In this paper, we elaborate on the notion of artifact for MAS (multi-agent system), aimed at providing a unifying abstraction to model and engineer the environment of agents. This notion, inspired by the studies on Activity Theory, generalises over existing infrastructures for MAS environments, and over previous work on coordination artifacts and agent coordination contexts. The artifact abstraction is used to represent at the agent cognitive level those tools, services, components, objects and entities that constitute and shape the agent environment.
In its most general acceptation, an artifact conceptually exposes (i) a usage interface, (ii) operating instructions, and (iii) a service description, as well as other relevant properties such as inspectability, malleability, formalisability and linkability. By enabling rational agents to reason about the artifact possible uses and effects, these features provide the agent with a well-engineered model of its environment, enabling the cognitive exploitation of its services toward the achievement of individual and social goals.
After discussing rational exploitation of artifacts by agents, we shift to the MAS engineer's viewpoint. On the one hand, we show how the notion of artifact can be used as a conceptual tool to classify, catalogue and compare a number of different agent-related abstractions. On the other hand, we suggest how it could potentially provide the basic foundation bricks of a principled methodology for the engineering of MAS environments. 

2nd International Workshop "Environments for Multi-Agent Systems" (E4MAS 2005), pages 62-77, 26 July 2005.
Danny Weyns, H. Van Dyke Parunak, Fabien Michel (eds.), AAMAS 2005, Utrecht, The Netherlands.