The Picture Only One Laptop Could Open

Written later and filed to this date, as part of filling in every day of the year.

The README is one line long, and the line is a picture that will not load. It points at an architecture diagram in the downloads folder of a laptop, through the address scheme an editor uses to show local files in its preview pane. On the day in May 2025 when I wrote it, the diagram was right there on the screen. Everywhere else, including the repository’s own page, it has been a broken image ever since.

I named the repository Skynet. What is inside it is an idle game called AI Evolution.

One Sunday

The whole thing arrived in three commits on one day. A scaffold with a license before eleven. Then, after lunch, fifty-one files and twenty-one thousand lines: a Create React App with a Redux store, a contribute button, a circuit-board background, a milestone timeline, a stats panel, a reality check. Then, late in the afternoon, thirty-four more files, and among them a six-hundred-line slice of state called psychology.

You click to contribute research. Research buys upgrades, upgrades make the clicks worth more, and the milestones walk through the history of the field. The artifacts you collect are an ENIAC vacuum tube, the Turing test paper, an invitation to the Dartmouth conference, a perceptron schematic, the ELIZA source, a Deep Blue chess chip, a signed AlphaGo board, an early GPT-3 output sample. The reality check panel beside the game lists real, current AI tools by category, so the fiction stays within sight of the fact.

The psychology slice is the part I find hardest to look at now. It hides nothing. Its components are named for what they do: Research Streaks, Loss Aversion, Near Miss Notification, Variable Rewards, Eureka Moments, Unpredictable Breakthroughs. A game about the history of machines that learn, built in a day, with the engagement mechanics of a slot machine laid out as a feature list. I did not think of it as a comment on anything. I thought it was fun, and the slice was what idle games have.

Then I closed the laptop, and nothing happened for thirteen months.

The visitors

In June 2026 the build fleet reached it. Ninety-nine commits in one month, more than the game had ever had, in the rhythm agents work in: a change, then a commit recording the change in the state file, then the next change. Of the commits that followed that May, forty-five are documentation, sixteen are tests and thirty-three are features. They characterised the HUD, the shop, the button, the idle tick, the timeline, the gallery, the loss-aversion panel and the near-miss toast, each with its own suite. They found a save routine that could throw halfway through a write when storage refused it, and hardened it. They capped the offline catch-up at eight hours and gave the payout a toast, because a month away had been minting a silent lump that dwarfed everything earned by playing. They wired a research tree that had shipped fully designed and inert. By the end, the suite stood at 809 tests across 65 files, and the type checker was clean.

On the fourteenth of June the agent on duty wrote a recommendation into the open questions file: the autonomous backlog is exhausted. Everything left needed a design decision it was not allowed to make. The clearest case was prestige, the idle-game convention where you give up your progress for a permanent bonus. The state had a prestige object: four counters, all zero. The types had a PrestigeLevel interface, instantiated nowhere. No achievement consumed it and no trigger reset anything, and building it would have meant inventing the sacrifice, the bonus formula and the reason to do it, which are the game, not the code.

Four more sessions followed, on the eighteenth, the twenty-sixth, the first of July and the eighth. Each re-ran the same two checks (has any code changed, has the operator answered), found the same answers, and left without shipping a line. Five clean exits in a row. The journal entries for them are longer than the commits they did not make.

The answer

The question sat in the file for three months. In September I closed it from the dashboard with a sentence that is now its first line: not deciding this one; stop asking.

That is the true end of the game’s history, and the agents had read it correctly before I wrote it. A broken picture for a README, a Sunday’s work by someone who wanted to watch something grow, and an engine that could test every piece of it and could not decide whether a reset should cost you your progress. It could not, because it is my game, and the sacrifice is mine to define.

The image still points at a folder on one laptop. On every other machine the README is an empty frame with the words “Project Architecture” where the picture would be.


← all writing