A computer-implemented method is provided for distributed computing. The method includes forming interaction specific shadow agents by extracting agent interaction data from an original agent at a source node utilizing agent-based simulators. The method further includes broadcasting aggregated agent interaction data collected from the original agent at a source node to other nodes in a distributed area to calculate agent interactions and behaviors therefrom. The agent interaction data extracted from the original agent consists of a constrained subset of data derived from a set of data of the original agent at the source node.