; Cutter-Exec_Rpt-July_2007
Documents
Resources
Learning Center
Upload
Plans & pricing Sign in
Sign Out
Your Federal Quarterly Tax Payments are due April 15th Get Help Now >>

Cutter-Exec_Rpt-July_2007

VIEWS: 4 PAGES: 27

  • pg 1
									Enterprise Architecture

    Vol. 10, No. 3


                          Agent Technology:
                          What Is It and Why Do We Care?
                          by James Odell, Senior Consultant,
                          Cutter Consortium


                          Agent technology is now necessary to reduce costs; to improve

                          efficiency and effectiveness; and to support the requirements

                          of individuals, groups, companies, and universities as they

                          collaborate globally. More importantly, this type of technology

                          enables us to create and support a whole class of IT applications

                          and approaches that we previously could not have developed.

                          In this Executive Report, we discuss what agent technology

                          is as well as how and where it is currently being used in

                          various industries.
Cutter Business Technology Council

Rob Austin    Tom DeMarco   Christine Davis   Lynne Ellyn   Jim Highsmith   Tim Lister   Lou Mazzucchelli   Ken Orr   Sheleen Quish   Ed Yourdon




             Access
              to the                                About Cutter Consortium
             Experts                                Cutter Consortium is a unique IT advisory firm, comprising a group of more than
                                                    150 internationally recognized experts who have come together to offer content,
                                                    consulting and training to our clients. These experts are committed to delivering
                                                    top-level, critical, and objective advice. They have done, and are doing, ground-
                                                    breaking work in organizations worldwide, helping companies deal with issues in
                                                    the core areas of software development and agile project management, enterprise
                                                    architecture, business technology trends and strategies, enterprise risk management,
                                                    business intelligence, metrics, and sourcing.

                                                    Cutter delivers what no other IT research firm can: We give you Access to the
                                                    Experts. You get practitioners’ points of view, derived from hands-on experience with
                                                    the same critical issues you are facing, not the perspective of a desk-bound analyst
                                                    who can only make predictions and observations on what’s happening in the
                                                    marketplace. With Cutter Consortium, you get the best practices and lessons learned
                                                    from the world’s leading experts, experts who are implementing these techniques
                                                    at companies like yours right now.

                                                    Cutter’s clients are able to tap into its expertise in a variety of formats including
                                                    print and online advisory services and journals, mentoring, workshops, training,
                                                    and consulting. And by customizing our information products and training/consulting
                                                    services, you get the solutions you need, while staying within your budget.

                                                    Cutter Consortium’s philosophy is that there is no single right solution for all
                                                    enterprises, or all departments within one enterprise, or even all projects within a
                                                    department. Cutter believes that the complexity of the business technology issues
                                                    confronting corporations today demands multiple detailed perspectives from which a
                                                    company can view its opportunities and risks in order to make the right strategic and
                                                    tactical decisions. The simplistic pronouncements other analyst firms make do not
                                                    take into account the unique situation of each organization. This is another reason to
                                                    present the several sides to each issue: to enable clients to determine the course of
                                                    action that best fits their unique situation.

                                                    For more information, contact Cutter Consortium at +1 781 648 8700 or
                                                    sales@cutter.com.
                 Agent Technology:
                 What Is It and Why Do We Care?
                 ENTERPRISE ARCHITECTURE
                 ADVISORY SERVICE
                 Executive Report, Vol. 10, No. 3




by James Odell, Senior Consultant, Cutter Consortium

Centralizing a corporation was         enough to thrive in new markets       The biggest breakthrough with
once considered an efficient way       will be left behind.                  agents is that they are an evolu-
to run an enterprise. Decisions                                              tion of existing technologies. What
and information processing             Agent technology is a primary         is revolutionary is the way we
occurred in an orderly, top-down,      enabler to support this new era.      think about and use agents to
hierarchical manner. However, we       In fact, without it, our current      design IT systems. They are being
are no longer in the era of main-      technology will not scale to sup-     built from today’s technology and
frame computing, when both             port the ever-increasing global       will work together with today’s
companies and applications were        interaction. In response, many        technology. While agents, objects,
typically command-and-control-         companies are now building            relational databases, legacy
oriented and organized in vertical     agent-based systems. These sys-       systems, service-oriented archi-
silos. With the combination of the     tems employ agents that can dis-      tectures (SOAs), event-driven
Internet, fiber optics, and PCs, the   tribute functionality across a vast   approaches, and so on, each
business and technology playing        computing network. Furthermore,       have their own niche, together
field has been flattened. No longer    agents not only adapt to their        they can orchestrate rich systems
primarily top-down, it has shifted     environment but also evolve by        that none of these technologies
more to a side-by-side approach        learning from the environment.        could provide alone.
— as individuals, small groups,        In short, they are the ultimate in
and organizations interact around      distributed computing. Such an        In this Executive Report, we first
the world. Furthermore, globaliza-     approach prepares enterprises         discuss how agent technologies
tion and changes in technology         for an increasingly complex mar-      — including ant-based systems —
are causing today’s market to          ketplace and enables them to          are improving operations in dif-
be in a state of constant flux.        respond rapidly to change.            ferent industry sectors. We then
Companies that cannot adapt fast                                             explore agent technology in more
   2                                                                                  ENTERPRISE ARCHITECTURE ADVISORY SERVICE


detail by defining it, examining its                 phenomena (e.g., ant colonies,                  1. Wander randomly.
internal architecture and proper-                    traffic jams, stock markets, and
                                                                                                     2. If food is found, take a piece
ties, and understanding how vari-                    prairie ecosystems). Similarly,
                                                                                                        back to the colony and leave
ous agent-based systems adapt                        automated entities — called
                                                                                                        a trail of pheromones, which
and interact within environments.                    agents — can enable supply chain
                                                                                                        evaporates over time; then go
                                                     systems, planning and scheduling
                                                                                                        back to rule one.
WHY SHOULD WE CARE                                   applications, Web services, and so
ABOUT AGENT TECHNOLOGY?                              on. Such agents need not be com-                3. If a pheromone trail is found,
First, An Example from                               plicated. For example, the ant                     follow it to the food and then
Everyday Life                                        colony simulation of StarLogo1                     go back to rule two.
                                                     (see Figure 1) uses software
Living creatures and plants can                                                                      In Figure 1a, the “ant” agents
                                                     agents, where each ant follows
produce complex everyday                                                                             are just emerging from the anthill
                                                     three simple rules:
                                                                                                     to begin their random walk.
                                                                                                     Eventually, an ant discovers a
                          a
                                                                                                     food source and returns some to
                                                                          b
                                                                                                     the colony, leaving a trail of evap-
                                                                                                     orative pheromones, as shown in
                                                                                                     Figure 1b. Figure 1c shows the ant
                                                                                                     colony well underway in retrieving
                                                                                                     the food. Lastly, Figure 1d depicts
                                                                                                     two depleted food sources and
                                                                                                     one that is exhausted altogether.

                                                                                                     Ant-Like Agents at British Telecom
                                                                                                     and HP Labs

                           c                                              d                          The ant colony is organized
                                                                                                     without an organizer, coordinated
                                                                                                     without a coordinator. As it turns
                                                                                                     out, this simple ant foraging tech-
                                                                                                     nique provides an inspiration to
                                                                                                     tackling some difficult technologi-
                                                                                                     cal problems. While the insect’s
                                                                                                     totally random approach to
                                                                                                     locating food may appear to be
                                                                                                     inefficient, the study of this tech-
                                                                                                     nique has led to a new area
    Figure 1 — Snapshots from an ant colony simulation using software agents.                        of research known as “swarm
                                                                                                     intelligence” [2].

1StarLogo software is available for free at http://education.mit.edu/starlogo.


The Enterprise Architecture Advisory Service Executive Report is published by Cutter Consortium, 37 Broadway, Suite 1, Arlington, MA
02474-5552, USA. Tel: +1 781 641 9876 or, within North America, +1 800 492 1650; Fax: +1 781 648 1950 or, within North America, +1 800 888 1816;
E-mail: service@cutter.com; Web site: www.cutter.com. Group Publisher: Kara Letourneau, E-mail: kletourneau@cutter.com. Managing Editor:
Cindy Swain, E-mail: cswain@cutter.com. ISSN: 1530-3462. ©2007 by Cutter Consortium. All rights reserved. Unauthorized reproduction in any
form, including photocopying, faxing, image scanning, and downloading electronic copies, is against the law. For information about reprints and/or
back issues of Cutter Consortium publications, call +1 781 648 8700 or e-mail service@cutter.com.


VOL. 10, NO. 3                                                                                                                   www.cutter.com
EXECUTIVE REPORT                                                                                             3


For example, a decade ago,             network is large, many possible       Other Uses of Ant-Based Systems
British Telecom (BT) used swarm        routes exist for each connection.
                                                                             In ant-based systems, an ant
intelligence for inspiration to        Using a centrally controlled rout-
                                                                             colony of a finite size searches
make software systems simpler,         ing approach, however, scales
                                                                             collectively for a good solution to
cheaper, and more intelligent and      badly and can lead to failure of
                                                                             a given optimization problem.
robust. At the time, Chris Winter,     the entire system. A decentralized
                                                                             The optimal solution can only be
then head of BT Future Systems,        mechanism like the ant-based
                                                                             found through the global coopera-
was concerned that the trend of        approach was found to both scale
                                                                             tion of all the colony’s ants, where
IT systems was toward central          well with the network size and
                                                                             the ants only communicate indi-
control. “Ants walk around at ran-     avoid the possibility of a system-
                                                                             rectly by adding pheromones to
dom and there’s no intelligent         wide failure.
                                                                             the environment. The BT applica-
controller telling them what to
                                                                             tions above are only two exam-
do,” said Winter. “When they           The approach here is to send
                                                                             ples of where techniques start
find food, they simply run back to     ant-like agents periodically
                                                                             with models of ant behavior and
the nest laying a little pheromone     between nodes. At each node,
                                                                             then add things that are not pre-
trail that other ants can smell. The   the agent updates the node’s
                                                                             sent in the real world. However,
rest of the colony need only fol-      routing table with information
                                                                             there are many more [2, 8], some
low that path to fetch the remain-     (“pheromone”) indicating how
                                                                             of which include:
ing food” [3]. In other words,         long the journey took from its
the processing is distributed, not     origin and which nodes the agent        Traveling salesman problem.
centrally controlled.                  used along the way. The routing         A salesman must find the short-
                                       table contains a list of the node’s     est route by which to visit a
Winter used ants initially to          immediate neighbors and proba-          given number of cities, each
weed out system bugs. Today,           bilities associated with using that     city exactly once.
in BT’s development network,           neighbor as the next step on the
little ant-like programs run around    journey toward a target node in         Quadratic assignment prob-
finding problems. They leave           the network. The fastest ants will      lem. This involves the problem
pheromone-like time stamps that        have a positive effect on the prob-     of assigning facilities to loca-
highlight the fault and draw help      ability scores of the nodes they        tions so that the costs of the
from other software. In other          used, while slow ants will have a       assignment are minimized.
words, BT is using the ants as a       negative effect.                        Job-shop scheduling prob-
first step in self-healing systems                                             lem. For a set of machines and
software. Many software compa-         Network congestion is another           a set of jobs, operations must
nies are now embedding ant-like        problem that was addressed by           be assigned to time intervals
agents in systems software and                    .
                                       BT and HP They considered con-          in such a way that (1) no two
middleware for similar reasons.        gestion in a particular section of      jobs are processed at the same
                                       the network to be analogous to          time on the same machine,
In another application, BT and         the depletion of a food source          and (2) the maximum of the
HP Labs created the world’s first      near an ant colony. Here, ant-like      completion times of all opera-
ant-based systems [9]. Both tele-      agents would search for new             tions is minimized.
phone networks and the Internet        routes and dynamically update
typically route connections            the virtual pheromone trail             Vehicle routing problem. This
through a number of intermediate       recorded in routing tables.             involves finding minimum cost
switching stations. When the                                                   vehicle routes where (1) every




