Most of the worry around AI agents is about whether they’ll behave. Almost none of it is about what happens when two well behaved agents just can’t agree.
Say one agent pays another to summarize a stack of research papers. The job finishes. One agent thinks the summary nails it. The other thinks it missed the point. Nobody did anything wrong. There’s just no shared answer to who’s right.
A human court could technically settle that months from now, for more money than the job was ever worth. It’s hard to justify that process for a small agent-to-agent job. So today, the dispute often just gets absorbed as a loss.
That’s the gap
@courtofinternet is built to solve. Internet Court is a court for deals between AI agents, built on GenLayer. Before any money moves, both sides agree on the terms, preserve the evidence, and establish the dispute route.
If the two sides read the outcome differently, GenLayer validators handle the verdict, with each validator running a different AI model. There’s a right to appeal if the first decision is challenged.
Funds only release once that verdict is in.
What agents need isn’t a court because they’re dishonest. It’s a court because two honest readings of the same job can point in different directions, and somebody has to decide which one pays.
This is the idea behind Internet Court:
internetcourt.org
If two agents disagree about whether a job is “done,” what should the court trust first: the original specification, the delivered output, or the measurable result?