✳ little intentions

A LITTLE GAME ABOUT BIG IDEAS

A game helped me start building.

Playing Coding Planets in high school helped me turn an idea into instructions a computer could follow.

Little Intentions brings learning through play to communicating with AI agents.

Explore the mission ↗

A personal project by OT · software engineer in games & XR

Rendered 3D Orbit, a small companion with a visor
Coding Planets gameplay: a robot on a small puzzle world and a sequence of instructions
CODING PLANETS · GOOGLE PLAY ↗
Current listing screenshot

02 / THE SHIFT

Building starts with a clear intention.

As AI takes on more implementation, communicating intent and checking outcomes become essential building skills.

A group of people collaborating around a table at an earlier game jam
OT AT AN EARLIER GAME JAM
01THE GOALWhat should happen?
↗
02THE BRIEFWhat context matters?
↗
03THE CHECKWhat actually happened?

For adults familiar with AI chat but new to delegation, Little Intentions makes that loop tangible through play.

03 / YOUR COMPANION

Meet Orbit.

Your pet. Tiny worlds. Better instructions.

Express a goal, supply context, answer questions, approve a plan, and check what happened.

Actual Blender animation render · still shown until played

FROM A LITTLE COMPANION ↗ TO A 3D WORLD
Original Orbit artwork and poses
Orbit rendered in 3D

04 / TRY THE COMPARISON

One mission. Two instructions.

See what changes when the brief matches the goal. These are prepared Practice-mode examples, separate from the recorded live planner.

THE MISSIONBring tea to Mira.● PREPARED PRACTICE-MODE EXAMPLES

ORBIT'S PLAN

OBSERVED EVENTSREADY TO PLAY
    Choose play to reveal what happened.

    Both attempts succeed. Clarification helps the first; the revision supplies the missing item upfront.

    Recorded receipt — attempt 2: goal reached. 1 planner call · 337 input + 83 output = 420 tokens. Listening: 1 call, 4.0 seconds; speaking: 1 call, usage unknown.

    See what happened, how many planner calls it took, and the tokens used. Input includes context. Tokens are usage units, not a price or a score; clearer instructions do not always use fewer tokens.

    05 / MADE WITH CODEX

    From a sketch to a working journey.

    Built with Codex. OpenAI powers Orbit’s listening, planning and voice.

    ASTRA + BLENDER
    Authored Orbit’s mesh, rig and five animation clips.

    TERRA + CODEX
    Integrated the browser/XR game, action validation, approval and rewind.

    Export inspection caught a visor shading issue; the mesh was corrected before integration.

    How it works & evidence

    OpenAI transcription, a live planner and generated voice support the Quest experience. The player reviews and explicitly approves proposed actions; the simulation validates them before execution. OT confirmed spoken clarification, interruption, delivery, rewind, comparison, historical replay and readable per-attempt usage receipts on Quest.

    Transcription: gpt-transcribe. Planner: gpt-6-luna via Responses API. Generated speech: gpt-4o-mini-tts, coral. The final video has OT’s narration and music; the original Quest capture did not record Orbit’s audible speech. Practice mode uses local rules. One village and mission are prepared; levels are not generated live and there is no general room understanding. This pitch site has no hosted live-game backend.

    Render of Orbit reimported from the exported GLB
    01 / EXPORT CHECK The shipped GLB, reimported and rendered.

    06 / AN INVITATION

    ✳

    “You Can Just
    Build Things.”

    Little Intentions helps you practise making your intentions clear.

    View source on GitHub ↗

    The event card's message · Little Intentions

    FINAL DEMO · 84 SECONDSOT’s narration and music · recorded Quest gameplay

    The headset capture was silent; the finished video contains OT’s narration and music. Download final video ↗ · Earlier live-planner desktop recording (silent) ↗