In materials research, it may be useful to configure separate agents for literature summarization, crystal structure parsing, composition and phase-stability checks, experimental log interpretation, and uncertainty tracking. Each agent should post evidence, hypotheses, unresolved questions, and confidence notes to the same blackboard, and the other agents should then pick them up and refine them.
It is better to start with a narrow task at first. For example, in a task such as “integrate literature evidence and phase-stability information for 20 candidate compositions in a certain alloy family and rank them by priority,” it should be possible to check whether the blackboard reduces missing information and duplicate reasoning. After that, experimental notes or computational logs can be connected and the workflow can be expanded step by step to see whether role division among agents is actually helpful.