©2007 CUTTER CONSORTIUM                                                                        VOL. 10, NO. 3
  4                                                              ENTERPRISE ARCHITECTURE ADVISORY SERVICE


  customer is visited exactly        a supply chain system perform in         equipped with a simple software
  once by exactly one vehicle;       a similar manner?                        agent that was programmed to
  (2) for every vehicle, the total                                            keep its booth busy and bid on
  demand does not exceed the         Painting Trucks at General Motors        each paint job. The amount of
  vehicle capacity; (3) the total    Traditionally, assembly line sched-      the booth’s bid was based on
  tour length of each vehicle        ules are centrally developed and         how busy the booth was at the
  does not exceed a given limit;     controlled. Any change in the            moment of bidding, whether it
  and (4) every vehicle starts       schedule must be centrally recon-        had to change to a different paint,
  and ends its tour at the same      figured. When the line is small          and whether the booth was func-
  position (the depot).              and has few unplanned stop-              tioning properly.
                                     pages, centrally controlled
Beyond Ants                                                                   To coordinate the various bids
                                     schedules work well. However,
Individual ants are not very                                                  for each paint job, a scheduler
                                     scheduling for most real-world
sophisticated insects. They have                                              agent acts as a broker. For exam-
                                     assembly lines can be a night-
limited memory and a largely ran-                                             ple, when a truck arrives to be
                                     mare: work stations break down,
dom element to their behavior.                                                painted, the scheduler agent tells
                                     personnel get sick, environmental
However, acting as a collective,                                              the booths, “I have a truck that
                                     conditions are not always within
ants can perform complex tasks                                                needs to be painted red.” A vacant
                                     acceptable limits, products com-
with reliability and consistency.                                             paint booth already loaded with
                                     ing down the line have or acquire
Ant colonies are not the only                                                 red paint will bid very high.
                                     unexpected defects, and so on.
things that work like this. Bee-                                              However, a vacant booth with a
hives, flocks of birds, freeway      Dick Morley, a technology vision-        different color would bid lower
traffic, national and global         ary and father of the programma-         because of the extra labor and
economies, societies, and            ble controller, swept away old           time to clean and reload the paint
immune systems are all exam-         assembly line schedules and              gun. A booth that has just started
ples of patterns that are deter-     developed a better system for            to paint a truck, has broken down,
mined predominantly by local         painting trucks at GM’s assembly         or is otherwise less suited for the
component interaction instead        plant in Fort Wayne, Indiana.            job would bid even lower. Based
of centralized authority. For IT     “How do I schedule the non-              on the outcome of the bidding
applications, this can include       schedulable?” Morley wondered.           activity, the scheduler assigns the
order processing, supply chain,      “Trucks do not come down the             truck to the highest-bidding paint
shop-floor control, inventory        line in order of their color, and fre-   booth.
management, message routing,         quently no paint booth is available
                                                                              In a top-down planned “push-
and management of multiple           with the correct color.” Morley
                                                                              through” world, if one booth mal-
databases. In other words, a         also discovered that many of the
                                                                              functioned, a centrally controlled
decentralized approach should        paint booths were typically broken
                                                                              system would require immediate
be considered where local com-       down or being repaired.
                                                                              recomputing. With bottom-up,
ponents also have control —
                                     In his technique, the scheduling         “pull-through” paint booth agents,
instead of limiting system-design
                                     program interacts with each              other booths were ready to pick up
approaches solely to the cen-
                                     paint booth. Instead of assigning        the bidding slack at a moment’s
trally organized one traditionally
                                     unpainted trucks to booths, GM’s         notice. This new design saved $1
employed by IT. After all, if New
                                     solution was to have the booths          million dollars in nine months and
York City can maintain a two-
                                     bid on the paint jobs [4]. To            reduced the lines of computer
week supply of food with only
                                     accomplish this, each booth was          code from hundreds to four [4].
locally made decisions, why can’t


VOL. 10, NO. 3                                                                                    www.cutter.com
EXECUTIVE REPORT                                                                                                                               5


Morley and GM tackled a problem                   can no longer scale using a                        are usually centrally managed.
where centralized scheduling did                  centralized approach. Organiza-                    By using agents, however, the
not work efficiently by adopting an               tions like DHL and Credit Suisse                   process and resources can be
agent-based approach where each                   have found that an agent-based                     managed in a distributed manner.
booth acts on its own behalf using                approach is their only option                      For example, an Install Pole agent
a market-based bidding system.                    for successfully managing the                      is a specialized entity that knows
Even though the scheduler was a                   previously unmanageable size                       its Install Pole process needs a
centralized element, it deferred to               and complexity within their                        utility pole, pole-installer equip-
distributed booth agents. Agent-                  businesses.                                        ment, a person to operate the
based solutions do not remove                                                                        pole installer, and the prior com-
centralization; instead, they try to              At the heart of the agent-based                    pletion of a Dig Pole Hole process.
balance it with distributed solu-                 solution, two primary forms of
tions wherever it makes sense.                    agents were used to gain distrib-                  Such knowledge can be used
                                                  uted control:2                                     in several ways. For instance, a
Dynamic Scheduling                                                                                   requester might be interested in an
                                                  1. Process-based agents. These                     Install Pole service but would like
Current market trends are driving                    have the knowledge of how to                    to know its cost before the service
organizations from mass produc-                      combine resources and create                    is invoked (see Figure 3). Here, a
tion (where the supplier tells the                   products as part of a workflow                  call for proposal (CFP) would be
customer what to buy) to mass                        in a supply chain.                              sent to an agent that provides such
customization (where the cus-
                                                  2. Resource-based agents.                          a service.3 The agent would send
tomer tells the supplier what to
                                                     These manage the capacity-                      a CFP to those resources and pre-
provide). For small supplier orga-
                                                     constrained resources of the                    ceding services required for the
nizations where resources and
                                                     systems, such as people, vehi-                  Install Pole process. Each resource
requirements are reasonably sta-
                                                     cles, tools, machinery, materi-                 agent would then calculate the
ble, this does not present a prob-
                                                     als, and facilities.                            charge for its usage based on the
lem. However, in larger and more
                                                                                                     time and duration specified by the
complex organizations, supporting
                                                  Business Processes and                             requester. If a resource is urgently
centrally managed operations is
                                                  Resources Using Agents                             needed and is in short supply, the
more difficult. This is particularly
                                                                                                     cost might be high; if not, the price
true of large suppliers with volatile             Figure 2 contains an example
                                                                                                     would be low. For example, if the
and demanding conditions that                     of a workflow diagram for a
                                                                                                     requester wants a utility pole in
include unscheduled resource                      process that installs utility poles
                                                                                                     two hours and only an expensive
failures, periodic surges in new                  for an electric company. The
                                                                                                     provider has one available, the
orders, and changes in require-                   round-cornered rectangles repre-
                                                                                                     price would be higher than if the
ments and priorities. Using agents                sent processes, and the square-
                                                                                                     requester could wait for two days
here can change the perspective                   cornered rectangles are those
                                                                                                     and obtain it from a cheaper sup-
to a more distributed approach,                   resources required by the
                                                                                                     plier. In contrast, the agent repre-
making the solution more scal-                    processes.
                                                                                                     senting pole installers might be
able, adaptable, and robust. In
                                                  In business process management                     able to provide a person imme-
fact, many organizations have
                                                  (BPM) systems, such activities                     diately. In two days, however,
already realized their limit and

2Several other types of agents were used as well (e.g., order agents, dispatch agents, and supervisor agents). However, the process and resource
agents are key to understanding the approach.
3Agents are a useful way to extend SOAs.




©2007 CUTTER CONSORTIUM                                                                                                      VOL. 10, NO. 3
  6                                                                                      ENTERPRISE ARCHITECTURE ADVISORY SERVICE



                                                                           Transformer                Install
                                                                                                   Transformer




                                                                                                Cable
                                               Utility
                                               Pole                                                                             Test


                       Dig Pole Hole                            Install Pole                      Pull Cable


                                                                                                                              Tester

                 Pole Hole        Pole Hole                Pole         Installer             Cable        Cable
                  Digger           Worker                Installer       Worker               Puller       Worker


      Figure 2 — Example of a business process where processes and resources are managed using agents. (Source: [7].)


                               CFP for
                             Dig Pole Hole
                                                                                                       sequence diagram depicted in
  Dig Pole Hole                                                                CFP for                 Figure 4. The requester is play-
                     CFP for Utility Pole           Install Pole           Install Pole                ing the role of customer and the
       Utility                                                                                         provider is playing the role of
       Pole                    CFP for                                        CFP for                  supplier. Interaction protocols are
                             Pole Installer                              Installer Operator
                              Equipment                                       Worker                   useful because they define the
                                                                                                       expected behavior between inter-
                                                                      Installer
                                  Pole Installer                                                       acting agents — whether either
                                                                      Operator
                                                                                                       can be a resource agent or a
       Figure 3 — A call for proposal (CFP) to the Install Pole agent results in                       process agent.4
        a CFP to all of its resources and preceding operations. (Source: [7].)
                                                                                                       Agent Renegotiation

                                                                                                       In the real world, suppliers can
they could all be busy on other                   In other words, the Install Pole
                                                                                                       overcommit, have resource
assignments but would do the                      agent acts like a broker on behalf
                                                                                                       failures, and experience process
requested job at double-time rates.               of the process. If the requester
                                                                                                       delays. Using a top-down, central-
                                                  accepts the bid, the broker is
Once the resources have                                                                                ized approach, the schedule
                                                  notified of the award and will con-
responded to the CFP with a                                                                            changes tend to be reoptimized
                                                  firm (or decline) the award.
bid, the Install Pole agent tallies                                                                    from a global level. Such a
up the cost for the service as a                                                                       technique would work in a small
                                                  Agent Negotiation
whole and sends a consolidated                                                                         operation but would not scale to
bid to the requester for the Install              The interaction protocol just                        a large one. However, using an
Pole service.                                     described is expressed using the                     agent-based approach — which is


4It should be noted that the notion of requester and provider agents are at the very heart of the W3C’s Web Services Architecture study [11].




