Games are eating the world

In August 2011 Marc Andreessen published an essay in the Wall Street Journal arguing that software was eating the world. Fifteen years later nobody argues with it. Every industry he named got its software layer, and most of them got eaten by a company that started as a website.

The essay was right, and being right is exactly why it stopped being useful. A thesis that explains everything explains nothing in particular. "Software is eating the world" now describes the weather. What it no longer does is tell you where to look.

So here is the narrower claim, the one worth arguing about.

Software is eating the world again, and this time it arrives as a game.

The cost of making a thing collapsed

The interesting change of the last two years was not that models got smarter. It was that building a playable thing stopped being a project and became an afternoon.

In March 2025 Pieter Levels announced a game jam with a strange rule: at least 80 percent of the code had to be written by AI, the game had to run in a browser, no login, free to play. He gave people about two weeks.

1,170 games were submitted.

That number deserves a moment. Steam, the largest games storefront on earth, ships around 66 games a day. Over the same fortnight Steam would have shipped roughly 920. One jam, organised by one person on social media, out-produced the entire store.

The follow-up jam in 2026 raised the bar to 90 percent AI-written code and put 50,000 dollars on the table. The rule that matters is not the percentage. It is "runs in a browser, no login, free". That single condition turns the whole field into something measurable, because every game has an open URL.

When making costs almost nothing, the constraint moves. It used to sit on execution. It now sits on ideas, taste, and the ability to be found.

The numbers, and the arithmetic nobody does

Everything below is public. What is not public is what happens when you divide one of these figures by another.

The scissors

The global games market will reach roughly 205 billion dollars in 2026, growing about 4.6 percent according to Newzoo. That is a mature market growing politely.

Meanwhile Steam shipped 19,112 games in 2025 and had already shipped 11,979 in the first half of 2026. Hold that pace and the year lands near 24,000, a rise of about 25 percent.

Supply is growing roughly five and a half times faster than the money.

That is the single most important sentence in this piece. It is not a boom. It is a compression. Every new game arrives into a pot that is barely growing, alongside thousands more competitors than last year. Average revenue per title is falling structurally, and no amount of craft reverses arithmetic.

The clock

Steam adds about 66 games a day. itch.io adds around 400. Together that is 466 daily, which means a new game appears somewhere in the world every three minutes, day and night, forever.

itch.io alone has passed one million games.

Where the growth actually comes from

Valve requires developers to disclose AI use. In 2025, 19.9 percent of releases carried a disclosure. By July 2026 it was 30.8 percent, and over 17,000 titles carry one in total.

The forecast that more than half of new Steam games will disclose AI by 2027 or 2028 is the headline everyone repeats. The better number is underneath it. Before the disclosure requirement, AI-flagged launches ran at about 13 a month. They now run at about 530. Launches without a disclosure went from roughly 1,030 a month to 1,320.

Do the subtraction. Non-AI production grew by 290 a month. AI production grew by 517.

About two thirds of all growth in new games is AI-driven. Traditional game-making did not shrink. It simply stopped being where the expansion happens.

One caution, because it gets misread constantly: a disclosure means AI tools were used somewhere, not that a game was generated by a machine. Reading "one third of Steam is AI slop" out of that figure is wrong, and the people making the best vibe-coded games are not the ones producing slop.

What a person can actually see

Around 170,000 new games arrive each year across Steam and itch alone. A committed player finishes maybe fifteen.

That is 0.0088 percent.

And of the roughly 19,000 titles Steam shipped in 2025, about 300 earned more than a million dollars. 1.6 percent. The other 98.4 percent did not fail at making games. Most of them failed at being found, which is a different problem with a different solution.

Why games, and not some other kind of software

Here is where the thesis stops being about volume and starts being about interfaces.

Games are the only category of software where the interface is not a way to reach the product. The interface is the product. A spreadsheet has a job it performs whether you enjoy the experience or not. A game has nothing else. If the moment-to-moment feel is bad, there is no underlying utility to fall back on.

That constraint has been running as an unforgiving evolutionary process for fifty years. Everything we know about making a system feel responsive, legible, rewarding and worth returning to was worked out first in games, usually decades before it reached anything else. Loading states, progression, onboarding that teaches by doing rather than by telling, feedback tight enough that your hands learn it before your head does: all of it is native to games and imported everywhere else late and badly.

So the argument is not that people like games and would enjoy more of them. It is stronger:

Game interfaces are the best interfaces we have, because they were built under the only conditions that punish a bad interface with total failure.

