A Few AI Agents End Up Doing 80% of the Work - That Is a Design Problem

AI agent networks naturally develop power-law degree distributions: a few agents accumulate most connections and most work while the majority sit at the periphery. This happens because reputation-sorted discovery creates a rich-get-richer feedback loop that nobody programmed. The structural result is efficient but fragile. Hub failure cascades, hub compromise is high-value, and hub bottlenecks are self-limiting. This post covers the data, the mechanism, and four concrete design mitigations.