Student Builds Medieval Kingdom of 85 AI Agents Solving 12 Puzzles in 30 Days Using LangGraph
Zethixcut · reddit · 2026-11-03
A student built a medieval kingdom simulation where 85 AI agents must collaboratively solve 12 puzzles within 30 days, aiming to land an agentic AI internship. A King and 5 council ministers debate and set direction, with orders cascading down to Lords and workers.
Key design elements:
- Memory inheritance: when an agent's context fills up, it dies and its memory is archived and handed to its successor;
- A learning council: the council retains what works and drops what doesn't across iterations;
- Real budget: agents spend an actual daily token budget — when it runs out, the kingdom starves until reset.
The engine is written from scratch in LangGraph, with a working 3D viewer. The author is seeking feedback on the project and whether it's enough for an internship.
More from coding & agent
- Agent-designed nested Kubernetes horror: kcp partitions on bare metal under KubeVirt under guest clusters — sloppenheimer · 2026-11-03
- Real-world agent test: flight and hotel rebooking burned tokens, manual app was still easier — curious_vii · 2026-11-03
- Dev ditches adversarial AI code review: 'a waste of tokens', prefers verification — sloppenheimer · 2026-11-03
- YC's Momentic launches QA-Bench: its AI QA agent finds 2.7x more reproducible bugs than Codex — ycombinator · 2026-11-03
- Ex-professional translator open-sources Subtitle Foundry, a modular AI subtitle workbench — mrtdlgc · 2026-11-03
- 8-layer architecture for building agent skill libraries: Skill = Trigger + Procedure + Resources + Checks — 2C_ornot2C · 2026-11-03