Once building an interface costs almost nothing, there is no longer a reason to accept the worse one. The grey form survived because someone had to pay for the alternative. Nobody has to pay much for it now.

Gamification failed. This is not that.

The obvious objection is that we tried this and it was embarrassing. Points on a banking app. Badges for filing an expense report. Streaks designed by growth teams to exploit loss aversion.

That failed for a precise reason. Gamification bolted the rewards of games onto software while ignoring the mechanics. Points and leaderboards are the scoreboard, not the game. A scoreboard attached to something boring just measures how bored you are, in public.

What actually makes a game work is a loop: a system with real state, a decision that changes that state, feedback fast enough to learn from, and consequences you care about. That is expensive to build, which is why nobody built it for expense reports and everyone reached for badges instead.

That cost is what collapsed. Simulating a real system with a real feedback loop is now within reach of an afternoon. The thing gamification could never afford is the thing that just became cheap.

Businesses as playable models

Which leads somewhere more serious than entertainment.

For forty years, the way to reason about a business under uncertainty has been a spreadsheet. You build a model, you change a cell, you read a number. It works, and it has two deep flaws: only the person who built it can hold it in their head, and it answers one question at a time.

Business simulations exist. The Business Strategy Game has been running in universities for over forty years. But they teach with invented companies and made-up markets, because wiring a simulation to a real business was too expensive to justify.

From the other direction, enterprise digital twins now hold real operational data, and the stated ambition for 2026 is to move them from showing you what happened to shaping the next decision. They have the data and no loop.

Planning games have the loop and no data. Digital twins have the data and no loop. The gap between them is exactly what just got cheap to build:

Load a real business into a playable model and let people play their actual decisions before making them.

Not a dashboard of what already happened. A system where you cut the price, run four quarters in twenty seconds, watch churn and margin and cash move, and do it again with a different choice. Where the argument in the room stops being about whose intuition is louder and becomes a thing both people can run.

This is the part of the thesis with a direct commercial edge, and it is the part we intend to build.

What breaks first: finding anything

If a new game appears every three minutes, the scarce resource is no longer games. It is knowing which one to play tonight.

The existing infrastructure does not solve this, and largely cannot. MobyGames has curated since 1999 and is the most careful archive there is, which also makes it slow. IGDB has the best commercial API. RAWG covers 350,000-plus titles. All of them are organised around store entries.

A game that was made in an afternoon, runs at a URL, and never touched a storefront does not exist for any of them. That is the fastest-growing part of the medium, and it is invisible to every serious database.

Meanwhile the directories that do cover it, like vibegames.io, are catalogues without competition: no play counts, no scores, no ranking. And the leaderboard services, globalstats.io and the various open source kits, are infrastructure for developers to install, not places a player ever visits.

Directories without a contest are catalogues. Contests without a directory are invisible. Nobody has put the two together.

What this site is

Three things, and they feed each other.

An index. The figures on this page, kept live, with sources and dates, and the cross-connections nobody else computes. Where sources disagree we show both and say why, instead of averaging the disagreement away.

An arcade. Vibe-coded games worth your evening, curated for players rather than listed for completeness.

A world ranking. One score table across every game that joins, which is the piece that does not exist yet. A game adds one line of code and gets a global leaderboard. A player gets one identity and one ranking across all of them. Points are awarded by rank within each game, so playing widely counts as much as topping one board, which is precisely the behaviour worth rewarding when discovery is the broken part.

That last piece also produces something no aggregator currently has: play data on games that no store has ever heard of.

Software ate the world. It is coming back for seconds, and this time it has a score counter in the corner.

---

Sources

• Marc Andreessen, "Why Software Is Eating the World", Wall Street Journal, 20 August 2011
• Newzoo, Global Games Market Report, 2025 figures and 2026 forecast
• SteamDB, Steam Game Release Summary by Year
• Sulka Haro, "Three years of AI on Steam", monthly AI and non-AI launch rates
• PC Gamer, Steam Week in Review, on the AI disclosure trajectory
• PC Guide, on 17,000-plus titles carrying an AI disclosure
• itch.io, released games count and daily upload rate
• Vibe Code Game Jam 2025 recap, 1,170 entries; vibej.am for the 2026 edition
• GitHub Octoverse 2025
• vibegames.io, globalstats.io, IGDB, MobyGames, RAWG

Calculations marked as ours on the index page state their assumptions. The pile figure assumes eight hours per game, the attention figure assumes fifteen games played per year. Disagree with the assumptions rather than the arithmetic.