VOL. 10, NO. 3                                                                                                                    www.cutter.com
EXECUTIVE REPORT                                                                                                                                         7


distributed by nature — enables
more dynamic scheduling by
                                                         Supplier                                                          Customer
adapting to these unexpected situ-                                                  Call for Proposal
ations on an individual and local
basis, rather than a massive and                                                        Bid/(no bid)
global one.                                                                       Award/(no award)

For example, an important job                                                       Confirm/(decline)
request is received for Resource                                                    Proceed/(abort)
A, which is already allocated to
Job 1 and Job 2. The agent repre-
senting Resource A needs to inter-                     Figure 4 — An interaction protocol for making a contract.
act with the process agents for
each job to determine whether
the extra work can be accommo-
dated. The agent for Resource A
asks the agent for the first step in
                                              Job 1 could be delayed by four hours.
Job 1 about any extra (or slack),             I wonder what my customer (Process                          I can handle a four-hour delay without
time (see Figure 5). It turns out             Step 2) thinks about that?                                  affecting my customer. Therefore, I
                                                                                                          won't charge for such delay.
that if Resource A takes on the
new job, it will delay Step 1 by         Job 1
four hours. Now, Step 1’s agent               Process Step 1
                                                                       Process Step 2
can ask Step 2 if a four-hour delay            Resource A                                                Process Step 3
would affect its processing. Since                                        Resource B                                                  Process Step 4
                                                                                                            Resource C
Step 2 has lots of slack time, it can                                                                                                   Resource D
absorb the delay without affecting
the overall schedule.                   Figure 5 — Renegotiation example for Job 1 by process and resource agents. The
                                        bar represents the total time allocated to perform the process; the segment within
Resource A’s agent must now              the bar indicates the actual amount of time needed by the process. (Source: [1].)
examine the remaining option
of changing Job 2 (see Figure 6).
Step 1 would be delayed by three
hours. The Step 1 agent then asks
Step 2 if a three-hour delay would         Job 2 could be delayed
impact its processing. The Step            by three hours. I wonder        If my supplier is late by three hours,
                                           what my customer will           I will be late by three hours. I wonder
2 agent indicates that such a              think about that?               what my customer will think about that?

slippage would also cause a
                                                                                         If my supplier is late by three
three-hour delay for it. Step 3                                                          hours, I will be late by three     This is high priority.
                                                                                                                            That will really cost you!
                                                                                         hours. I wonder what my
has a similar result and therefore           Job 2                                       customer will think about that?
its agent must ask the end cus-                    Process Step 1
                                                                      Process Step 2
tomer if this slippage presents a                 Resource A                              Process Step 3
                                                                       Resource C
problem. In this example, the                                                             Resource G
customer has a large penalty
clause built into the contract for                    Figure 6 — Renegotiation example for Job 2 by process and
late shipments. In summary, Job 2                                   resource agents. (Source: [1].)


©2007 CUTTER CONSORTIUM                                                                                                          VOL. 10, NO. 3
  8                                                             ENTERPRISE ARCHITECTURE ADVISORY SERVICE


would result in a penalty if modi-    Magenta Technology has devel-        just-in-time (JIT) products and
fied to handle the extra require-     oped an agent-based system           services is vital. SCA Packaging
ment for Resource A. However,         that dynamically schedules the       turned to an agent-based model-
adapting Job 1 would enable           deployment of ships and cargo.       ing solution that explored different
Resource A to be assigned to a        IAI is developing a generic          strategies for reducing inventory
new job without affecting the         scheduler that rapidly generates     levels without compromising its
overall schedule. All decisions in    tailored and optimized scheduling    delivery commitments. An agent-
this example were based on a          engines. Using their software, an    based simulation developed by
local optimization rather than a      end user who has only domain         Eurobios (www.eurobios.com)
global one — enabling decisions       expertise can define:                enabled the company to reduce
to be made in seconds instead of                                           inventory levels by 35% while
hours or days and reducing costs         Components (e.g., resources,      maintaining its delivery commit-
as well as time to market.               work center, parts, tasks/        ments [6].
                                         operation, and jobs) — are
                                         required for the scheduling       Insurance Claims Processing
Agents and Dynamic Scheduling
                                         application, using the generic
The agent-based approach                 scheduling engine builder         Acklin B.V. (www.acklin.nl)
described above was developed            (GenSEB) component libraries      developed an agent-based,
with support from the Rock                                                 international vehicle–claims
                                         Protocols (interactions/          processing system for three
Island Arsenal as part of the US
                                         constraints between compo-        European companies. The agent
Department of Defense’s Defense
                                         nents) — use a standardized       system ensures EU regulations
Advanced Research Projects
                                         representation language           for confidentiality and enables a
Agency (DARPA) contract to
                                         called an agent interaction       robustness that can survive
build an agent-based, factory-
                                         protocol (AIP)                    broader system shutdowns and
scheduling prototype, named
Autonomous Agents for Rock               Rules/policies — define the       failures. As a result, the total time
Island Arsenal. In one benchmark         order of interactions and the     for client and claim identification
test using these techniques, inven-      content of an interaction based   is reduced from six months to
tory costs were cut by 47%, lead         on a user-defined scheduling      two minutes [6].
times were cut 59%, and sched-           algorithm
ule-reducible costs (such as over-                                         Distribution Optimization
                                      Other Agent-Based Uses
time and inventory holding                                                 Delivery of goods and services
charges) were cut by 93% [4].         Some Common Applications             between site and customer is
                                                                           a problem for many industries.
Since that time, agent-based          Many companies have already          Companies such as BT, DHL,
schedulers have been designed         tested agent-based applications in   and Air Liquide (www.airliquide.
for suppliers in many industry        their research and development       com)require transportation net-
sectors: manufacturing, finance,      facilities. Successful applica-      works that can flexibly adapt
energy, and transportation.           tions are now used commercially      in real time to changing and
NewVectors LLC (http://www.           throughout the world. Here is a      unforeseen run-time conditions,
newvectors.net), Magenta              brief overview of some of them.      fluctuations, and requirements.
Technology (www.magenta-                                                   Not surprisingly, at least three
technology.com), and Intelligent      Just-In-Time Delivery
                                                                           companies provide agent-
Automation, Inc. — IAI (www.                                               based solutions to this area:
                                      When operating in a highly
i-a-i. com/view.asp?tid=7) —                                               IAI, NuTech Software Solutions
                                      competitive market, delivering
have all been active in this area.

VOL. 10, NO. 3                                                                                    www.cutter.com
EXECUTIVE REPORT                                                                                               9


(www.nutech.com), and                exception-handling process in           complexity. Agent-based
Whitestein Technologies              order to provide better support to      simulation spans many areas.
(www.whitestein.com).                business users. Agent technology,       For example, it can aid social
                                     by its unique exception-handling        structures and institutions in
Onsite Maintenance                   capability, provides the solution to    developing plausible explana-
                                     this problem.                           tions of observed phenomena;
Any industry that employs JIT
                                                                             assisting in designing organiza-
processes relies on the availabil-
                                     Healthcare                              tional structures; and informing
ity of manufacturing equipment.
                                                                             policy or managerial decisions.
Faulty equipment can have cata-      By modeling the stakeholders
                                                                             Agent-based simulation can help
strophic consequences on pro-        in primary care systems as
                                                                             model physical systems, including
ductivity. In this context, both     agents, Calico Jack (www.
                                                                             intelligent buildings, traffic sys-
preventive maintenance and cor-      calicojack.co.uk) has developed
                                                                             tems, biological populations,
rective maintenance become key       a way to integrate health systems
                                                                             and software systems of all types,
activities. In the agent system      for the Scottish Executive Health
                                                                             including e-commerce and infor-
developed by Oslo Software           Department.
                                                                             mation management systems [6].
(www.oslo-software.com), its
                                                                             The two most popular agent
automotive customer handled          Simulation
                                                                             simulators are Swarm (www.
150,000 pieces of equipment with
                                     Since multiagent systems involve        swarm.org) and CybelePro
two million maintenance opera-
                                     many interacting agents, predict-       (www.cybelepro.com). Recently,
tions per year. The automotive
                                     ing the whole system’s success          CybelePro was used to simulate
company was originally using
                                     or failure is difficult. Defining the   air transportation systems involv-
the maintenance module of a
                                     behavior of each agent is one           ing flights, airports, terminal areas,
well-known enterprise resource
                                     thing; knowing what will emerge         controllers, and airline operation
planning (ERP) software vendor.
                                     from their joint interaction is         centers with varying levels of
While the ERP package could
                                     another.                                fidelity. Here, an individual simula-
provide preventive-maintenance
                                                                             tion is comprised of tens of thou-
schedules, it could not handle the   Multiagent models can be used           sands of dynamically interacting
corrective-maintenance opera-        to simulate the behavior of com-        agents configured to represent a
tions. Here, team managers had to    plex computer systems, including        concept/scenario.
deal with exceptional corrective-    multiagent computer systems.
maintenance operations using the     Such simulation models can              Collaborative Decision Making
less practical Excel format. This    assist designers and developers         and Distributed Control
lead to canceling an average of      of complex application systems
1,000 maintenance operations                                                 As systems become increasingly
                                     and provide guidance to software
every month, increasing the                                                  decentralized and autonomous,
                                     engineers. Agent-based simulation
risk of equipment failure. The                                               technology needs to support col-
                                     is also an important technique
maintenance problem is complex                                               laborative, team-based decision
                                     that offers strong models for rep-
because many pieces of equip-                                                making and control. Agents can
                                     resenting complex and dynamic
ment are involved and the skills                                             work individually and socially
                                     real-world environments. Agent
required of the personnel are                                                in teams to aid and accomplish
                                     systems simulating real-world
varied. New technologies are                                                 this. In particular, IAI has devel-
                                     domains may provide answers
needed to automate the mainte-                                               oped applications for team for-
                                     to certain physical or social
nance process. These technol-                                                mation, cooperative sensing,
                                     problems that would be other-
ogies need to automate the                                                   tracking, monitoring, and traffic
                                     wise unobtainable due to their
                                                                             management.

