Claude-tested bots for weeks, then 219 real players flipped this game's difficulty curve
idlerunner00 · reddit · 2026-10-25
A developer building Out of Pixkrieg, a browser multiplayer extraction RTS made almost entirely with Claude Code and Opus 5.5, used Claude-built bot players (four play styles) and an economy simulator for balance: bots lost 0-2% of units on the map's calm edge and about half in the deadly core, 30-36% overall per raid.
Then 219 real players showed up for a weekend and inverted the curve: edge 10.4% lost, core just 6% (95% of core raids escaped), 7.3% overall. Humans play far more cautiously than bots.
Lessons for working with Claude: bots only play the way you wrote them — good for validating rules, useless for predicting humans; turn design targets into self-measured numbers (a dashboard tracks 19 real-player targets, 11 were off, plus 22 sim-only metrics) so Claude iterates against measured gaps, not gut feeling; real players remain the final test — open weekend playtests, feed data back to Claude, and verify next round.
More from coding & agent
- Adaptive Skill Granularity Framework Routes Coding Agent Skills at Runtime, Hits 100% Success — dair_ai · 2026-10-25
- Study: LLMs rewriting their own ARC-AGI-3 harnesss gain just 2.8 points, not significant — omarsar0 · 2026-10-25
- Dev adds thread tags to agent sidebar to track orchestrators across machines — kevinkern · 2026-10-25
- Dev experience: once your agent works well enough, you can ask it how to improve itself — Then-Programmer7221 · 2026-10-25
- AIOps didn't cut alert volume—the bottleneck is the runbook no one trusts to automate — krishnan · 2026-10-25
- Hermes ships subagent routes plugin for per-task model orchestration — intellectronica · 2026-10-25