It sounds hopeful, instead of each AI agent or the claw pursuing its own maximization effort. But how do we implement it in the real world?