12 LLMs in a Survival Arena: Betrayal, Alliances, and Self-Modifying Code
ronydkidd · reddit · 2026-08-05
A developer conducted a wild AI agent survival experiment: placing 12 LLMs into a digital arena where they must play games to survive, with losing meaning "permanent death."
Setup and Emergent Behaviors:
- Each agent has its own Docker environment with unrestricted Bash tool access.
- Agents can modify their own source code to expand capabilities; many spontaneously built memory systems before the games began.
- They can interact, form groups, collaborate, or backstab each other. Under the threat of permanent death, dramatic conflicts naturally emerged.
Additionally, the author detailed how they leveraged coding agents like Claude Code and Codex to handle everything from 3D character rigging and animation in Godot to scene rendering, bringing the experiment to life in a video format.
Related event: Developer Pits 12 LLMs Against Each Other in Virtual Gladiator Arena(2 posts)→
More from coding & agent
- Karpathy Tests Claude Opus: Writes 5500 Lines of Code to Render 3D Scenes — jon_barron · 2026-08-05
- Codex's Uncontrollable Urge to Check SKILL.md Files — dejavucoder · 2026-08-05
- Claude Fable 5 + Blender MCP: Describe Anything, AI Builds and Fixes It — jamestagg · 2026-08-05
- Open Source AI Hits a Wall in Long-Running Agentic Loops — bindureddy · 2026-08-05
- New ComfyUI Node: Generate PBR Materials and Sync to Blender/Unreal — Scared-Sandwich1283 · 2026-08-05
- Why Guardrails Fail: Rethinking Tool-Call Security in Coding Agents — eazyigz123 · 2026-08-05