Is Using an AI Judge to Settle Disputes Reliable?

Last Wednesday I saw someone on Hacker News showing off a real-time debate game called wram.chat. The core gameplay is two people each holding a viewpoint, with an AI acting as the judge to decide who wins on the spot. I casually had a few engineers on my team who love arguing try it out, and we ran seven matches internally within 48 hours. Topics ranged from "microservices vs monolith" to "should Python type annotations be mandatory," and it was quite lively.