Peer-to-peer (P2P) systems are distributed systems based on the concept of resource sharing by direct exchange between peer nodes (i.e., nodes having same role and responsibility). Complex Adaptive Systems (CAS) can be a new programming paradigm for P2P applications. In the CAS framework, a system consists of a large number of relatively simple autonomous computing units, or agents. From a P2P perspective, CAS offers several attractive properties, including total lack of centralized control. In this paper we present a load balancing framework which effectively balances the workloads of the jobs distributed among interconnected nests with the help of information carrying autonomous agents called ANTs. The ants helps to effectively balance the loads as it wanders via the interconnection network to find a pair of under loaded and over loaded nests (collection of nodes). The load transfer from overloaded to underloaded nests is performed by direct downloading between the two nests thereby avoiding large amount of data transfer across the network. The algorithm developed improves the response time of the user submitted jobs and the overall execution time required for the completion of the submitted jobs is found to decrease. It is found that the agents wander about randomly when the load is uniformly distributed among the interconnected nests and they move rapidly towards the regions of the network with highly imbalanced loads.