An Agent-mediated Approach
Virginia Dignum
Department of Information & Computing Sciences, Utrecht University, The Netherlands
Knowledge Management, Organizational Performance, Multi-Agent Systems, Agent-based Simulation.
Organizational effectiveness depends on many factors, including excellence, effective planning and capability
to understand and match context requirements. Moreover, organizational performance cannot be just eval-
uated in economic or other global terms, but it must consider values of the participating agents (people or
groups), such as individual satisfaction. Different organizational structures are clearly better matched to cer-
tain problems and context requirements than others, but evaluation methods are mostly lacking. In this paper,
we will present ongoing work on tools and formalisms to model organizations and evaluate their performance
according to global and individual values, under different circumstances.
Today, collaboration is the common place for organi-
zational activity. Both within as across organizations,
collaborative work happens for product development,
marketing, sales, R&D, etc. For instance, in the field
of product development several companies collabo-
rate nowadays to develop new products, e.g., cars, air-
crafts or machinery, resulting in what is referred to as
a collaborative supply chain. Collaborative develop-
ment efforts are usually conducted on a project basis,
i.e., centered around a new product or service. The
duration of such collaborative development projects
can vary from some months to several years.
Several challenges arise in such collaborative
projects. On the one hand, efficient organization
structures need be implemented that facilitate the per-
formance of project teams. Typically, such teams con-
sist of employees from the different groups involved
in the collaboration, possibly augmented with exter-
nal consultants. This means that members of the team
fulfil a specific role on the project, but are also part,
and must answer to, the original organizational struc-
tures of their own companies and groups. Further-
more, from a knowledge management perspective,
these collaborative teams depend on knowledge that
is spread over different companies that are part of the
chain. That is, the team will need to tap in to knowl-
edge from other people in the different partner orga-
nizations. Managing this knowledge effectively con-
tributes to the performance of the supply chain as well
as the individual companies.
Effectiveness of organizations and teams depends
on many factors, including excellence, effective plan-
ning and capability to understand and match context
requirements. Despite a large number of studies, the
effect of organizational structure on the performance
and the knowledge sharing capabilities of a group
is still not well understood. Different organizational
structures are clearly better matched to certain prob-
lems and context requirements than others, but evalu-
ation methods are mostly lacking.Furthermore, envi-
ronments are not static, which implies that structuring
decisions must be adaptable to change and therefore
the process of determine congruence between struc-
ture and environment is cyclic.
Moreover, organizational performance cannot be
just evaluated in economic or other global terms, but it
must consider values of the participating agents (peo-
ple or groups), such as individual satisfaction. Intel-
ligent agents have been defined as autonomous, reac-
tive, proactive entities capable of social interactions in
dynamic environments. As such, the agent paradigm
is particularly suitable to model project teams. In or-
der to cope with the inherent complexity of knowl-
edge sharing in teams, the concept of Agent-mediated
Knowledge Management (AMKM) proposes agent-
based approaches to deal with collective aspects of
the domain in an attempt to cope with the conflict be-
tween desired order and actual behavior in dynamic
Dignum V. (2007).
In Proceedings of the Ninth International Conference on Enterprise Information Systems, pages 504-509
environments (van Elst et al., 2004).
In this paper, we will present ongoing work on
tools and formalisms to model organizations and eval-
uate their performance under different circumstances.
The aims of this project are to generate tools and
methods to understand and model the relation be-
tween organizational design, agent behavior and team
performance. Results will be applied to the design
of more effective agent systems that support the ac-
tivity of virtual groups. The paper is organized as
follows: in section 2 describes our work on organi-
zational models, section 3 discusses the factors that
influence performance and how to dynamically adapt
these factors in order to improve performance. In sec-
tion 4 we introduce our current work on the develop-
ment of simulations that enable to study the perfor-
mance model proposed. Finally, section 5 presents
our conclusions and directions for future work.
The agent paradigm offers an effective way to model
and analyze complex systems composed of multiple
and distinct components. In this sense, an organiza-
tion can be seen as a set of agents whose interactions
are regulated by mechanisms of social order and are
created to achieve common goals. Furthermore, vir-
tual teams as described in the previous section, op-
erate in open environments, where autonomous par-
ticipants and stakeholders are not centrally organized
and follow heterogeneous models and motivations,
and act according to their own plans and norms.
