All articles
ai

AI isn't killing content, it's exploding it

In 1975, BusinessWeek ran a piece called “The Office of the Future” and predicted that paper would be largely obsolete by 1990. The reasoning was clean. Screens were arriving, documents were going digital, and the cost of producing a memo on a terminal was a fraction of typing, printing, and filing one. Less friction, less paper. The paperless office became the standard shorthand for where work was heading.

It did the opposite. Office paper consumption roughly doubled between 1980 and 2000. The technology that was supposed to kill paper turned out to be the thing that multiplied it, and the worst offender was the one nobody saw coming.

Email.

In The Myth of the Paperless Office (2002), Abigail Sellen and Richard Harper found that introducing email to an organisation drove roughly 40% more paper, not less. The mechanism is the interesting part, and it is the same mechanism playing out with AI right now. Email did not replace documents. It lowered the cost of creating and circulating them to nearly zero. So people created and circulated far more. Every cheap email became a thing someone wanted on paper, to read properly, to mark up, to take into a meeting, to keep.

Make production cheaper and you do not get less output. You get more.

It is not a quirk of paper, either. The same pattern shows up wherever automation makes a unit cheaper. When the ATM arrived, the fear was that it would gut bank branches and the tellers who staffed them. The opposite happened. By making each branch cheaper to run, the ATM let banks open far more of them. Economist James Bessen documented the numbers: the tellers needed to run a typical urban branch fell from around 20 to 13, so banks opened roughly 43% more urban branches, and total teller employment actually rose, from about 500,000 in the 1980s to nearly 600,000 by 2010, even as the number of ATMs went from around 100,000 to 400,000. The machine that was supposed to replace the branch made the branch cheap enough to multiply.

That is the lesson worth holding onto, because a similar instinct is quietly taking hold about AI. If a model can summarise, compress, and consolidate, then surely it should mean fewer documents, fewer apps, fewer near-identical pages, with the machine trimming the fat for us. Less noise, more signal. It is a reasonable thing to assume.

I do not think it is what is happening. I think we are living through the paperless office again, at a scale the 1975 version could not have imagined.

The cost of making things fell off a cliff

The pattern repeats whenever the marginal cost of producing something collapses. The constraint was never demand. The constraint was the effort of production. Remove the effort and the latent demand floods in.

AI has just removed the effort from making almost every kind of digital artifact. A document, an image, an app, a website, a block of code. The thing that used to take a day takes an afternoon. The thing that took an afternoon takes a prompt. So we are not making fewer of them. We are making more of all of them, at once, in a way that compounds across every category at the same time.

The numbers are starting to show it.

Apps: a decade-long decline, reversed

App store submissions had been declining for years. The land grab was over, the cost of building and maintaining a real app was high, and the marginal new app was not worth the effort. Then the trend snapped.

In Q1 2026, worldwide app releases were up around 60% year on year, and close to 80% on iOS alone, reversing a near-decade of decline. Appfigures, reported via TechCrunch, attributes the reversal directly to AI vibe coding tools lowering the barrier to shipping software. People who could never have built an app are now building one in a weekend.

The flip side shows up in the moderation numbers. Google blocked over 1.75 million policy-violating app submissions to the Play Store in 2025. When production gets cheap, volume goes up across the whole quality distribution, the good and the junk together. That is not a reason to panic about AI. It is exactly what the email-and-paper curve predicts.

The web: most new pages now carry AI

The web is the clearest signal of all, because it is the easiest place to measure.

None of this content is replacing the web that already existed. It is being added on top of it. And the demand for somewhere to put it has not slowed either. Verisign’s Domain Name Industry Brief recorded 378.5 million registered domains in Q3 2025, up 4.5% year on year. More content, more pages, more places to host them, all at once.

The engine behind the volume is mainstream now, not niche. ChatGPT reached 800 million weekly active users in October 2025. Producing content with AI is no longer an early-adopter behaviour. It is how a meaningful share of the planet now drafts, builds, and ships.

Developers: the ecosystem is widening, not shrinking

The same expansion is happening in places you would expect AI to consolidate.

Take a single API as a proxy for one developer ecosystem. Strava reported more than 175,000 developers building on its API, with 25,000 added in the past year at its 2025 Developer Summit. One platform, one API, and the number of people building against it keeps climbing.

