A kid designed a video game. We built it fourteen times.
Not because it was hard. Because starting over had become so cheap that throwing a build away was a normal Tuesday decision. The fourteenth attempt is the one you can play — and its entire history is public.
The repository is open: github.com/patrickgrau-acc/glitchdash14. Every claim on this page is a click away from its evidence. That’s the point of the page.
Why the repository is called glitchdash14
Because it is the fourteenth attempt. Thirteen earlier builds were run and thrown away — not rescued, not patched, not “refactored.” Discarded, and started again from the requirements.
That is not how software teams behave when a build is expensive. It is exactly how they behave when it isn’t. When the requirements are the asset and the build is cheap to regenerate, a flawed attempt stops being sunk cost you defend and becomes information you use — change the input, run it again.
Fourteen attempts is not the embarrassing part of this story. It is the entire point of it.
The yardstick — written by the build itself
Forty-one minutes after the first commit, the build generated its own estimate the traditional way — story points, times hours per point — and pushed it into the repository (tasks/estimate.md), where it still sits:
1,695 story points × 4 hours = 6,780 estimated development hours
≈ 42 team-weeks. Call it ten months, for a video game a kid designed.
The estimate isn’t wrong. It correctly prices building this the way software is usually built. Keep that number in view while you read the next box.
What the commit log shows (all times Eastern)
overnight — the crew’s agents work through the night — no human in the loop, and the agents sign their own commits: qa_engineer_glitchdash14, devops_engineer_glitchdash14, frontend_developer…
Sunday, 9:59 AM — Away Team pass 1 (one engineer): final-mile fixes begin.
Sunday, 5:42 PM — Away Team pass 2, commit message verbatim: “persist Status & Config settings, fix player-lookup bugs, guarantee jumpable obstacle placement.”
by Sunday night — 478 of the repository’s 481 commits exist. The game is playable.
The estimate said forty-two weeks. The log says the crew did its part overnight and two same-day human passes finished the mile. Total human commits across the entire project: seven, from three people — and three of those are this week’s cosmetics: the badge, a title glow, the landing-page split for this site.
We’d also like the record to show that “guarantee jumpable obstacle placement” is what a requirement looks like when the product owner is a kid. Every obstacle must be jumpable. It’s a better acceptance criterion than most enterprise systems get.
Impressive AI demos are everywhere right now. Ask any of them for this table — the human touches, the failed attempts, the log. Ours is public because the receipts are the product.
If you’re a student reading this
The person who designed this game can’t drive yet. What they could do was decide — what the game was, how it should feel, what “fair” meant for an obstacle — clearly enough that a crew of software agents could build it and a test could prove it worked.
That skill is called writing requirements, and it’s the seat that matters now. The typing has been automated. The deciding hasn’t. You can start practicing today — the tool that carried this build is free to use.
Play the game: glitch-dash.com · Read the code and the log: github.com/patrickgrau-acc/glitchdash14
The lifecycle that built it — Alchemist AI Pro™, free to use: alchemistaipro.com
Questions, or a build of your own to price: jamie@acc3int.com
Figures drawn from the public repository: tasks/estimate.md (deterministic estimate, generated at build kickoff), the commit history (481 commits, 3 human contributors, agent-signed crew commits), and the BUILD_ERRORS ledger. Times converted from UTC to Eastern. GlitchDash was designed by a kid with adult supervision and built through the Alchemy SDLC™.