The future success of EUD depends on

creating tools that end users are motivated to

learn and use in daily work practices.

META-DESIGN: By G. Fischer,
E. Giaccardi, Y. Ye,

A MANIFESTO FOR A.G. Sutcliffe,

and N. Mehandjiev

END-USER DEVELOPMENT
End-user development (EUD) activities range Therefore, EUD is only a
from customization to component configuration partial success story. Here, we
and programming. Office software, such as the argue the spread of EUD
depends on a fine balance
ubiquitous spreadsheet, provides customization between user motivation, effec-
facilities, while the growth of the Web has added tive tools, and management sup-
impetus to end-user scripting for interactive func- port. We explore that balance
tions in Web sites. In scientific and engineering and investigate a future
domains, end users frequently develop complex sys- approach to EUD—meta-
design—that proposes a vision
tems with standard programming languages such as in which design, learning, and
C++ and Java. However, only a minority of users development become part of
adapt commercial off-the-shelf (COTS) software everyday working practice.
products. Indeed, composing systems from reusable Designing language for user-
components, such as enterprise resource planing computer communication poses
a conflict between complexity
(ERP) systems, defeats most end users who resort to and power. More complex lan-
expensive and scarce expert developers for imple- guages can address a wider range
mentation. of problems but impose an

COMMUNICATIONS OF THE ACM September 2004/Vol. 47, No. 9 33

increased learning burden on the user. Text-based grams as instructions from user manipulations of
languages tend to be more complex because the syn- agent worlds. The graphical agent worlds must still
tax and lexicon (terminology) must be learned from be designed, but once present, programming by
scratch, as with any human language. Consequently, example [5] can infer instructions from the users’
languages designed specifically for end users repre- actions; for example, in a robot game the user
sent the programmable world as graphical metaphors demonstrates an agent bumping into a wall followed
containing agents that can be instructed to behave by by reversing two steps and changing direction. The
condition-action rules. The aim is to reduce the cog- system infers the condition-action rule of detect-a-
nitive burden of learning by shrinking the conceptual collision followed by the appropriate reverse-and-
distance between actions in the real world and pro- change-direction response. This approach reduces
gramming. learning by semiautomatic rule acquisition, but the
A key trade-off in EUD languages is between their downside is the learning system can make mistakes.
scope of application and learning costs, as illustrated The learning styles range from more complete infer-
in Figure 1. In the high cost, high scope cell are tradi- ence to direct instruction, where the system learns
tional programming lan- only when given a command.
guages, Java and C++, Cost of learning Direct instruction requires
employed by highly moti- High Low the user to anticipate all the
vated end users particularly possible rules and learning
Java EUD
in scientific domains. At the C++ ideal situations, while the com-
convergence of this cell and Current EUD envs
plete inference approach is
the high scope, lower cost High Agentsheets limited by the system’s
Alice
cell are the majority of cur- Java Script domain knowledge. Devel-
rent EUD languages that VB Script Excel macros oping the model is the diffi-
have evolved as simplified Scope Office Applications
cult part and therein lies the
versions of full program- Report writers real challenge for end-user
Query screen
ming languages, for exam- Domain engineering builders design—abstract conceptual
ple, Web scripting Low languages
SDL Domain-specific thinking. Complex domains
languages. The low scope, Hardware design languages require sophisticated analysis
Customization
high cost cell is occupied by Adaptation and modeling skills; pro-
only a small number of gramming is only part of an
domain-specific program- end-user developer’s needs.
ming languages developed Figure 1. Cost-scope The goal for EUD tools is to reduce the learning
trade-offs in EUD tools.
to address the requirements burden while providing powerful facilities to address
in complex engineering domains, such as device con- a wide range of problems. Given that some learning
trollers. These languages impose a considerable burden will always be present, tools must motivate
learning burden, but are worth it for improving effi- their users. We propose a meta-design approach [4],
ciency over a general-purpose language. The low cost, where users are motivated to learn by examples and
low scope cell contains domain-specific EUD lan- demonstrations of working systems to show them
guages that lower the learning burden but at the price what is achievable.
of addressing only a specific application area. In this
cell, EUD languages merge with the customization Managerial and Social Perspective
of COTS software packages so the act of program- EUD is a long-standing concern within organiza-
ming is reduced to entering parameters in a form-fill- tions. Managerial issues, illustrated in Figure 2, are
ing dialogue. Closer to the higher scope boundary based on previous surveys of end-user computing
are macro languages that extend the office-style [2, 9] and our more recent investigation into the
applications, for example, formulae for Excel spread- task-organizational fit of EUD technology [6]. Do-
sheets, and database query languages. Finally, the it-yourself development is a balance of benefits and
high scope, low cost cell is the EUD ideal, although cost. User motivators are empowerment from being
still largely unattained. The current state-of-the-art able to complete a job more effectively, speed of
EUD environments provide graphical worlds to cre- development, flexibility and local control so pro-
ate programmable agents that still impose a learning gramming can be on demand. Another benefit is
burden of instructing agents with condition-action eliminating potential miscommunications of
rules, and designing agent models. requirements to specialist software engineers, thus
Active EUD environments attempt to infer pro- avoiding the frustration with perceived poor ser-