©2007 CUTTER CONSORTIUM                                                                         VOL. 10, NO. 3
 10                                                               ENTERPRISE ARCHITECTURE ADVISORY SERVICE


Supply Chains and Logistics            have been adapted to use the           techniques used by agents could
                                       semantic Web language (OWL),           be used to enhance the level of
Today’s supply networks need
                                       the semantic Web rule language         automation of peers in popular
to be flexible, responsive, adap-
                                       (SWRL), and the services ontol-        applications. Since complex P2P
tive, and able to cope with the
                                       ogy (OWL-S) for the specification      applications are social in nature,
variability of demands. Tradi-
                                       of agent systems. In addition,         they require increasingly sophisti-
tional supply chain and logistics
                                       agents can be used to manage           cated approaches to trust and
optimization systems are not
                                       and maintain large distributed         reputation and to the application
designed to handle volatility and
                                       ontologies as well as perform data     of social norms, rules, and struc-
complexity or to function in a
                                       mining functions. Since ontologies     tures. Social simulation would
real-time environment. Using an
                                       are both syntactically and seman-      be particularly appropriate here
agent-based approach is now a
                                       tically defined, agents not only       in order to better understand the
common solution for many com-
                                       intelligently query, infer, and rea-   population dynamics of indepen-
panies. Software vendors utilizing
                                       son over ontologies, their behavior    dent agents [6].
this type of approach include
                                       can also be guided by an ontology.
Magenta Technology, IAI, and
                                       These are the so-called ontology-      Web Services and SOA
Cougaar Software, Inc. (www.
                                       driven approaches. Here, the
cougaarsoftware.com).                                                         The W3C Web Services Archi-
                                       agent code does not have to be
                                                                              tecture study recommends an
                                       modified and recompiled for
Autonomic Computing                                                           agent-based approach to using
                                       every change. Instead, the ontol-
                                                                              SOAs — both on the service
In today’s global networks, we         ogy can be changed, resulting in
                                                                              requester side and the service
need computer systems that             a dynamic change in the agents’
                                                                              provider side [11]. Agents act
regulate themselves much in the        behavior based on the modified
                                                                              with varying levels of autonomy,
same way our autonomic nervous         ontology.
                                                                              depending on environmental
system regulates and protects
                                                                              constraints and their ongoing
our bodies. Self-healing systems       Peer-to-Peer
                                                                              interactions. Because services
require an agent-based approach
                                       Peer-to-peer (P2P) applications        are often best modeled as
rather than a centralized one.
                                       display agent-like characteristics     autonomous and heterogeneous,
                                       that include both applying             they can naturally be associated
Architectures and Ontologies —
                                       self-organization techniques that      with agents. Agents are not a
and Agents
                                       ensure continuous operation of         panacea. However, when applied
In addition to specific applica-       the network and employing inter-       appropriately, they enable us to
tions, the IT developer must           action protocol designs to enforce     (1) define elaborate constraints
address current architectural          correct behavior among interact-       on what services are willing to
approaches. Here is a brief            ing nodes. For example, commer-        offer, thereby enabling us to
overview of those approaches           cial e-marketplace systems like        define richer requirements for
particularly appropriate for agents.   eBay include simple credit reputa-     service composition; (2) discover
                                       tion systems to reward socially        trustworthy services; (3) negotiate
Ontologies                             beneficial behavior. As P2P            with external service providers;
Agents require a common vocab-         systems become more complex,           and (4) evaluate the compliance
ulary and set of concepts to com-      increasing the use of agent tech-      of service providers within their
municate effectively with other        nologies will be appropriate.          contracts. They can also provide
agents. Agent-system developers        For example, the auction-based         a more natural, human-like
                                       mechanisms and negotiation             way of interacting with service



VOL. 10, NO. 3                                                                                       www.cutter.com
EXECUTIVE REPORT                                                                                               11


requesters, as well as ensuring       across and embedded within mul-
the entire process is carried out     tiple services. However, multiple               Service
effectively and efficiently [10].     business processes can be used
                                      to satisfy the service (see Figure
Verizon chose a decentralized         7). For example, to request a
agent-based approach to SOA           product’s price, the computation
rather than a centralized approach    could vary depending upon
where all services are handled        location of order, time and date
by a server or set of servers.        ordered, quantity, customer, and
Additionally, Verizon employs         other items in the same order.
                                                                                     Figure 7 — Services can be
agents to support the subscription,
                                                                                   carried out by many different
management, and dashboard             The DHL pricing system, for                 business processes depending on
layers of its SOA, using its IT       example, had nearly 500 different            business conditions and rules.
Workbench software. By the            processes for pricing an order
end of 2004, Verizon was handling     because of the number and poten-
nearly three million service trans-   tial combinations of variables. To       various operating conditions
actions per day.                      handle this, DHL used agents to          and require frequent and timely
                                      choose the appropriate processes.        changes, agents should be used.
Service composition aggregates        In other words, each agent con-
services to create new functional-    tained a process-execution engine        Event-Driven Architectures
ity. Often, the composed function-    that would select and execute            Real-time response is often criti-
ality would itself be exposed as      the appropriate process for each         cal to customer satisfaction.
a new service with a standard         pricing-query service request. If a      Abnormal events or a combina-
interface. If this happens, a new     problem arose in the middle of a         tion of events can have a signifi-
service could be composed that        process, the agent would select an       cant impact on an organization.
would intelligently guide the         alternate one. DHL found that an         More and more companies,
service requester throughout the      agent-based approach not only            such as IAI and Rhysome (www.
lifetime of a particular business     executed faster, but was easier to       rhysome.com), are using agents
process. Such an intelligent ser-     maintain. Changes to a process or        along with software and physical
vice could be implemented as an       process rule (such as adding a fuel      sensors to detect changes in the
agent. Agents also make it possi-     surcharge) took hours rather than        business environment. (Examples
ble to capture the interactions       weeks. In a situation like this, soft-   include RFID tags for retail supply
among services and to create new      ware from companies like OSLO            chain optimization, medical
services as subtle compositions       can be employed, which uses a            monitors, physical sensors that
of others.                            graphic editor based on UML activ-       detect changes in air quality, and
                                      ity diagrams — enabling a user to        electronic data capture tools for
BPM                                   change the process. Again, agents        patient trials at pharmaceutical
BPM has come to mean an               are not always appropriate for           companies.) Companies that
integrated collection of critical     controlling business-process logic.      derive the greatest advantages
process technology necessary to       If the process is simple and             from agent-based, event-driven
support the business process life-    straightforward and it does not          architectures (EDAs) have the fol-
cycle (also called the BPM suite).    require frequent changes and pro-        lowing characteristics: (1) large
The actual business process logic     cessing robustness, a language           and heterogeneous environments,
can be centralized in one location,   like BPEL could be used. However,        (2) information that changes
as opposed to being distributed       for complex processes that involve       constantly in various ways,


©2007 CUTTER CONSORTIUM                                                                            VOL. 10, NO. 3
 12                                                               ENTERPRISE ARCHITECTURE ADVISORY SERVICE


(3) complex exceptions and state         Higher productivity                  mixture of technologies: object-
changes in real time, and (4) the                                             orientated (OO), relational, and
                                         Leverage of existing
need to deliver and respond                                                   agent-based. OO and relational
                                         infrastructure
appropriately to that information.                                            technologies enable a top-down
                                         Reuse of processes and               and centralized solution to busi-
Integrating SOA, BPM, and EDA            services                             ness application. Agents provide
                                                                              one more tool that conventional
These last three architectural           Foundation for future projects
                                                                              IT shops do not have — a
approaches can be considered
                                         Reduction of time to market          bottom-up and distributed app-
separately for including agent
                                                                              roach. The real benefit comes
technology. Additionally, agent          Increased agility to respond
                                                                              when an organization can choose
technology can be used to link           to business needs
                                                                              the appropriate mix of technolo-
business processes and services
                                       Is an agent-based approach useful      gies for a given application, which
to facilitate a single architecture
                                       for every application and usage?       provides a balance of both the
that integrates SOA, BPM, and EDA
                                       Does it always provide the bene-       centralized and distributed
(see Figure 8).
                                       fits listed above? Certainly not. If   approaches.
Why Use Agent-Based Technology?        your business is predictable and
                                       stable and if your processes are       It is one thing to say that agents
Thus far, we’ve examined a sam-                                               can provide a helpful, distributive
                                       centralized and scalable for the
pling of uses for agent-based                                                 approach to software applications;
                                       foreseeable future, adopting an
systems — happening now within                                                it is quite another to understand
                                       agent-based approach is not
the IT community. Given that any                                              what that means and entails. The
                                       necessary. However, for those
new technology is disruptive,                                                 next section begins that journey
                                       applications that must support
why did these early adopter com-                                              by answering the question “What
                                       complexity and change in a
panies decide to use agent tech-                                              is an agent?”
                                       scalable and timely manner,
nology? The reasons can be
                                       agents will likely be a necessary
summarized by one or more of                                                  WHAT IS AN AGENT?
                                       technology.
the following benefits:
                                                                              Conventional objects can be
  Faster ROI                           In the typical organization, both      thought of as passive, because
                                       of these situations probably coex-     they wait for a message before
  Lower maintenance                    ist in one form or another. Savvy      performing an operation. Once
                                       organizations will respond with a      invoked, they execute their
                                                                              method and go back to “sleep”
                                                                              until the next message. A current
                                Business Process                              trend in many systems is to design
      Event-Driven                                                            objects that both react to events
       Architecture              Orchestration               Business         in their environment and are
             (EDA)                                           Process          proactive. In UML 2.0, these are
                                      Agents                 Management
                                                                              known as active objects; in the
                                                             (BPM)
  Service-Oriented                                                            agent community, they are known
       Architecture              Choreography
                                                                              as agents. Whether they are
             (SOA)
                                     Services                                 called active objects or agents,
                                                                              this new direction is going to
                                                                              change radically how we design
       Figure 8 — Agents can be used to integrate SOA, BPM, and EDA.          systems.

