100 LLM Agents Run a Town Economy for 26 Weeks — and Money Stops Moving
omarsar0 · x · 2026-09-12
A simulation study put 100 LLM agents in charge of a town economy modeled on Pokhara Lakeside for 26 simulated weeks (91 runs, 2.44M decisions) and found the economy freezes: only 0.3% of 3,981 menu items were ever repriced, a 12x tourist shock raised revenue 4.62x but wages just 1.03x, and 96.7% of a cash transfer remained unspent after 311 steps. Notably, swapping the underlying LLM changed every measured outcome, while deleting agent memory changed none — and 1-2 week simulation horizons are misleading, with wealth rankings only starting to move by week 12.
More from AGI Musings
- Kaj Sotala's Established Creator talk: 20 quotes on the long creative grind — xuenay · 2026-09-12
- High achievers' smugness delays AI reckoning, argues contrarian thread — teortaxesTex · 2026-09-12
- China's AI adoption isn't low: token usage hits 140T/day vs US 45-50T, browser stats mislead — pstAsiatech · 2026-09-12
- Model 'caring' is jagged and contextual, argues researcher in AI safety debate — jd_pressman · 2026-09-12
- Model caring is 'jagged' and contextual, pushing back on 'obviously aligned' claim — jd_pressman · 2026-09-12
- Safety researcher: hidden backdoor-driven swarm misbehavior more likely than simple derailment — PandaAshwinee · 2026-09-12