34 September 2004/Vol. 47, No. 9 COMMUNICATIONS OF THE ACM

vice from the IS department. Success stories can hands, often using standard programming languages.
create motivational capital to help users over the However, in most business domains, training, techni-
hump of learning until actual benefits arrive in the cal, and management support are vital for helping
form of working applications. User motivation EUD flourish. A culture of cooperation shares the
should be encouraged during the early stages of responsibility for developing accurate and effective
adoption by management support, training, and solutions. Local experts among the end-user commu-
task forces to spread best practice and expertise. nity spread expertise and advice, although power
This counteracts user costs such as selecting appro- users can be prone to migrating to the wrong side of
priate technology, installing and learning it, pro- the “us” and “them” (IT department) fence [7, 8].
gramming, and debugging. Technology should provide easy integration with
A number of context and management issues other information systems and optimized support for
influence the balance between costs and benefits. For EUD tasks. Progress in the technology area is still
example, EUD can be dangerous in safety-critical necessary to unlock the true potential for EUD.
domains where software must be reliable and accu- The set of EUD critical success factors suggests the
rate. User costs can be sig- need for a socio-techni-
nificantly influenced by Culture
cal approach to increase
power, control reliability
the scale and complexity responsibility Management accuracy
issues
Training
Support user motivation and
standards security Local experts
of the domain, so safer, cost effectiveness risk decrease cognitive and
less complex domains potential can
reduce
organizational costs.
conflict increase
should be selected for empowerment
User
learning Such an approach sug-
flexibility User programming Critical success
EUD. Changeability of speed of delivery motivations trade-off costs debugging factors gests a future technolog-
local control
the domain can be a poor IS Dept.
service
ical framework with
motivator for EUD adop- influence reduce tools for discovery-led
tion, since end users can Context complexity
criticality
Technology:
ease of use design to balance learn-
issues integration
respond to rapidly evolv- changeability task support ing costs with results-
ing requirements more driven motivation. We
quickly than traditional propose meta-design,
development; however, Figure 2. Relationships which is an evolution of domain-oriented design
between social and managerial
rapid change can lead to issues in EUD. environments (DODEs) [3] as a vision in which
throw-away software and design, learning, and development become everyday
lost development effort. Management issues include working practice.
risks associated with EUD, perceived by IT manage-
ment to create unreliable and unmanageable soft- Meta-design
ware. Other risks are inaccurate information and Meta-design characterizes objectives, techniques,
security with increased exposure to hacking attacks. and processes for creating new media and environ-
The conflict between IT management and end ments allowing “owners of problems” (that is, end
users over power, authority, and control of IT systems users) to act as designers. A fundamental objective
may be a productive force for change or it may lead of meta-design is to create socio-technical environ-
to disruption, mistrust, and failure. It can be argued ments that empower users to engage actively in the
that enforcing standards and controlling end users continuous development of systems rather than
leads to more cost-efficient development and less being restricted to the use of existing systems.
waste from unreliable software. However, rigid top- In all design processes, two basic stages can be dif-
down control may only cause resentment among end ferentiated: design time and use time. At design time,
users. The control-power conflict between users and system developers (with or without user involve-
IT management will not evaporate; but constructive ment) create environments and tools. In conven-
support and training fosters success, encourages tional design they create complete systems. Because
responsibility, and enables management to control by the needs, objectives, and situational contexts of users
leadership. can only be anticipated at design time, users often
Critical success factors for EUD depend on the find the system unfit for their tasks at use time, thus
domain. In a culture of high end-user motivation and requiring modification of existing systems. To
low managerial influence—a common situation in accommodate unexpected issues at use time, systems
scientific and engineering domains, educational must be “underdesigned” at design time. Underde-
applications, and interactive art—success is simply a sign represents a fundamental shift in the approach to
matter of users taking development into their own the creation of systems, but it does not mean less

COMMUNICATIONS OF THE ACM September 2004/Vol. 47, No. 9 35

work or fewer demands on the design team. