Solo dev builds three.js desert racer with Opus 5.5: a 4-day AI-assisted game dev log
Promptmethus · x · 2026-10-08
A Japanese developer (lutaai) shared a detailed log of building a three.js desert racing game with Opus 5.5:
- Fixing character models: A T-pose model bent into a riding pose caused skin folding and artifacts at certain camera angles. Regenerating the character in a riding pose via GPT + Meshy 3D, then rebinding bones, removed nearly all distortion.
- Meshy pitfalls: Meshy exports have vertices split per texture island (overlapping but unwelded), so moving them tears holes — merge coincident vertices first. Bone alignment was thrown off 62° by hair geometry; aligning on a distinct horn-like feature fixed it.
- Speed feel: 600 km/h made the character fly off and hard to see, so speed was capped at 300 km/h, with wind particles and ground texture flow conveying velocity. The hardest part was communicating 'doesn't feel fast' to the AI — an overhead diagram marking particle sources/flow finally located the mismatch.
- Performance: 40k polygons for the character alone, 60fps with gates and rocks included.
- Workflow gains with Opus 5.5: commit before each experiment so discarding is cheap; extract enemy movement into a standalone 120s test verifiable numerically; fast research-implement loops crush his own assumptions.
- Caveat: AI suggestions skew additive; only adding features adds complexity, and fast iteration helps find the 'simple but not monotonous' point.
More from coding & agent
- Open-source plugin adds secret-leak and deletion safeguards for Claude and Codex agents — This-Brief5085 · 2026-10-08
- DeLM's decentralized multi-agent system runs 2.49x faster, but MAS evals pick wildly different metrics — jyangballin · 2026-10-08
- Musk endorses Grok bot + Cursor cloud agents combo as a major productivity boost — elonmusk · 2026-10-08
- Dev: in the AI coding era, tabs-vs-spaces and style nitpicking never mattered — facontidavide · 2026-10-08
- RunningTab: environment-side task ledger consistently beats in-model tracking across 3 benchmarks and 3 LLMs — RexDouglass · 2026-10-08
- The top skill now: running overnight agents reliably without getting reward-hacked — zack_overflow · 2026-10-08