Abstract. Natural forming multi-agent systems have the ability to grow to enormous sizes and perform seemingly complex tasks without the existence of any centralized control. Their success comes from the fact that agents are simple and the interaction with the environment and neighboring agents is local in nature. In this paper we describe how swarm intelligence can be used in the implementation of a Linda-based system called SwarmLinda. It is argued that SwarmLinda achieves many desired characteristics such as scalability, adaptability and even some level of fault-tolerance. This paper describes our current implementation and discusses approaches for a full-fledged SwarmLinda implementation.