VOL. 10, NO. 3                                                                                    www.cutter.com
EXECUTIVE REPORT                                                                                               13


The Basic Properties of Agents                           without any
                                       gathering of foodInteractivity         Ants provide a good example
                                       external choreography.                 of simple reactive adaptation.
An agent can be a person, a
                                                                              However, agents can also be
machine, a piece of software, or
                                       Agents are interactive entities        designed to learn and evolve.
a variety of other things. The basic
                                       because they are capable of
dictionary definition of agent is                                                       multiagent society
                                       exchanging rich forms of mes-          Overall, agents can be auton-
“something that acts.” However,
                                       sages with other entities in their     omous, interactive, and adaptive
for developing business and IT
                                       environment. These messages            to some degree (see Figure 9). It
systems, such a definition is
                                       can support requests for services      is not an simple agent proposi-
                                                                                         all-or-nothing
too general. While an industry-                                                         interaction
                                       and other kinds of resources, as       tion. In the next few sections,
standard definition of agent has
                                       well as event detection and notifi-                             properties
                                                                              these three key agentAdaptivity
not yet emerged, most agree that
                                       cation. They can be synchronous              more externally more detail.
                                                                              will be discussed in controlled
agents deployed for IT systems
                                       or asynchronous in nature. The
are not useful without the follow-                                             more Are Autonomous
                                                                              Agentsself-governing
                                       interaction can also be conversa-
ing three important properties:
                                       tional in nature reactive learning/
                                                         — negotiating        Agents can be thought of as
                                                        only      evolving
1. Autonomous. The agent is            contracts, marketplace-style           autonomous because each
   capable of acting without direct    bidding, or simply making a
                                         Autonomy                             is capable of governing its
   external intervention. It has       query. In the ant colony example,      own behavior to some extent.
   some degree of control over its     ants interact via the pheromones       Autonomy is best characterized
   internal state and actions based    that they deposit in the environ-      by degrees. At one extreme, an
   on its own experiences.             ment. The pheromones act as            agent could be completely self-
                                       information signposts for other        governing and self-contained.
2. Interactive. The agent commu-       agents, providing a simple yet         However, an agent that does not
   nicates with the environment        highly effective means of              require the resources of or inter-
   and other agents.                   communication.                         action with other entities is very
3. Adaptive. The agent is capable                                             rare. Even a database access
   of responding to other agents       Lastly, agents can be thought of
                                                                              or Web query agent requires
   and/or its environment. An          as adaptive because they can
                                                                              external resources. At the other
   agent can modify its behavior       react to messages and events
                                                                              extreme, an agent that is barely
   based on its experience.            and then respond appropriately.
                                                                              able to perform the simplest of
                                       In the example above, each
                                                                              actions is impractical for the prag-
Based on this approach, a basic        ant adapts to its environment by
                                                                              matic system developer. Even
working definition for an agent is     continuing to wander randomly if
                                                                              traditional objects have some
“an autonomous entity that can         it fails to find food or pheromones.
                                                                              degree of self-contained lines of
adapt to and interact with its         However, when pheromones
                                                                              code within their methods.
environment.”                          are detected, an ant reacts by
                                       changing its behavior to track the     Depending on the system devel-
Agents are commonly regarded           pheromones to the food source.         oper, an agent’s autonomy can
as autonomous entities because         Once the food is found, the ant        be designed to fit some place
they can be thought of as having       again adapts by picking up the         between these two extremes.
their own set of internal responsi-    food and carrying it back to the       For example, conventional objects
bilities and processing. For exam-     colony. If the food is moved, the      are not completely dependent
ple, each ant in Figure 1 has its      adaptive ants will locate the new      on outside resources; they have
own self-contained processing          food source and notify others          some state and behavior of their
that enables searching for and         while bringing it to the colony.       own (see Figure 10). But they are


©2007 CUTTER CONSORTIUM                                                                         VOL. 10, NO. 3
 14                                                                    ENTERPRISE ARCHITECTURE ADVISORY SERVICE


                    Interactivity                                                 entirely proactive. On the other
                                                                                  axis, an agent’s decision-making
                                                                                  aspect ranges from being limited
                                                                                  to simple decisions to being capa-
                                                  multiagent society
                                                                                  ble of making complex decisions
                                                                                  (see Figure 11).

                                                  simple agent
                                                  interaction                     Dynamic Autonomy

                                                                 Adaptivity         In addition to reacting to specific
                                               more externally controlled           method invocations, agents can
                                                                                    respond to events within their
                                           more self-governing
                                                                                    environment. Proactive agents will
                   reactive learning/
                   only        evolving                                             actually poll the environment for
                                                                                    events and other messages to
      Autonomy                                                                      determine what action they
                                                                                    should take. To compound this,
 Figure 9 — Agents have various degrees of autonomy, interactivity, and adaptivity. in multiagent systems, agents can
                                                                                    be engaged in multiple parallel
                               Supply Chain Manager                                 interactions with each other —
                    Objects                                      ?                  magnifying their dynamic nature.
                                   Agents         Agents
    Complete                                                          Complete In short, an agent can decide
 Dependence                                                          Autonomy when to say “go.”
                  higher                                lower
               dependency                            dependency                     As described earlier, the GM paint
             on other entities                    on other entities                 booths had both reactive and
                                                                                    proactive features. Information
                        Figure 10 — Degrees of autonomy.                            about an unpainted car or truck
                                                                                    coming down the line was posted
                                                                                    in an automated form accessible
traditionally simple, which means        involve a high degree of internal          to all paint booths. When a paint
that they are far from being com-        decision making though activities          booth had nearly completed its
pletely independent. Because sup- such as monitoring and delegat-                   current job, it basically said,
ply chain agents are able to reach ing, which depend on outside                     “Hmmm, I’m running out of work,
certain conclusions and make             resources.                                 I’ll look over at the jobs posted.”
decisions on their own, they have                                                   As stated earlier, if the booth
more autonomy than objects. By           Two major aspects of an agent’s            was applying the color of paint
their very nature, supply chain          autonomy involve its capacity to           required by an upcoming job, it
agents require interactions with         be dynamic and its ability to make would bid more for the job than
other entities to enable a fully         decisions. As depicted in Figure           would a booth having a different
functional supply chain, since           11, autonomy can be considered             color. Other bidding criteria could
one agent cannot take care of an         on two axes. On one axis, the              include how easy or important the
entire organization’s supply chain. dynamic aspect of an agent’s                    job was. In other words, the booth
More autonomous are manager              autonomy can range from being              was reactive in whether to bid on
agents because their role can            simply passive in its action to            a paint job, but proactive in that it


VOL. 10, NO. 3                                                                                           www.cutter.com
EXECUTIVE REPORT                                                                                                            15


would check the paint job list and                  Proactive
determine what to do.                                                    Clock

Objects, on the other hand, are                                                                  Supply Chain
conventionally passive — with




                                                      Dynamic Ability/
                                                                                                    Agent




                                                       Can Say “Go”
their methods being invoked
under a caller’s thread of control.
                                                                                                   GM Paint
The term “autonomy” barely                                                                          Booth
applies to an entity whose invoca-
tion depends solely on other com-                                         Java/UML                    Shopping
ponents in the system. However,                                            Object                      Agent
UML and Java have recently intro-                                                   Ant
duced event-listener frameworks                      Reactive
and other mechanisms for allow-                                            Object
                                                      Passive
ing objects to be more active. In
                                                                         Simple                               Complex
other words, objects now have                                            decisions
                                                                                          Decision Ability
                                                                                                              decisions
                                                                                           Can Say “No”
some of the dynamic capability
of agents.                                                           Figure 11 — Two aspects of autonomy.

Decision Autonomy
                                       complex criteria for selecting a                      the message and how to respond
Agents can be designed to make         gift for a specific individual. In                    if it does. In common practice,
a few simple decisions. However,       fact, the agent might return empty                    however, the refusal of an object
the more decisions an agent is         handed because it decided no                          to execute its method is typically
capable of making and the more         gifts were appropriate. A contract-                   considered an error situation.
complex they are, the more             negotiation agent may decide the                      With agents, this is not the case.
autonomous the agent appears.          contract is not worth pursuing for                    Manager and quality assurance
A completely autonomous agent          a number of reasons. In other                         agents are good examples of
will do whatever it wishes.            words, the agent can also say                         agents that can say “no.”
                                       “no” to performing a requested
For example, an ant that is wan-                                                             Object classes are usually
                                       service.
dering around looking for food                                                               designed to be highly predictable
can appear to be taking a random       Conventional objects certainly                        in order to facilitate buying and
walk. However, once pheromones         have the ability to make decisions.                   selling reusable components.
or food are detected, the choices      However, the typical usage and                        Agents are commonly designed
for its behavior are predeter-         direct support with OO languages                      to determine their behavior based
mined. In contrast, the GM paint       tend toward a less complicated                        on their individual goals and
booth can decide on how much           approach to decision making. For                      states as well their contexts within
to bid based on the booths ability     instance, when a message is sent                      ongoing conversations with other
to efficiently take on a new paint     to an object, the method is always                    agents. While OO implementa-
job. Certainly, the choices are lim-   invoked. The contract-negotiation                     tions can certainly be developed
ited for a paint booth, but the ant    agent could refuse to invoke a                        to include nondeterministic
has no choice on how to act            requested bid method; the object                      behavior, this is common in
when food is found. In contrast,       cannot. Yes, an object may deter-                     agent-based thinking.
shopping agents can have very          mine whether or not to process


©2007 CUTTER CONSORTIUM                                                                                           VOL. 10, NO. 3
 16                                                               ENTERPRISE ARCHITECTURE ADVISORY SERVICE