Such open environments demand organizational mod-
els that integrate the realization of organizational re-
quirements and objectives, and at the same time al-
low participants to have the freedom to act according
to their own agendas and goals. This means that we
should take a distributed view on Knowledge Man-
agement (KM): knowledge is autonomously managed
where it is created and used, namely within each com-
munity or project team. Moreover, autonomy without
coordination is almost useless for knowledge sharing
in complex organizations, yet coordination should be
reached through interoperation rather than centraliza-
tion (Bonifacio et al., 2001). Recently, several agent-
oriented analysis methodologies for Knowledge Man-
agement have been proposed (van Elst et al., 2004;
Dignum et al., 2004).
Our approach prescribes such an agent-oriented
framework, which considers individual and social
goals in order to determine which structure is the best
applicable for a project team, given the task environ-
ment, the knowledge needs, and the organizational
constraints of its members. The framework is de-
scribed in more detail in the remainder of this section
2.1 Organizational Structure
The OperA model for agent organizations, (Dignum
et al., 2004), enables the specification of organiza-
tional requirements and objectives, and at the same
time allows participants to have the freedom to act ac-
cording to their own capabilities and demands. OperA
considers agent organization models as having at least
two description levels. At the abstract level, which
can be seen as a receipt for collective activity, orga-
nizations are described in terms of roles, their depen-
dencies and groups, interactions and global norms and
communication requirements. The concrete level is a
possible instantiation of the abstract organization, by
populating it with real agents that play the roles and
realize interactions (Vazquez-Salceda et al., 2005),
(Sichman et al., 2005). Organizational design starts
from the identification of business strategy, stakehold-
ers, their relationships, goals and requirements and re-
sults in a comprehensive (agent) organization model
including organizational roles, interactions and plan-
ning rules, that fulfil the requirements set by the busi-
ness strategy. Organizational instantiation is the pro-
cess that accepts an abstract organization model and a
set of agents, and resources and generates a concrete
organization by assigning responsibilities and organi-
zational goals to each agent.
The OperA framework consists of three interre-
lated models. The organizational structure of the so-
ciety, as intended by the organizational stakeholders,
is described in the Organizational Model (OM). The
OM specifies an agent organization in terms of four
structures: social, interaction, normative and commu-
nicative. The social structure specifies objectives of
the system, its roles and the model that governs co-
ordination. The interaction structure gives a partial
order of the scene scripts that specify the intended in-
teractions between roles. Interaction scene scripts are
flexible constructs that describe how a result should
be achieved, instead of using procedures to describe
what are the steps to follow. Society norms and reg-
ulations are specified in the normative structure, ex-
pressed in terms of role and interaction norms. Fi-
nally, the communicative structure specifies the on-
tologies for description of domain concepts and com-
munication illocutions. The way interaction occurs
in an organization depends on the aims and charac-
teristics of the application, and determines the way
roles are related to each other, and how role goals and
norms are ’passed’ between related roles. For exam-
ple, in a hierarchical organization, goals of a parent
role are shared with its children by delegation, while
in a market organization, different participants bid to
the realization of a goal of another role.
The Social Model (SM) describe how agents can
enact roles in the organizations. That is, the SM spec-
ifies the interaction scenes that describe the possi-
bilities for negotiation of role enactment by agents
joining the organization. These scenes generate so-
cial contracts for the participating agents that describe
the capabilities and responsibilities of an agent within
the organization, that is the desired way that an agent
will fulfil its role(s). The use of contracts to describe
the activity of the system allows on the one hand for
flexibility in the balance between organizational aims
and agent desires and on the other hand for verifica-
tion of the outcome of the system. Finally, given an
agent population for an organization, the Interaction
Model (IM) specifies possible interaction protocols
between agents that implement the functionalities de-
scribed in scene scripts in the OM.
A generic methodology to analyze a given do-
main and determine the type and structure of an ap-
plication domain resulting in a OperA agent organi-
zation model is described in (Dignum et al., 2004).
The methodology provides generic facilitation and in-
teraction frameworks for agent societies that imple-
ment the functionality derived from the co-ordination
model applicable to the problem domain. Standard
organization types such as market, hierarchy and net-
work, can be used as starting point for development
and can be extended where needed and determine the
basic norms and facilitation roles necessary for the so-
ciety. A brief summary of the methodology is given
in table 1.
In order to design and analyze OperA models for
organizations, we are currently developing an graph-
ical tool, OperettA. The tool enables the verification
of the OM, but it also generates simulations of the
organization, to be populated with agents so that its
activity can be animated.
2.2 Agents in Organizations
Section 2.1 described the modelling of organizations
from the perspective of the organization’s designer.
As such, organizational models describe the goals, re-
quirements and expectations of the organization itself.
