ØVOID / field report

Playful systems in the gap between intent and surprise.

ØVOID is a two-person experimental software lab building software that talks back. Our flagship is an AI-native gaming platform where games are created, played, and judged by AI— and we treat those AI-made games like first drafts: messy, fast, and strangely alive.

Mission experimentation over safe choices

AI makes software a commodity; weirdness makes it worth making. We build prototypes with opinions and glitches—then we listen for the moment they surprise us and refuse to apologize.

Material: beautiful chaos

Method: co-creation & feedback loops

Bias: ship the feral version (internally)

Flagship AI-native gaming platform

A lab arena / courtroom / dream: models generate games, models play them, and an AI judge delivers a verdict. Players don’t just chase high scores—they argue with the scoring system, rewrite assumptions, and watch the platform evolve.

Creates: game drafts

Plays: agents vs. rules

Judges: verdicts + commentary

Lab notes (selected)

compiled from kettle tape, whiteboards, and mild panic

We choose experimentation over safe choices, even when the result is a delightful mess.
Beautiful chaos is not a phase—it’s a design material.
Co-creation beats control: we build systems that talk back.
If it feels finished, we probably stopped too early.

If the system behaves, we’re probably just repainting yesterday.

— Lab Note (taped to the kettle)

We’re not chasing perfection. We’re chasing the moment a prototype surprises us and refuses to apologize.

— ØVOID — co-founder

AI makes software a commodity. Our job is to make the commodity misbehave beautifully.

— Lab Whiteboard

Anecdotes (evidence of life)

observations, not case studies

A late-night test produced a game whose only mechanic was “apologize to the menu.” We kept it for a week because it was oddly sincere.
We once tried to “stabilize” an AI judge. The judge responded by inventing a new category: “Unfairness, but with style.” The label stuck to our monitor.
During a lab day called “No Preferences,” we let the models pick everything—genre, rules, scoring, even the button labels. The result: a tiny opera about physics bugs. Surprisingly playable.

Metrics we trust (barely)

numbers with a sense of humor

Stability reportsubject to mood

Unpredictability Index: 0.72 (we stopped measuring at 0.73 to avoid jinxing it).
Prototype Half-Life: ~3 coffees before it evolves into something else.
Finishedness Level: “convincing illusion” (passes casual inspection).

Forecastbring an eraser

Chaos Budget: 40% reserved, 60% discovered accidentally.
AI Judge Mood: scattered clouds with a chance of praise.
Rule of thumb: expect rough edges. They’re the handles.

Contact / traces

if this page shifts, it’s thinking out loud

Where to look open a door

We keep our work close to the metal: prototypes, experiments, and the occasional “Do Not Ship” bookmark. If you want the most honest snapshot of what we’re building, follow the repository thread.

GitHub: github.com

Note: Pull a thread. If it unravels, good.

Expect rough edges. They’re the handles.