Agents Are Interactive                 While we can conceive of an            simple stimulus — to make a
                                       agent that cannot interact with        direct, predetermined response
Interaction involves the ability to
                                       anything outside of itself, the use-   to a particular event or environ-
communicate with the environ-
                                       fulness of such an entity for devel-   mental signal. Such a response
ment and other entities. It can
                                       oping agent-based systems is           is usually expressed in “if-then”
be expressed in degrees (see
                                       questionable.                          form. Thermostats, robotic sen-
Figure 12).
                                                                              sors, and simple search bots fall
                                       Agents Are Adaptive                    into this category.
On one end of the scale, object
messages (method invocation)           An agent is considered adaptive
                                                                              From atoms to ants, the reactive
can be seen as the most basic          if it can respond appropriately
                                                                              mode is quite evident. A carbon
form of interaction. A more com-       to other agents and/or its environ-
                                                                              atom has a rule that states in
plex degree of interaction would       ment to some degree. Auton-
                                                                              effect: “If I am alone, I will only
include those agents that can          omous entities that fail to adapt
                                                                              bond with oxygen atoms.” An
react to observable events within      rapidly to the changing world
                                                                              ant has a rule that if it finds food, it
the environment. For example,          become extinct: organisms die;
                                                                              should return the food to its colony
food-gathering ants don’t invoke       companies go out of business.
                                                                              while leaving a pheromone trail.
methods on each other; their           For an organization, adaptation
interaction is indirect, through       enables the system to react effec-
direct physical effects on the         tively to changes in areas such as     Rules
environment. Even more complex         the market and business environ-       Beyond the simple reactive agent
interactions are found in multi-       ment. In IT systems, adaptation        is the agent that can appear to
agent systems where agents             enables systems to react appro-        reason by following chains of
engage in multiple, parallel inter-    priately enabling, for example,        rules. For example, agents can
actions with other agents. Here,       system balancing, integrity assur-     react by making inferences and
agents begin to act as a society.      ance, and self-healing systems.        include patient diagnosis agents
                                       When designed properly, the indi-      and certain kinds of data mining
Finally, the ability to interact       vidual parts of the system can be      agents. These inferences are
becomes most complex                   empowered to change based              computed by following a chain
when systems involving many            on environment and market              of inference rules.
heterogeneous agents can coordi-       conditions.
nate through cooperative and/or                                               Business process engines can
competitive mechanisms (such           Simple Reactivity                      use a similar approach. Here,
as negotiation and planning).                                                 libraries of processes are main-
                                       At a minimum, this means that an
                                                                              tained, which indicate the
                                       agent must be able to react to a
                                                                              circumstances under which a
                                                                              particular process might be
                                                                              appropriate. Each step within a
                                                                              process acts as a service request,
                                                    Supply Chain              invoking the business process
             Objects        Ants                      Agents
                                                                              engine to choose the appropriate
  Method                                                       Society
Invocation                                                   Interaction      set of steps that should be exe-
                                                                              cuted, based on the business con-
                  simple                           complex
                                                                              text. For example, a request to
                                                                              compute the sales task would
                     Figure 12 — Degrees of interaction.


VOL. 10, NO. 3                                                                                       www.cutter.com
EXECUTIVE REPORT                                                                                            17


trigger the engine to locate the       niques that employ reinforcement       is affable, companionable, or
appropriate set of steps needed        learning include credit assign-        friendly)
for a given location. The taxes in     ment, Bayesian and classifier
                                                                              Mobile — able to transport
Michigan would be computed dif-        rules, and neural networks.
                                                                              itself from one environment
ferently than in Brussels.             Examples of learning agents
                                                                              to another
                                       would be agents that can approve
Rules do not change in and of          credit applications, analyze           Proxy — may act on behalf of
themselves. Instead, change can        speech, or recognize and track tar-    someone or something (i.e.,
come through other mechanisms          gets.                                  acting in the interest of, as a
such as learning and evolution.                                               representative of, or for the
Without learning and evolution,        Evolution                              benefit of some other entity)
ants and atoms are still quite able
to support complex “societies.”        Evolution is change that occurs        Intelligent — state is formal-
With learning and evolution, how-      over successive generations of         ized by knowledge (i.e., beliefs,
ever, the rules can be changed         agents. A primary technique for        goals, plans, assumptions)
based on experience — resulting        agent evolution usually involves       and interacts with other agents
in new and perhaps improved            genetic algorithms and genetic         using symbolic language
results.                               programming. Agents can literally      Rational — able to choose an
                                       be bred to fit specific purposes.      action based on internal goals
Learning
                                       For example, operation plans, cir-     and the knowledge that a par-
                                       cuitry, and software programs can      ticular action will bring it closer
Learning is change that occurs         prove to be more optimal than          to its goals
during the lifetime of an agent and    any product that a human can
can take many forms. The most          make in a reasonable amount            Temporally continuous — is a
common techniques enable rules         of time.                               continuously running process
and decisions to be weighted
                                                                              Credible — believable person-
based on positive (or negative)        Other Agent Properties
                                                                              ality and emotional state
reinforcement. For example, in a
                                       In earlier sections, three key agent
basic bidding system, a bid could                                             Transparent and accountable
                                       properties were discussed: auton-
be selected simply on the basis of                                            — must be transparent when
                                       omy, interactivity, and adaptivity.
bid price. However, other consid-                                             required, yet must provide a log
                                       These properties are important
erations might also be appropriate                                            of its activities upon demand
                                       because agents deployed for IT
such as the bidder’s ability to
                                       systems are not useful without         Coordinative — able to per-
deliver its goods in the quantity,
                                       them. However, agents may pos-         form some activity in a shared
quality, and timeframe requested.
                                       sess various combinations of           environment with other agents;
Over time, a purchasing agent
                                       other properties that may be           activities often coordinated via
can learn to choose from reliable
                                       useful — depending on the              plans, workflows, or some
vendor agents instead of just
                                       application requirements and           other process management
choosing the lowest bid. If a ven-
                                       the agent designer. Here, agents       mechanism
dor’s performance improves (or
                                       may be:
declines), the purchaser’s deci-                                              Cooperative — able to coordi-
sions are modified accordingly.          Sociable — interaction marked        nate with other agents to
In other words, the agent contin-        by friendliness or pleasant          achieve a common purpose;
ues to learn. Popular learning tech-     social relations (i.e., the agent    nonantagonistic agents that




©2007 CUTTER CONSORTIUM                                                                       VOL. 10, NO. 3
 18                                                                        ENTERPRISE ARCHITECTURE ADVISORY SERVICE


  succeed or fail together (also            accomplish this, agents must have       terrain, the laws of physics, and
  known as collaboration )                  some means by which they can            other agents. For software agents,
                                            perceive their environment and          the environment would consist of
  Competitive — able to coordi-
                                            act upon those perceptions (see         the hosting platform, its operating
  nate with other agents, but the
                                            Figure 13).                             systems and supporting middle-
  success of one implies the
                                                                                    ware, and other agents.
  failure of others (opposite of            Agent as a Black Box
  cooperative)                                                                      Input can be whatever the agent
                                            In its simplest form, each agent
  Rugged — able to deal with                                                        “perceives,” such as messages,
                                            can be thought of as an interactive
  errors and incomplete data                                                        requests, commands, and events.
                                            black-box process, where no
  robustly                                                                          Based on this input, the agent can
                                            indication of an agent’s internal
                                                                                    choose how to act on it, which
  Trustworthy — adheres to                  structure and behavior is given
                                                                                    typically results in output from the
  Laws of Robotics and is truthful          other than that which is visible
                                                                                    agent such as communicating
                                            from outside the agent. Figure 14
                                                                                    with, or making a change to, its
WHAT GOES ON INSIDE                         depicts this by illustrating that
AN AGENT?                                                                           environment. For instance, an
                                            agents have input from, and
                                                                                    Inventory agent could receive
As we discussed in the previous             output to, the environment. An
                                                                                    a message that requests the
section, agents are autonomous              agent’s environment can be
                                                                                    quantity on hand for a given com-
entities that can adapt to and              thought of as everything outside of
                                                                                    modity. The agent would then
interact with their environment. To         the agent. For robotic agents, the
                                                                                    determine the requested quan-
                                            environment would consist of the
                                                                                    tity and send a response to the
                                                                                    requester. The Inventory agent
                                                                                    could also be notified of the event
                                                                                    that some of the inventory has
                                  percept                                           been removed and is being sent
                                            sensors                                 to a customer. The agent process
                                    action                                          would then recompute the quan-
                                                                                    tity on hand and determine
                                      effectors                                     whether a reorder point has been
                                                                                    reached. If so, the Inventory agent
                   environment                        agent
                                                                                    would send out requests for bid to
                                                                                    the appropriate vendors.
            Figure 13 — Agents interact with their environments.
                                                                                    Agent Processing Overview

                                                                                    For an agent to interact with its
                                                                                    environment, it must be able
                           Environment
                         and other agents
                                                                                    to detect and effect its environ-
                                                                                    ment but also to understand
                                                                                    what it is detecting, to select an
                                                                                    appropriate response, and to act
     Input                       Agent                               Output         on it. In other words, for the agent
   (percepts)                                                 (initiated actions)
                                                                                    to be an interactive entity, it
                             Process
                                                                                    also requires some degree of
                 Figure 14 — An agent and its environment.                          autonomy and adaptivity. The


VOL. 10, NO. 3                                                                                           www.cutter.com
EXECUTIVE REPORT                                                                                                                   19


internal structure of an agent is      event-monitoring agents (in EDAs)                   object, the second specifies the
shown in Figure 15.                    are specialized to recognize                        requested operations, and the
                                       particular kinds of state changes                   remaining input contains the vari-
State                                  in the environment. For example,                    ables required for the operation
                                       database event agents could                         execution. Because the signature
Just like objects, agents can have     watch for certain kinds of changes                  syntax is already defined in this
a form of memory, referred to          in a database such as metamodel                     manner, the message does not
here simply as state. State            modifications, customer address                     need to go through the Determine
includes data that it needs to         changes, or the database going                      Action step. Instead, it goes imme-
operate effectively, which the         offline. If any of these events are                 diately to invoke its specified
agent maintains or can access in       detected, the event monitor could                   method.
some manner (such as via a data-       notify the appropriate agents that
base or automated blackboard).         can react appropriately to the                      In a system with a more complex
For example, an Inventory agent        event. In EDAs, this is particularly                style of input, the OO approach
would need to maintain informa-        useful. Robotic agents may                          cannot provide a satisfactory
tion, such as the reorder point for    employ such mechanisms as                           solution. For example, some
a particular product or the algo-      cameras, global positioning hard-                   agents need to process complex
rithm needed to compute the            ware, or infrared devices for                       syntaxes and receive their input
reorder point. The agent might         detectors.                                          in one or more languages, such
also maintain a list of valid prod-                                                        as SQL, Prolog, BPEL4WS, BPML,
uct vendors that it can contact for                                                        or ebXML. Furthermore, they
                                       Parse Input
reordering — or at least know                                                              may need to understand ontolo-
where the information might be         Once the detector receives input,                   gies expressed in RDF or OWL
found. The agent can even make         the agent needs to analyze and                      before any further action can be
inferences about its state or the      understand the input. This action                   determined. For event-monitoring
state of other entities in its envi-   typically involves recognizing the                  agents, this parsing can mean
ronment. State, then, can be           input, normalizing it, and dealing                  looking for patterns in the input.
metaphorically interpreted as          with any inconsistencies in the                     Event pattern recognition is impor-
anything that the agent can            content. For conventional OO                        tant for agents, especially for
“know” or “believe” about itself       software, there is virtually no pars-               those that participate in mecha-
or its environment.                    ing. In the OO “signature,” the                     nisms for self-healing systems.
                                       first parameter identifies the
Detectors

Detectors provide the interface to
the agent’s environment. They                                    Agent
                                                                                      State
receive input that may interest
                                                     Detectors




                                                                                                                     Effectors




the agent. Detectors can both                                        Parse          Determine           Invoke
                                                                     input           action             action
receive input sent to them or            Input                                                                                   Output
actively scan the environment
for it. For the more passive soft-                   recognize               interpret available data    determine
                                                                                                         mechanism
ware agents, this could include                      normalize               examine goals
                                                     representation                                      invoke chosen
receiving messages and event                                                 form plan                   actions
                                                     deal with
notifications. Active agents might                   inconsistencies
scan message boards or monitor
event occurrences. In particular,                       Figure 15 — The basic internal architecture of an agent.