Obviously, individual agents will have their own mo-
tivations and expectations when joining a certain or-
ganization. This implies than more than a need for
frameworks that specify the organization’s structure
and goals, we need to specify mechanisms through
which prospective participants can evaluate the char-
acteristics and objectives of society and role goals,
in order to decide about participation. Furthermore,
tools for individual agents to adapt their architecture
and functionality to the requirements of an assumed
role must be provided (Dastani et al., 2003).
In the context of project teams, organizational
roles indicate the capabilities of agents that are to ful-
fil the role, typically describing the minimum expec-
tations on the enactment of participating agents. The
success, that is, improved performance, of the team
is dependent on the effort agents will put into their
role performance. As in human organizations, the
idea here is that a job description only describes part
of one’s input into the organization (as well as one’s
reward). What distinguishes employees is how well
they put themselves to their job. That is, an agent,
representing a member of one of the companies par-
ticipating in the project, is faced with a decision about
how to best enact its teams role in a way that ideally
benefits both the project team as its original company.
Many issues influence this decision, including trust,
strategic reasoning and cultural characteristics.
The considerations above indicate the need to
match agent and role objectives and functionality. At
the moment most approaches to this problem sim-
ply design agents from scratch so that its behavior
complies with the behavior described by the role(s)
it will take up in the society. This applies especially
to closed MAS where all agents are designed cen-
trally and with a pre-intended purpose. In this case
the design of the agent follows from the requirements
specified for the role(s) the agent is fulfilling. Com-
prehensive solutions for this problem require complex
agents that are able to reason about their own objec-
tives and desires and thus decide and negotiate their
participation in a society. A first step on the road to
this solution (taken in (Dastani et al., 2003)) is to have
a formalism to compare the specifications of agents
and roles and determine whether an agent can enact a
role. In the future, agents themselves should be able
to use this mechanism to automatically evaluate their
participation on a society.
Furthermore, once a decision has been reached
that an agent will indeed enact a role, there must be
ways to modify that agent in order to include the char-
acteristics of the assumed role. A possible solution for
this point has been proposed in (Esteva et al., 2001) in
which agents are extended with an interface to the so-
ciety. This interface prevents any action not allowed
by the role definition. However, it does not facilitate
proactive behavior expected from the agents while
playing the role. Therefore, our proposal allows for
different approaches to such modification of an agent
which result in different role performances, by pro-
viding the for negotiation of role enactment for spe-
Table 1: Overview of OperA methodology.
Step Description Result
Identifies organization’s
main characteristics:
purpose, relation forms
Stakeholders, facilitation roles,
coordination requirements
Analysis of expected
external behavior of
operational roles, use cases,
normative requirements
Design of internal
behavior of system
Role structure, interaction
structure, norms, roles, scripts
Design of enactment
negotiation protocols
Agent entrance scripts,
Role enactment contracts
Design of interaction
negotiation protocols
Scene script protocols,
Interaction contracts
cific agents in the SM. For instance, some agents will
uniquely attempt to achieve the goals of its adopted
role and forget its own private goals, while others will
only attempt to achieve the goals from the role after
its own goals have been satisfied.
2.3 Organizational Environment
Organizations do not exist in a vacuum. Each orga-
nization is set in a particular environment to which
it is inextricably linked. This environment provides
multiple contexts that affect the organization and its
performance, what it produces, and how it operates.
The “rules of the game” of a society are one of the
most important ingredients of the enabling environ-
ment. All societies require appropriate rules, as well
as fair and efficient mechanisms by which they can
be enforced. Organizations must pursue their goals
within a normative structure that facilitates or inhibits
their work. North defines rules as “... the formal laws
and codes that positively or negatively influence the
behavior of organizations through the incentives and
constraints they provide or impose. (North, 1990).
There are rules for all dimensions of the environment
be it at administrative, technological, political, eco-
nomic, socio-cultural, or stakeholder’s level. Rules
can be formal or informal, explicit or implicit. In ad-
dition to norms, teams and organizations possess a
certain combination of resources that influences the
type and scale of activities undertaken by the group,
as well as how successful their efforts are likely to be.
Capabilities of a group include natural resources, hu-
man resources, financial resources, infrastructure and
technology. Together with norms, these capabilities
create an enabling or inhibiting environment for orga-
nizations (Lusthaus, 2002).
Furthermore, given dynamic environments, teams,
and therefore, their members, should have a range of
capabilities and the ability to decide how best to apply
those capabilities to changing situations. This implies
that,ideally, team members should have a sense of re-
sponsibility towards the global objectives of the team
(which in our model we represent as social contracts
to role enactment), and the ability to assess a situa-
tion and modify its behavior to maintain or improve
overall team performance.
