A button that refused to stay put
A late-night build produced a button that wouldn’t stay in one place. We almost fixed it—then realized it was the most honest part: software that won’t pretend it’s finished.
Current project: an AI-native gaming platform
created • played • judged by AI
We’re building a platform where games are created, played, and judged by AI—so the loop is fast enough for weird ideas to actually survive.
Think of it like a stock ticker that doesn’t report prices—it reports possibilities. Every game is a draft; every verdict is taste, not just rules.
Note: The website is the brochure. The product is the engine.
Loop sketch
co-creation by default
Create
A human prompts (or pokes), an AI drafts mechanics, tone, and constraints—then we keep the parts that feel alive.
Play
AI players explore the edges: speedruns, exploits, “what if I refuse?” behavior—stress-testing vibe and balance.
Judge
An AI judge scores (and writes) feedback that can sound like a referee doing poetry. It judges the experience, not just compliance.
Rule #1 Make it readable, then make it risky.
Rule #2 Co-creation is the default: systems respond, humans respond back.
Rule #3 If it behaves perfectly, it probably isn’t alive yet.
Philosophy: experimentation over safe choices
beautiful chaos, with intent
Our north star is not “best practices.” It’s “best questions.” We ship drafts with intention and let the next version surprise us.
We treat misbehavior like a signal: sometimes it’s a bug; sometimes it’s personality. The job is learning the difference.
Working definition: Beautiful chaos is the moment your tools surprise you and you decide not to punish them for it.
AI makes software a commodity; taste becomes the differentiator. expand
We design for preference, not perfection: what feels like a room you want to walk into twice.
We choose experimentation over safe choices, even when it’s louder. expand
“Loud” is fine when it carries meaning. We avoid noise that doesn’t teach us anything.
Beautiful chaos beats polished certainty—especially early. expand
Early polish can hide the wrong idea. We’d rather reveal the right problems first.
We ship drafts with intention and let the next version surprise us. expand
Shipping isn’t “done.” Shipping is how you find out what the thing actually is.
If it behaves perfectly, it probably isn’t alive yet. expand
We look for edges: contradictions, exploits, and accidental delight—then decide what to keep.
Signals (we measure what we’re willing to keep)
ticker, but truthful
Chaos Index
0.72
Higher means more delightful accidents survived review.
Draft Velocity
3.4 / week
First versions shipped per week (counting the ones we immediately regret).
Novelty Drift
+18%
Measured in “wait, what?” moments per scroll—surprise with structure.
Co-creation Surface
42 touchpoints
Places where the system can respond back (and sometimes disagree).
Lab notes (things we keep on purpose)
pinned drafts
“Every game in AI-native gaming platform where games are created, played, and judged by AI is a first draft. That’s not a limitation; it’s a distribution strategy for weird ideas.”
“If AI makes software cheap, then our job is to make it interesting—like a room you want to walk into twice.”
“We’re not chasing ‘best practices.’ We’re chasing ‘best questions.’”
“Beautiful chaos: the moment your tools surprise you and you decide not to punish them for it.”
The judge wrote like a referee-poet
While testing AI-native gaming platform where games are created, played, and judged by AI, an AI judge delivered feedback that sounded like poetry written by a referee. We didn’t get it at first—then we realized it was judging the vibe, not just the rules.
Artifacts (evidence of process)
kept for personality
Index card “Chaos Budget: 15% UI mischief, 85% clarity.”
List “Do Not Solve” problems we refuse to over-optimize.
Changelog “Today’s Mood” describing what the interface felt like.
Folder “almost-features” prototypes too strange to ship, too alive to delete.
Timer “Dare Mode: 17 minutes.” When it ends, you must ship a draft.
Spec “Make it readable, then make it risky.”
Screenshot AI judge delivering a verdict like a game show host.
Visual notes (text-only references): image://neon-lab-notebook-scan • image://ai-judge-scorecard-surreal • image://procedural-typography-swarm • image://two-desks-one-cable-chaos