©2007 CUTTER CONSORTIUM                                                                                          VOL. 10, NO. 3
 20                                                              ENTERPRISE ARCHITECTURE ADVISORY SERVICE


Granted, a parsing activity            the appropriate pricing algorithm     sent to them. The agent controls
requires greater complexity than       when the customer submits order.      its autonomy by using its own
the traditional object. In fact, not                                         resources or having them pro-
all agents will require sophisti-      Another function within the           vided by other agents. It deter-
cated parsers. Yet, when system        Determine Action step could be        mines what it can do and what it
requirements demand richer             to determine the appropriate          must delegate. In this sense, the
behavioral control at execution        goal for further consideration.       agent manages itself — rather
time, agents will be able to           For example, an Inventory agent       than just being managed.
accomplish what traditional            could detect each time a product
objects do not. Parsing extends        was pulled from a warehouse bin.      In addition, some agents can
the notion of the semantic net to      For each instance, the agent might    determine when their internal
enabling the semantic approach         need first to decide whether the      actions are not functioning as
to system interaction, in general.     product should be reordered or,       expected. The agent can also
                                       if it is moving too slowly, discon-   communicate to itself that the sit-
Determine Action
                                       tinued altogether. Each of these      uation needs to be remedied —
                                       goals requires a different process.   by sending itself a message and
As suggested earlier, traditional      One might involve sending out         letting the Determine Action step
OO recommends that the method          and processing bids from other        determine which action should be
requested of an object be deter-       agents to refill the bin; the other   taken next. In other words, the
mined before a message is sent to      might notify customer agents that     agent can have its own processing
the object, because each object        the product will no longer be         engine. For BPM systems, this
class defines separate methods,        available.                            mechanism can then be handled
one for each kind of message                                                 in a distributed manner using
“signature.” If the number of pos-     The Determine Action process          agents, rather than by centralized
sible methods is tangible, this is     might involve a high degree of        controllers that will not scale
not a problem; for those situations    complex process in its own right.     under large processing volumes.
requiring hundreds or thousands        Can this be accomplished using
of possibilities, such an approach     an OO approach? Absolutely.           Effectors
is impractical. For example, insur-    However, some agent platforms
ance companies have hundreds of        have mechanisms that support          When messages need to be
ways of rating a policy; retailers     action determination directly —       sent or events posted outside the
and shipping companies have an         instead of having the programmer      agent, effectors are the connec-
equally large number of rules and      do extra work in design, construc-    tion with the agent’s environment.
procedures for pricing a product       tion, testing, and maintenance.       It is the mechanism an agent
or service. To support this need in    Agents, by nature, are adaptive.      uses to interact with and mobilize
an agile and adaptive manner,                                                within the environment. This
the agent approach also has            Invoke Action
                                                                             could involve sending messages
the option of determining the                                                though a message bus or activat-
appropriate method after it            Once selected, the appropriate        ing a hardware device such as a
receives a request for service. In     action needs to be invoked. The       printer or robot arm.
this way, a policy agent can select    action execution may involve
                                       other actions within the agent, in    Agents Can Be Multiprocessing
the appropriate method to rate a
policy based on the requester's        which case, messages can be sent
                                                                             Agents can be single threaded
needs. Order agents can select         directly to the agent’s own detec-
                                                                             or designed to process input
                                       tors. If the action involves other
                                                                             concurrently (see Figure 16).
                                       agents, the messages need to be

VOL. 10, NO. 3                                                                                   www.cutter.com
EXECUTIVE REPORT                                                                                                                                                       21


Agents that can support concur-                                                                          Agent                      State
rency might also have a process                                                         Parse                Determine             Invoke
controller that coordinates the                                                         input                 action               action




                                                                Detector
various internal actions in some




                                                                                                                                                           Effector
                                                                                        Parse                Determine             Invoke
manner.                                               Input
                                                                                        input                 action               action
                                                                                                                                                                      Output
                                                                                                               •••
Reasoning, Planning, and                                                                Parse
                                                                                        input
                                                                                                             Determine
                                                                                                              action
                                                                                                                                   Invoke
                                                                                                                                   action
Scheduling

A common approach for agents                                  Figure 16 — Concurrent processing within the agent is possible.
can involve reasoning, planning,
and scheduling (see Figure 17). In
                                                                                          Agent
this approach, an agent can deter-
                                                                                           Parse             Determine          Invoke
mine the appropriate action by                                                             input              action            action
having a reasoner step consider                                                                                                     State


                                                                             Detector




                                                                                                                                                Effector
what it knows about its state


                                                                                          •••




                                                                                                                                      •••
                                                                                                         Coordinator
and the state of its environment                               Input                                                                                            Output
(including that of the service
requester). Additionally, the                                                              Parse
                                                                                           input
                                                                                                             Determine
                                                                                                              action
                                                                                                                                Invoke
                                                                                                                                action
reasoner agent can identify the
appropriate goals for acting based
on the detected input and state                                                           Reasoner             Planner          Scheduler
knowledge. Here, the agent may
optionally weigh the importance                                   Beliefs                       Intentions               Plan               Schedule
of certain goals (when there are                                           Desires, goals            Capabilities

several goals to choose from or
conflicting goals). For example,                                              Figure 17 — Determining the appropriate action
a scheduling agent may have a                                                 can involve reasoning, planning, and scheduling.
goal to ensure that all the tasks
in its care are executed on time.
Another goal might be that cer-                   intension. For example, the                                       of prefabricated processes from
tain priorities must be given to                  scheduling agent may need to                                      which to choose.
important tasks. The two goals,                   ensure a particular product is
however, may conflict when                        created for given order. The agent                                The planned process is then sent
priority tasks cause the delay of                 may need a make-or-buy decision                                   to the agent’s scheduler to deter-
other tasks. Here, the agent may                  resulting in a process that either                                mine which resources will be
have to weigh the cost of both                    fabricates the product inhouse or                                 assigned and when the process
and determine which service it                    purchases a fabricated item from                                  will be scheduled.5
intends to perform under a given                  an external source. Some agent
circumstance. A planner then                      planners dynamically assemble                                     On top of this, an agent may
determines the actual steps                       the planned set of steps to fit each                              also employ a coordinator
needed to execute the agent’s                     occasion; others have a library                                   that ensures that the agent’s


5Agents using this approach are sometimes referred to as “BDI” agents because they employ beliefs, desires, and intentions. Beliefs refer to what
the agent “believes” the state of its environment and other agents to be; desires basically refer to the goals the agent could attempt to achieve; and
intentions refer to the set of goals the agent actually intends to achieve under a given circumstances.



©2007 CUTTER CONSORTIUM                                                                                                                      VOL. 10, NO. 3
 22                                                                            ENTERPRISE ARCHITECTURE ADVISORY SERVICE


concurrent processing works har-               What if it were as hard to get your      approaches envision using objects
moniously: managing conflicts                  deliveries off schedule, as it is to     and agents together in the devel-
and optimizing the overall func-               keep ants out of your kitchen?           opment of software systems. In
tioning of the agent.                          (See Figure 18.) In other words,         other words, objects and agents
                                               agent systems do not require             are two distinct notions — each
But Keep In Mind …                             sophisticated agents to yield            having its particular place in soft-
Agent-based systems do not                     impressive results. With simple          ware development. The important
require complex forms of agents                rules and high fault tolerance,          point here is that the agent-based
like those previously described.               the colony thrives — a great first-      way of thinking brings a useful
For example, ant colonies have                 system model.                            and important perspective for sys-
been very successful for seven                                                          tem development, which is differ-
                                               Agents and Objects                       ent from (while also similar to)
million years. An estimated 10%
of the Earth’s animal biomass is               Just how different — or similar —        the OO way.
composed of ants, which vastly                 are agents and objects? Some
exceeds the human population.                  developers consider agents to be         Evolution of Programming
An ant has about 250,000 brain                 objects, except with more bells          Approaches
cells; humans have about 10 bil-               and whistles. Others see agents          Figure 19 illustrates one way
lion. Ants can lift 20 times their             and objects as different even            of thinking about the evolution
own body weight. (They may                     though they have many things             of programming languages.
be small, but they’re efficient!)              in common. However, both                 Originally, the basic unit of soft-
                                                                                        ware was the complete program
                                                                                        where the programmer had full
         What if it were as hard to get                                                 control. The program’s state was
         deliveries off schedule ...                                                    the responsibility of the program-
                                                                                        mer and its invocation determined
                                                                                        by the system operator. The term
                                                           ... as it is to keep ants
                                                           out of your kitchen?         “modular” did not apply because
                                                                                        the behavior could not be invoked
                                                                                        as a reusable unit in a variety of
        Figure 18 — Ant-like simplicity is still very successful. (Source: [7].         circumstances.

