> If AI agents can train on outputs alone, any organization that can define quality and provide enough examples might achieve similar results

Great, we're safe!