to-app category, which is competing for the attention of people who already have alternatives. The interesting customer is the one for whom there is currently no alternative at all, and who therefore has no expectation about what this should cost or look like.

India is the obvious first market, not for patriotic reasons, but because it has the highest density of sophisticated operators with no software serving them and a population already comfortable transacting by voice.

I am building here. Disclosure, not hedge.

The prediction

The measurable turn is not app-store counts. It is the first year in which more new business software is created by non-developers than developers, in any single large market. When that happens the incumbent tooling companies discover their distribution assumed a developer at the other end, and the entire go-to-market of the category has to be rebuilt.

I think that inversion happens in India before it happens in the US, and that this will surprise people who have only ever watched the US.

← all theses