How to use Codex to tune a browser rally game
Dean Hipkiss used Codex to build DUSTLINE, then playtested handling, rivals and collisions, turning each observation into fixes before publishing the game.
deanhipkiss.com
Original prompt
Shared by the source author. See source notes for context.
Make AI cars more competitive, it is currently too easy.
Reach people exploring AI agents
See placements →Evidence at a glance
Sourced account- Source author
- Dean Hipkiss
- Original account
- Read the source ↗
- Published
- Source link checked
- Reported result
- DUSTLINE playable on Cloudflare Pages.
- Human involvement
- Author playtested and supplied gameplay-video reference; judged camera and drift feel.
A linked account supports attribution. It does not mean we independently reproduced its result.
Report an issue with this case
Open correction emailOpens your email app with the case link included. Nothing is sent automatically.
How the agent did it
- Codex implemented TypeScript/Three.js/Vite game from a detailed original-assets brief.
- Author drove browser builds and reported weak rivals, collisions and drift feel.
- Agent adjusted handling, terrain sampling, tree collisions and race settings.
Dean Hipkiss used Codex and ChatGPT Astra to build DUSTLINE, an original browser rally game. His brief specified a car and circuit, chase camera, drifting, road surfaces, opponents, race timing and several control methods. Codex implemented a TypeScript application with Three.js and Vite, generating geometry and assets in code.
The useful work continued after the first playable version. Hipkiss drove the game and reported that rivals were too easy, collisions did not represent the car body well and the drift felt wrong. Codex adjusted opponent pace and passing, body collision bounds, road surfaces and airborne behavior. A supplied gameplay video gave further visual context for handling changes, although it could not expose the reference game's internal physics or player inputs.
Playtesting also found specific rendering and collision defects. Cars used road elevation while off-road, the starting grid flashed, and trees had no collision bounds. The agent revised ground sampling, the grid material and tree collisions. It later added configurable race settings.
DUSTLINE was published on Cloudflare Pages. Automated checks covered mechanical behavior, but real-device performance testing remained incomplete.
Prompt
Original prompt
Source: Blog · Dean Hipkiss
Frequently asked questions
What did playtesting catch?
Hipkiss reported weak opponents, unsatisfying drifting, cars using the wrong off-road elevation, a flashing starting grid and missing tree collisions.
What did the gameplay video provide?
It supplied visual context for handling and camera changes. The author notes that sampled frames could not reveal player inputs or the reference game's internal physics.
Did automated checks prove the game felt good?
No. The author judged driving feel through browser playtesting and says real-device performance testing remained incomplete.
More details
Remaining steps
- Built/tests run before deployment.
About this account
Codex implemented and revised game systems, investigated concrete playtest reports and added checks; the author judged handling and playability.
Primary first-person account opened and checked on 2026-10-14. Real-device performance testing remained incomplete; passing tests did not establish fun. Outcomes and test claims are attributed to the source and were not independently reproduced. Original prompt excerpt: reproduced verbatim from the first-person account.
Credit: Dean Hipkiss on Blog
Source published
Added to Agents In Use
Source checked
Tools and topics
Codex · ChatGPT Astra · Three.js · TypeScript · Vite · Cloudflare Pages
Browser game · Playtesting · Driving physics
Editorial completeness: 100/100. Measures documented fields, not agent performance or verified outcomes.