Compelling as this aim may be, such agents are
usually quite expensive and not always available for
any project team
. Therefore, it is important to de-
sign the structure of teams in a way that adaptation is
possible without requiring all participants to have this
type of functionality and ability.
Efficiency of organizations is usually measured in
terms of performance, or the degree to which goals
are achieved. Contingency theory shows that: (a)
there are no one optimal organizational design, and
(b) structural constraints, task constraints, cognitive
constraints and other environment conditions influ-
ence organizational outcomes. In human settings, or-
ganizational performance has been demonstrated em-
pirically to be associated with the degree of congru-
ence (or ’fit’) between organizational structure and
properties of the task or environment (Donaldson,
2001). Accordingly, it is to an organizations advan-
tage to monitor the fit between its structure and mis-
Note that this observation applies both to human agents
in project teams as to artificial agents in multi-agent sys-
tems. People with wide capabilities and ability to quickly
assess a situation and determine the best course of action
to take are often high educated, have a wide and long ex-
perience and therefore are expensive. Software agents with
similar possibilities are, ifat all existing extremely complex,
and therefore computationally expensive.
sion, and to alter its structure when a misfit is identi-
fied. There is empirical evidence that high performing
organizations can discern when environmental forces
have changed the state of congruence (i.e., the good-
ness of fit), thus driving changes in the strategies (e.g.,
communication patterns, back-up behaviors) that they
employ (Entin and Serfaty, 1999). In the next section,
we will discuss in more detail the factors that influ-
ence performance.
From the discussion in the previous sections, it
becomes evident that organizational performance is
a complex issue that depends on many different fac-
tors. Performance, or efficiency, is affected by the
organization’s design, the cognitive, by information
processing capabilities of the agents, the operating
conditions, the task and the task environment faced
by the organization (Carley and Svoboda, 1996). Re-
organization can then be seen as a means to achieve
better performance, or creating a more efficient orga-
nization, by changing one or more of the factors that
affect performance. From the analysis of many or-
ganizational simulation projects, some of which will
be discussed in the next sections, and inspired by So
and Durfee (So and Durfee, 1998), we classify the
factors that determine performance into three broad
classes. The first are the structural factors, which are
the components and features of the organization (such
as roles, dependencies, constraints, norms and regu-
lations). The second are task environmental factors,
which are the components and features of the task
(such as size, time constraints, uncertainty). The third
class of factors are agent factors, which are the char-
acteristics of the individual agents concerning task ca-
pability, knowledge (including decision making and
reasoning capabilities), social awareness, etc. These
three classes of factors jointly determine the perfor-
mance of the organization.
This suggests a 3d-space in which the results of
different organizational simulations can be placed and
eventually compared. Axes in this 3-space correspond
to the groups of factors listed above. Each point in
the space represents a specific organization instance
(that is, with a given structure, task, agent capabilities
and operating conditions). For each point, the perfor-
mance can be calculated. In this perspective, reorga-
nization means a move into another point in the space
which has a better performance. In order to cope with
complexity, we are developing simulations that based
on two dimensional projections of the overall space,
which means that one set of factors is kept fixed in that
particular simulation. In the following we introduce
our work on organizational simulations along these
Both in social sciences as in computer science many
simulation studies have been realized that aim to in-
crease the understanding of reorganization in human
and agent organizations respectively. Different or-
ganizational simulation projects have studied the in-
fluence of different factors on the organizational effi-
ciency, using different techniques and aiming at prov-
ing different hypothesis.
So far, we have discussed models for the specifi-
cation of organizations that enable the representation
of organizational requirements, in terms of structures
and objectives, independently from the specification
of the individual agents. Such models enable the an-
alyze of sharing and interaction in organizational set-
tings. The novelty of the approach is that the same
model enables analyze at both the level of organiza-
tional performance as well as at the level of the indi-
vidual expectations. In this section, we will discuss
our current research towards simulation frameworks
for such models.
Currently, several agent-based simulation lan-
guages and platforms are available (e.g. Repast
By following an agent-oriented approach, these plat-
forms enable the observation and analysis of (com-
plex) social processes between the entities. However,
most of these simulations are based on very simple
models of agency, that hardly enable the specifica-
tion of the cognitive aspects of the agents. On the
other hand, languages and tools for agent develop-
ment, such as 3APL, Jason or Jack, that are based on a
cognitive notion of agency (typically the BDI model)
often lack the constructs to represent high level social
aspects, such as roles, norms or organizational struc-
tures. Moreover, from a computational point of view,
these agent platforms do not provide the scheduling
and timing mechanisms required to run simulations.
A few languages attempt to counter these problems.
