Agglomerative Hierarchical Clustering (AHC) is a classical explanatory approach allowing to organize instances of a database into a binary tree whose nodes express different levels of generality. However, the topology of a binary tree is not always semantically optimal. In many contexts, it would be more relevant to create a shallow Directed Acyclic Graph (DAG) instead of a tree, thus allowing to express overlaps. In this paper, we propose a new AHC algorithm which is based on the parallel use of multiple agents (aggregation, merging, sharing, ...) to build such DAG.