The shift in how that code gets written is just as telling. A quarter of the startups in Y Combinator’s Winter 2025 batch had codebases that were roughly 95% AI-generated. The barrier to producing a working software product has dropped far enough that the output is exploding, not contracting.

Corporate documents tell the same story as supporting colour. Google Workspace is past 3 billion users and Google Docs past 1 billion monthly actives. Microsoft 365 Consumer sits around 89 million subscriptions as of FY2025. These were already enormous surfaces for producing documents. AI is now wired directly into them, on every one of those desks. I am not going to invent a documents-per-day figure, because no credible one exists. But the direction is not subtle. The capacity to generate corporate content just got an AI multiplier applied to the largest installed base of document tools in history.

What this actually means

The instinct to read all of this as a problem is understandable. More content, more apps, more pages, much of it low quality, all generated faster than anyone can review it. That is real, and it is the same downside the paper curve had. Cheap production raises the floor of volume across the whole quality range.

And there is a catch the optimistic version skips over. When production becomes nearly free, the bottleneck does not disappear. It moves. It moves to consumption. A document still has to be read. An app still has to be reviewed. A pull request still has to be checked by a human who is now staring at code a machine wrote in seconds. We have made the writing close to infinite and left the reading exactly as slow as it has always been. So in the short term the explosion does not feel like liberation. It feels like more inboxes, more decks to sit through, more near-identical pages to wade past, more AI-produced work to read on the way to the thing you actually wanted. The cost of all that production did not vanish. It moved to whoever has to consume it. For a while, that means more of us doing the unglamorous work of reading what the machines are writing.

But the more useful read is the one the paperless office got wrong for two decades.

When you make production cheap, you do not reduce output. You change who gets to produce. Email did not just mean existing memo-writers wrote more memos. It meant people who never would have written a memo now sent fifty a day. AI is doing that for software, for writing, for design, for entire categories of work that used to sit behind a skill barrier.

The barrier was never the idea. It was the execution. AI removes the execution cost, and the latent supply of ideas, the ones people had but never had the time or skill to build, comes flooding through.

So the question stops being “how do we produce more”. We have solved that, permanently and accidentally. The question becomes “how do we make the things worth producing stand out in a world where producing anything is free”. That is a curation problem, a taste problem, and a judgment problem. Those are human problems, and they get more valuable, not less, as production gets cheaper.

Which brings me to vibe coding

The clearest version of this whole shift is what people now call vibe coding. The term was coined by Andrej Karpathy in February 2025 for AI-assisted development where you describe what you want and let the model write the code. It became Collins Dictionary’s Word of the Year for 2025. I have written before about what vibe coding means inside the enterprise, where the real opportunity is everyone who is not a developer finally being able to build the small tools they have always needed.

But the everyday version of vibe coding is even simpler than that. It is being able to have an idea on a Tuesday and have it live by Friday, just because you felt like it.

A small, real example. I vibe coded a thing called Bradsolutely. It is a single page where you type anything at it, a yes or no, an excuse, a decision you are stuck on, a regular Tuesday, and Brad, a sun-soaked, Jimmy-Buffett-flavoured pirate-captain hype man, rules on your life and hands back a verdict. It is gloriously unnecessary. It exists purely because the cost of building it had dropped to roughly the cost of describing it. A static page, one cloud function, a model behind it, a weekend.

Ten years ago that idea dies in my head, because it is not worth a week of evenings. Today it ships, because it is worth an afternoon. Multiply that by 800 million weekly ChatGPT users and 175,000 developers on a single API, and you have your explosion. Not one big surge of serious software, but millions of small, cheap, specific things that previously would never have justified the effort.

The paperless office was never wrong about the technology. Screens did get cheaper. Documents did go digital. It was wrong about human behaviour. It assumed that making production easier would make us produce less, when the entire history of every tool we have ever built says the opposite.

AI is not the exception to that pattern. It is the largest example of it we have ever seen. It is not killing content, or apps, or documents, or code. It is doing to all of them what email did to paper, exploding the volume by removing the cost.

The smart move is not to fight the explosion. It is to get very good at the one thing it cannot do for you, which is deciding what is actually worth making.

Written by Bradley Hunt. For more on AI-first building, see the writing index.