Mediator.ai launched a platform that uses large language models and 70-year-old game theory to negotiate disputes between two parties. The tool applies John Nash's 1950 cooperative bargaining solution, which mathematically identifies agreements where neither side can improve their outcome without hurting the other. Each person privately describes their situation, then the system generates candidate agreements, scores them against both accounts, and iterates until it can't find a better deal.
In one example on the site, two bakery co-founders deadlocked over equity got a creative fix: a 60/40 split where the minority partner can claw back 10% by returning full-time for six months or forgoing $24k in distributions. Neither had proposed this. The system found a deal that looked ahead instead of arguing over past grievances, adding a management salary and a shotgun buy-sell clause.
Hacker News commenters saw real potential. One user facing $20,000 in mediation costs for an HOA dispute wanted to try it. Another pointed to co-parenting conflicts as a natural fit. Someone mentioned andshake.app, a tool for clarifying agreements upfront.
The concept is sound. Nash bargaining is proven math. The question is whether LLMs can reliably translate messy human disagreements into structured terms the math can actually optimize. If they can, mediation becomes cheap and accessible. If they can't, you're getting game-theory window dressing on a black box. Worth watching either way.