Software Has No Value

More than two trillion photographs will be taken this year.¹ Kodak is bankrupt.

Hold those two facts next to each other, because they are the same fact. Demand for photographs did not soften. It went vertical. We shoot more pictures before lunch than a family in 1985 took in a decade. And the company whose name was photography, the one whose own engineer built the first working digital camera in 1975, filed for bankruptcy in 2012. The output exploded by orders of magnitude. The industry that sold the means to produce it died.

The value was never in the film. It was never in the camera. It was in the photograph, the moment you wanted kept. Film and cameras were only ever what that moment cost, until they stopped costing anything. When capturing became free, photographs went from scarce to infinite, and the apparatus built on their scarcity fell over.

Software is film.

Let me be precise, because the claim sounds larger than it is. Software has no intrinsic value. None. It is worth exactly the problem it solves and not one cent more, and that has been true the entire time. We are only about to find out we meant it.

You Were Paying for Scarcity

For forty years the price of software and the value of software were the same number, so nobody had to tell them apart. They were never the same thing. The value of a program is the problem it makes go away. The price was a rent on the fact that you could not make it yourself.

Building working software was hard. It took people who were rare, expensive, and slow to train, so a class of companies grew up around that scarcity the way a guild grows around any craft that takes a decade to learn, and they charged guild prices. The seat fee, the license, the per-user-per-month line: none of it measured what the software was worth to you. It measured how few people on earth could have built it. The artifact was a stand-in for scarce labor. You were paying the toll on a bridge that was expensive to build and that you had no way to build for yourself.

Marc Andreessen wrote that software was eating the world, back in 2011, and he was right. But eating the world is not the same as being one. A thing that is everywhere, that everyone makes and uses without a second thought, is not an industry. It is closer to a utility, or a habit. Software spent forty years as a utility that happened to be priced like a luxury good. That was the anomaly. We mistook it for the nature of the thing.

Everyone Makes It Now

The scarcity is ending in front of us. Not in a forecast. On screens, this quarter, in the hands of people who would laugh at being called engineers.

Early last year Andrej Karpathy gave the activity a name, vibe coding, and the name stuck because everyone abruptly recognized what it described. You say what you want. Something builds it. You look, say what is wrong, and it changes. No syntax, no stack, no four years of prerequisites in the way. I watch people with no technical background ship a working tool over a lunch break, for an audience of one, then close the laptop without a thought about who else might want it. That last part is the tell, and we will come back to it.

This is not “AI helps engineers write code faster.” That framing is the guild keeping the conversation inside the guild hall. The actual event is that the bridge is gone. The gap between having a problem and having software that solves it, the gap the toll was charging for, is closing for everyone, not only for the people who used to get paid to cross it.

But Can It Handle a Million Users

Here the engineer’s reflex fires, and it always fires the same way. Sure, but can it scale? What happens at a million users? How does it hold up under load, state, concurrency, the long tail of edge cases?

It won’t scale. And the question is a fossil.

Nobody ever centralized software because centralizing was good. They did it because building was expensive, so you spread that cost across the largest crowd you could gather. One rigid application served ten million people because building eleven applications, let alone ten million, was unthinkable. Then you spent a decade making those ten million people bend around the compromises required to serve all of them with one thing. Take the cost of building to zero and the logic turns inside out. You stop building one app for a market and forcing the market to fit it. You build the exact thing for the exact need, and you throw it away when the need is gone.

There will not be one chat app with ten million daily users. There will be two hundred thousand chat apps with fifty daily users each. Every one of those fifty will be using something shaped precisely to their group, instead of contorting themselves around a product designed for an average user who does not exist.

The reflex objects: chat has network effects, you cannot fragment it, everyone has to be on the same app. Look at what the network actually has to hold. Not the earth. Your family. Your team. Your fifty. That is the whole network you were ever really using, sitting inside an app built for two billion strangers so the vendor could afford to build it once. The belief that a chat app must serve millions to deserve to exist is not a property of software. It is the industry, talking through you. The fifty people you message are a rounding error the old model could not have served on purpose, and the new one serves them by default.

It won’t scale. Correct. It does not need to. Scaling was the answer to one question, how do we build this a single time and sell it to everyone, and fewer and fewer people will have reason to ask it.

We Have Already Done This Once

If this sounds like a prediction, it isn’t. We ran the experiment decades ago and the results are sitting on every computer you own.

By the only measure that matters, the number of humans who use it, the most widely used programming language on earth is the spreadsheet. Microsoft’s own researchers have said it plainly: more people write Excel formulas than write C, C++, C#, Java, and Python combined, and it is not close.² Hundreds of millions of people program, and almost none of them would ever use that word about themselves.

Now notice what did not happen. There is no industry of individual spreadsheets. Nobody productized the budget your aunt built for her catering business, or the tracker your manager keeps for the team, or the model an analyst threw together on a Tuesday to settle one question and never opened again. There is a tool, and beneath it an effectively infinite tail of bespoke little programs, most with exactly one user, most built by someone solving one problem for themselves and then walking away. A staggering share of the operating logic of the actual economy runs on these. None of it scaled. None of it had to. Each one worked for the person who made it, which was the entire point.

That is the future of software, in a picture that already exists. The spreadsheet only got there first because its language was simple enough for ordinary people forty years ago. The language is English now, and the domain is everything.

Software Becomes a Literacy

So the thing growing exponentially is not the software industry. It is software, the way writing grew once the printing press stopped being the bottleneck. There is a publishing industry. There is no writing industry, because everyone writes all day and nobody bills for the existence of a memo. Writing became a literacy, ambient and unpriced, while a much smaller industry kept selling the few things that genuinely still needed selling at scale.

The clever version of the bull case reaches for Jevons paradox, the way Satya Nadella did when cheap models spooked the market: make a thing more efficient and people consume far more of it, not less. That is correct, and it is the point. Jevons tells you the consumption of software is about to go vertical. It says nothing about that consumption running through an industry. Abundant software gets used the way abundant light gets used, everywhere and without thought, by people who never once picture themselves as customers of a lighting company. Cheaper, better lighting did not save the candlemakers. It drowned the world in light and moved the value somewhere else.

Somewhere else, for software, is down. The value drains out of the application layer, which de-industrializes, and pools beneath it: in the models and the platforms and the rails that let a hundred million people mint software on demand, and in the genuinely hard problems where scale and trust and coordination are the actual work rather than a side effect of cost. I made a version of this argument about knowledge work, where free generation moved the whole job onto judgment; software is the same story arriving a little later. The economy underneath is real, and parts of it are enormous. It is just not the one being forecast, the thousand new companies each renting one workflow to a million users. You do not rent what your own tools will mint for you while you describe it.

The people calling for exponential demand have the curve right and the conclusion backward. Demand for software was never demand for a software industry. It was demand for problems to go away. For one strange and profitable stretch, the only way to make a software problem go away was to buy software from someone who could build it, so we wrote the price on the artifact and started to believe the artifact was the thing.

It wasn’t. The photograph was always the point, and the film was only ever what it cost to get one. Software is the film. The problem you needed gone is the photograph. We are about to take two trillion of those a year, and almost none of it is going to look like an industry.

Footnotes

1. Photutorial estimates annual photographs past two trillion, with smartphones taking roughly 94 percent of them. ↩

2. “Excel formulas are written by an order of magnitude more users than all the C, C++, C#, Java, and Python programmers in the world combined.” Microsoft Research, announcing the LAMBDA function (2021). ↩