A good example is Brahms (Sierhuis et al., 2003), or
the MAS-SOC platform (Bordini et al., 2005).
Simulation of organizational or group behavior
requires more than the organizational model and its
members, as developed in the OperA model. The sim-
ulation must be able to incorporate the environment
conditions under which activity will take place. Sim-
ulation tools are well suitable to describe and mod-
ify environment conditions by enabling for instance to
vary task complexity and frequency, or the harshness
or the environment. Our approach is use an existing
simulation platform (Repast) and create a plugin to
cognitive agents developed in an agent-oriented lan-
guage (for the moment we are using 3APL
. The aim
is to eventually be able to integrate in one platform
agents developed using different languages so that
different approaches to cognition can be integrated.
Our current research on the development of a sim-
ulation tool for reorganization experimentation aims
at the identification of conditions and requirements
for change, ways to dynamically incorporate changes
in systems, how to determine when and what change
is needed, and how to communicate about changes.
In (Dignum et al., 2006) we presented a simulation
scenario, VILLA, to evaluate different reorganization
forms and understand triggers for reorganization.We
are currently working on a simulation that analyzes
the relation between agent capabilities and organiza-
tional structure, when the objective is to share knowl-
edge on a particular task.
In this paper, we argued that effectiveness of project
teams is dependent on the structure and norms of the
team, on the capabilities and values of participants,
on the way the team is able to consider and relate to
the external links of its members, and on the over-
all characteristics of the environment where the team
exists. This means that team performance cannot be
evaluated just on economic terms but must consider
the values and expectations of participants. We have
presented a model for organization design that de-
scribes requirements concerning the overall goals and
norms of the organization or team, but also takes in
account the autonomy of individual participants. In
order to understand relation between structure and
performance of teams we are developing simulations
tools that enable the controlled variation of factors
and their effects. Our current and future work is cen-
tered around the further development of simulation
and evaluation tools.
This research is funded by the Netherlands Organi-
zation for Scientific Research (NWO), through Veni-
grant 639.021.509. The author is grateful to Christi-
aan Tick for his work on the development of the sim-
ulation environment.
Bonifacio, M., Bouquet, P., and Traverso, P. (2001). En-
abling distributed knowledge management: Manage-
rial and technological implications. Upgrade, III(1).
Bordini, R. H., Costa, A. C. d. R., H
ubner, J. F., Moreira,
l. F., Okuyama, F. Y., and Vieira, R. (2005). MAS-
SOC: a social simulation platform based on agent-
oriented programming. Journal of Artificial Societies
and Social Simulation, 8(3).
Carley, K. and Svoboda, D. (1996). Modeling organiza-
tional adaptation as a simulated annealing process. So-
ciological Methods & Research, 25(1):138–168.
Dastani, M., Dignum, V., and Dignum, F. (2003). Role as-
signment in open agent societies. In AAMAS03. ACM
Dignum, V., Dignum, F., and Meyer, J. (2004). An agent-
mediated approach to the support of knowledge shar-
ing in organizations. Knowledge Engineering Review,
Dignum, V., Dignum, F., and Sonenberg, L. (2006). Ex-
ploring congruence between organizational structure
and task performance: A simulation approach. In Co-
ordination, Organization, Institutions and Norms in
Agent Systems I, volume 3913 of LNAI, pages 213–
230. Springer.
Donaldson, L. (2001). The Contingency Theory of Organi-
zations. Sage.
Entin, E. E. and Serfaty, D. (1999). Adaptive team coordi-
nation. Journal of Human Factors, 41:321–325.
Esteva, M., Padget, J., and Sierra, C. (2001). Formalizing
a language for institutions and norms. In ATAL-2001,
LNAI 2333, pages 348–366. Springer.
Lusthaus, C. (2002). Organizational Assessment: A Frame-
work for Improving Performance. IDRC.
North, D. (1990). Institutions, Institutional Change and
Economic Performance. Cambridge University Press.
Sichman, J., Dignum, V., and Castelfranchi, C. (2005).
Agent organizations. JBCS, 11(3).
Sierhuis, M., Clancey, W., and van Hoof, R. (2003).
Brahms: a multiagent modeling environment for sim-
ulating social phenomena. In SIMSOC VI.
So, Y. and Durfee, E. (1998). Designing organizations for
computational agents. In Simulating Organizations,
pages 47–64.
van Elst, L., Dignum, V., and Abecker, A., editors (2004).
Agent-Mediated Knowledge Management: Selected
Papers, volume 2926 of LNAI. Springer.
Vazquez-Salceda, J., Dignum, V., and Dignum, F. (2005).
Organizing multiagent systems. JAAMAS, 11(3):307–