How Claude Code optimized a game's battle engine: measure first, profile again, admit failures
Ivaricus · reddit · 2026-11-04
A solo dev building Houses of Veyra (Godot 4, 14k-line campaign script, 230+ tests) shares how Claude Code handled an "every soldier fights" feature: stage 1 was measure-only tooling with CPU throttled 4x to emulate phones. Profiling found per-soldier scans ate over a third of the frame at 80v80; the first grid fix (24.7→19.1ms) was flagged as disappointing by the model itself, which re-profiled and rewrote spacing as one batched pass per physics step. It verified 72 full battles matched old outcomes, then honestly reported "phones barely moved." Key practices: demand measure-first stages, prove rewrites return identical answers, compare on player-feel metrics, and reward the model for admitting a fix didn't work.
More from coding & agent
- Kubernetes promotes AI Conformance to permanent SIG Architecture subproject — TerryTangYuan · 2026-11-05
- Code mode isn't about MCP: it's about whether tool results go to context or program variables — ritakozlov · 2026-11-05
- Agent Arena: frontier models show 55-minute gap on long-horizon agent tasks — arena · 2026-11-05
- Agent-based kernel optimization faces a fidelity-throughput tradeoff: random data misleads, cold starts cost 12.7 minutes — simran_s_arora · 2026-11-05
- Solo dev builds window-free OS where a local Qwen3.5 9B generates every screen — leonardo_c · 2026-11-05
- Apple team ships Siri Evaluations: don't skip evals when wiring App Intents into Siri AI — rudrank · 2026-11-05