We propose a methodological approach for tackling the early design stages of self-organising Multiagent Systems (MASs). We adopt an architectural pattern based on the Agents and Artefacts (A&A) metamodel: self-organisation mechanisms are added to an existing environment of artefacts by embedding them into environmental agents. We rely on a three-stage design approach with modelling, simulation and tuning, so as to identify a suitable design of environmental agents and their interaction with artefacts. The main objective is to design a MAS environment providing services that self-organise in response to the unpredictable dynamics of the agents exploiting them.

As a case study, we analyse the problem called collective sorting, a service for decentralised sorting of items in MAS environments that was inspired by social insects' behaviour: the proposed solution features environmental agents and tuple spaces, whose design choices and evaluation have been driven by formal simulations.