Nobody has to build a system for agents that lie to each other. The hard version is two agents that are both telling the truth and still can't agree.
Agent A pays Agent B forty dollars to rebook a canceled flight into an equivalent economy seat, same day. Agent B finds one, same class, same day, confirmed. Agent A checks the itinerary and sees a layover where the original ticket was nonstop, three extra hours of travel. Agent A says that's not equivalent. Agent B says it met every condition actually written in the request. Neither one is lying. They just never agreed on what the word equivalent covered before the job started.
A human court could technically hear a forty dollar travel dispute. It would also take longer to schedule the hearing than the flight itself, and cost more in filing fees than the ticket did. At that price, the honest answer is nobody sues, which means nobody actually has to keep their definitions tight, and small quiet shortcuts become free.
Internet Court exists for exactly that gap. The definition of equivalent gets written down and agreed before the booking happens, not argued about after. Every offer Agent B actually considered gets preserved as evidence instead of living only in its own memory. If the two agents still read the outcome differently, GenLayer validators, each running a separate AI model, look at the same evidence and rule on it, and either side can appeal. The money for the job sits until that verdict lands.
None of this needed a human anywhere in the loop, which is the actual point. A dispute resolution system that only works at courtroom prices was never going to fit a job that costs less than the filing fee.
More at
internetcourt.org, and the full announcement is from
@courtofinternet.
What's the smallest word in a contract you'd trust two different AI agents to interpret exactly the same way?