Scenario: A decentralized autonomous organization is designing a multi-agent system with extremely long context windows. Which of the following approaches is recommended?