The concept of agent and of object were compared and contrasted, to explore their differences as units of software modularisation [19]. An agent-oriented approach provides a semantically richer framework for specifying distributed, heterogeneous information systems.
One way in which agents generalise objects is that their behaviour can evolve. Whereas usually the behaviour of objects is fixed at specification time, a generalisation proposed in [20] allows the behaviour of objects to change during runtime of a system in a certain restricted way. The ability to change behaviour can (for example) be used to model a change of goals for an agent.
The Lisbon group, in collaboration with IC (London), worked on the formalisation of reuse techniques based on an object/agent-based organisation of software systems. We formalised different levels of modularity in the development of reactive systems as categories -- specifications as theories, process models, programs -- together with relationships between them as functors, showing how reusability is a property of the way two levels of modularity fit together, and how verification techniques can be defined that take into account these relationships [3,8,2,1,4].
The Namur group has studied the use of agents in the modelling of requirements. Two agent-oriented formalism were compared and coupled: the Albert language was used to specify requirements, in terms of states and actions, and information and perception. The i* framework was used to understand and redesign organizational processes, in terms of strategic relationships and rationales. In a paper [9], a small banking example is used to illustrate how the requirements process may iterate between the two levels of modelling and analysis towards a requirements specification.
In conjunction with SG2, operational semantics for TROLL with emphasis on object interaction was developed [8,9]. A calculus of objects for designing information systems was also developed, and it is planned to generalise this to agents [2].
Work on using agents in the modelling of the software process has also continued. The IC London group has explored the consequences of decentralised development within the ViewPoints framework. In this context we have investigated a mechanism for decentralised process modelling and enactment. Our approach provides process guidance by analysing individual ViewPoint work records and recognising significant `situations' upon which to act. We have implemented the mechanism in the Objectworks/Smalltalk. In our prototype, ViewPoint situations are represented as regular expressions which are matched against pre-specified rules/patterns [6,7]
We have also continued exploring techniques for inconsistency management in evolving specifications and have made progress towards constructing a framework for doing so. We have examined the use of inconsistency to provide more focused inconsistency handling actions, particularly in the context of requirements elicitation. We plan to work with SG1 towards developing the formal underpinnings of our approach [5].