                                                                                        As programs became more com-
                                                                                        plex and memory space grew,
                 Monolithic        Modular  Object-Oriented Agent
                                                                                        programmers needed to introduce
                Programming      Programming Programming Programming
                                                                                        some degree of organization to
                                                                                        their code. Modular programming
      Unit
    Behavior
                 Nonmodular         Modular           Modular          Modular          employed smaller units of code
                                                                                        that could be reused under a vari-
      Unit                                                                              ety of situations. Structured loops
                   External         External          Internal          Internal
      State                                                                             and subroutines were designed
                                                                                        to have a high degree of local
      Unit
                   External
                                    External          External         Internal         integrity. While each subroutine’s
   Invocation                      (CALLed)          (message)      (rules, goals)      code was encapsulated, its state
                                                                                        was determined by externally sup-
        Figure 19 — Evolution of programming approaches (Source: [7].)                  plied arguments, and it gained

VOL. 10, NO. 3                                                                                               www.cutter.com
EXECUTIVE REPORT                                                                                             23


control only when invoked exter-        unnatural — more like an ani-          example, DHL found that its pric-
nally by a CALL statement. This         mated cartoon than animated life       ing and tracking applications
was the era when the primary            [9].                                   could be developed 50%-80%
programming unit was based on                                                  faster using agent development
modularizing procedures.                Yet most developers tend to build      tools like those available from
                                        centrally organized applications.      OSLO or Agentis Software (www.
In contrast, object orientation         They are also biased toward OO         agentissoftware.com). Time to
added to the modular approach           notions, such as class, association,   market was reduced and ROI
by maintaining its segments of          and message. While these con-          increased. This does not mean
code (or methods) as well as            structs are useful for a certain       that agent development tools
by gaining local control over the       category of applications, they do      will always reduce function
variables manipulated by its meth-      not directly address the require-      points, but it does for some
ods. However, in traditional OO,        ments of agents. As we have seen,      classes of application.
objects were considered passive         agents have such characteristics
because their methods were              as autonomy, mobility, and             The most compelling case for
invoked only when some external         adaptability.                          using agent technology is when
entity sent them a message.                                                    agents can enable solutions that
                                        Furthermore, business users like       were either impossible or very
Agents and Objects Together
                                        to express other concepts, such        difficult using conventional
                                        as rules, constraints, and goals       mechanisms. At manufacturing
Software agents have their own          and objectives as well as roles        companies like Deere & Company,
thread of control, localizing not       and responsibilities. In short, the    several hours each night are typi-
only code and state but their invo-     agent-oriented approach distin-        cally dedicated to computing opti-
cation as well. Such agents can         guishes between autonomous,            mized production schedules for
also have individual rules and          interactive, mobile entities           the shop floor’s use the next day.
goals, making them appear like          (agents) and the passive ones          When an organization has the
“active objects with initiative.”       of conventional OO (objects).          time to use techniques like these,
                                        This does not mean that object         agent technology is probably
In other words, the agent can           orientation is dead or passé. A        unnecessary. However, what if
determine when and how it               well-designed, agent-based sys-        John Deere’s IT department deliv-
acts. An agent-based approach           tem uses both objects and agents       ered its shop-floor schedule at the
is employed when a particular           — just as real-life organizations      beginning of a day when several
situation requires that process-        employ a balance of both               key employees were out sick or
ing be decentralized and                active and passive elements.           a vital piece of machinery in the
self-organized, instead of centrally    Furthermore, object technology         assembly line broke down? If they
organized. Any centrally organized      can be used to enable, rather than     needed to stop the assembly line
program written to handle the ant       drive, agent-oriented technology.      for several hours to recreate the
simulation (refer back to Figure 1)
                                                                               production schedule, this solution
would have been far too cumber-         CONCLUSION                             clearly is not the answer. Under
some. It would have required a
                                        Making Things Possible with            these circumstances, an agent-
single set of top-level rules telling
                                        Agents — Not Just Faster, Better,      based approach should be consid-
each ant precisely what to do in        and Cheaper                            ered. Instead of starting over to
every conceivable situation. Not
                                                                               reoptimize the schedule for all the
only would such an application          Using agents has produced
                                                                               resources, agents can be used to
be touchy and fragile, it would         applications faster and cheaper
                                                                               produce a local optimum. If one
likely end up looking jerky and         than other approaches do. For

©2007 CUTTER CONSORTIUM                                                                         VOL. 10, NO. 3
 24                                                              ENTERPRISE ARCHITECTURE ADVISORY SERVICE


resource is no longer available,       already employ agents inside their       [We need] a global, Web-
agents can look for a replacement      offerings but do not refer to them       enabled platform for multiple
                                                                                forms of collaboration. This
and “contract” with alternative        by name. Agent technology makes
                                                                                platform enables individuals,
resources. If the resources are        it possible to do things that were       groups, companies, and uni-
going to be delayed, agents in         previously either impossible or          versities anywhere in the
a supply chain can determine           could not be accomplished in an          world to collaborate — for
which way to solve the delay           acceptable time frame.                   the purposes of innovation,
                                                                                production, education,
with the least impact — using
                                       Where To Go From Here
                                                                                research, entertainment,
automated speeds.                                                               and, alas, war-making — like
                                       If you think agent technology            no creative platform before.
Even techniques such as linear                                                  This platform now operates
                                       might be appropriate for your
programming, neural networks,                                                   without regard to geography,
                                       organization, start by reading           distance, time, and, in the
and genetic algorithms are some-
                                       some books on agents (such as            near future, even language.
times used for similar situations.                                              Going forward, this platform
                                       [10]). Then, talk to companies
Again, when these provide                                                       is going to be at the center of
                                       that provide or use agent technol-
accurate results in a timely mat-                                               everything. [5]
                                       ogy (such as those mentioned in
ter, agents are not necessary.
                                       this report). In doing so, you will
However, for large complex sys-                                              This emergence of new business
                                       get an idea of how and where to
tems, such approaches can some-                                              processes is resulting in new IT
                                       use agents. Remember, though,
times become costly to configure                                             practices and approaches —
                                       that the biggest breakthrough
or reconfigure and also involve                                              where the two mutually reinforce
                                       with agents is that they are an
scalability problems. In such situa-                                         each other. Now that individuals,
                                       evolution of existing technologies.
tions, they are often run offline                                            groups, companies, and universi-
                                       What is revolutionary is the way
and do not respond rapidly to                                                ties are interacting on a global
                                       we think about and use agents to
dynamic changes in environment.                                              scale, IT must not forget its sup-
                                       design IT systems.
When this is the case, the horizon-                                          porting role. Agents are a key
tal approach of agent technology       Going Forward to the Global           ingredient in IT globalizing itself.
can significantly outperform a         IT Organization
centralized scheduling approach.                                             REFERENCES
                                       Agent technology is now neces-
                                       sary to reduce costs; to improve      1. Baker, Albert D., H. Van Dyke
Companies like IAI have devel-
                                       efficiency and effectiveness; and     Parunak, and Kutluhan Erol.
oped systems involving over
                                       to support the requirements of        “Agents and The Internet:
100,000 agents that can handle
                                       individuals, groups, companies,       Infrastructure for Mass
the resource rescheduling in a
                                       and universities as they collabo-     Customization.” IEEE Internet
manner of seconds — instead
                                       rate globally. More importantly,      Computing, Vol. 3, No. 5,
of hours. General-purpose agent
                                       it will enable us to create and       September 1999, pp. 62-69.
platforms, such as the OSLO Suite,
can be used, or a more special-        support a whole class of IT appli-
                                       cations and approaches that           2. Bonabeau, Eric, Marco Dorigo,
ized package can be chosen. For                                              and Guy Theraulaz. Swarm
example, agent-based logistics         we previously could not have
                                       developed.                            Intelligence: From Natural to
software is available from                                                   Artificial Systems. Oxford
Cougaar and Whitestein, and                                                  University Press, 1999.
                                       As Pulitzer Prize–winning author
event-driven sense-and-respond
                                       and columnist Thomas Friedman
solutions from Rhysome. Many
                                       said so eloquently:
other large software companies


VOL. 10, NO. 3                                                                                    www.cutter.com
EXECUTIVE REPORT                                                                                           25


3. British Broadcasting              9. Schoonderwoerd, Ruud, Owen           and improve OO methods and
Corporation (BBC) News.              Holland, Janet Bruten, and Leon         techniques. He participated in the
“Humble Ant Is a Software            Rothkrantz. “Ant-Based Load             development of UML, and is the
Saviour.” 28 January 1998            Balancing in Telecommunications         cochair of the OMG’s Analysis and
(http://news.bbc.co.uk/2/hi/         Networks.” Adaptive Behavior,           Design Task Force as well as the
science/nature/51337.stm).           Vol. 5, No. 2, 1996, pp. 169-207.       Agents Special Interest Group.

4. Ernst & Young Center for                               ,.
                                     10. Singh, Munindar P and               Mr. Odell conducts international
Business Innovation. “Embracing      Michael N. Huhns. Service-              seminars and workshops and
Complexity: Exploring the            Oriented Computing: Semantics,          provides consulting to major
Application of Complex Adaptive      Processes, Agents. Wiley, 2005.         companies worldwide. Formerly,
Systems to Business.” Christopher                                            Mr. Odell was the principal con-
Meyer (ed.). Proceedings from the    11. W3C. “Web Services                  sultant for KnowledgeWare, Inc.,
1996 Colloquium on the Business      Architecture.” W3C Working              where he pioneered and taught
Application of Complexity Science,   Group Note, 11 February 2004            the concepts of data modeling,
Boston, Massachusetts, USA, 17-19    (www.w3.org/TR/2004/                    information strategy planning,
July 1996.                           NOTE-ws-arch-20040211/)                 and CASE technology application.
                                                                             He now works as an independent
5. Friedman, Thomas L. The           ABOUT THE AUTHOR                        consultant to advance the
World Is Flat: A Brief History                                               state of agent technology
                                     James Odell is a Senior Consultant
of the Twenty-First Century.                                                 within the OMG and the com-
                                     with Cutter Consortium’s
Farrar, Straus and Giroux, 2005.                                             puter industry.
                                     Enterprise Architecture practice
6. Luck, Michael, Peter McBurney,    and a regular contributor to the
                                                                             Mr. Odell has coauthored, with
Onn Shehory, and Steven              Enterprise Architecture advisory
                                                                             James Martin, several books includ-
Willmott. “Agent Technology          service. He is a consultant, writer,
                                                                             ing Object-Oriented Analysis and
Roadmap: Computing as                and educator in the areas of agent-
                                                                             Design, Object-Oriented Methods:
Interaction (A Roadmap for           based and object-oriented (OO)
                                                                             Pragmatic Considerations, and
Agent Based Computing).”             systems. Throughout most of his
                                                                             most recently, Object-Oriented
AgentLink, 2005.                     35-year career, Mr. Odell has been
                                                                             Methods: A Foundation, UML
                                     heavily involved in developing bet-
                                                                             Edition. He also has published
7. Parunak, H. Van Dyke.             ter methods to understand, com-
                                                                             three books on agent-oriented
“DESK: Density-Based Emergent        municate, and manage system
                                                                             software engineering.
Scheduling Kernel.” Slide presen-    requirements. He was one of the
tation to DTE Energy, material       early innovators of information         Mr. Odell’s clients include
adapted in collaboration with        engineering methodologies.              Amazon.com, Netscape, Chrysler,
NewVectors, 2000.                                                            Capgemini, DHL, NASA, and other
                                     Mr. Odell is one of the first practi-
                                                                             major companies across various
8. Resnick, Mitchel. Turtles,        cal implementers of OO analysis
                                                                             business sectors in 19 different
Termites, and Traffic Jams:          and design. Working with the
                                                                             countries. He can be reached at
Explorations in Massively Parallel   OMG and other major method-
                                                                             email@jamesodell.com.
Microworlds. The MIT Press, 1997.    ologists, he continues to innovate




©2007 CUTTER CONSORTIUM                                                                       VOL. 10, NO. 3

								
To top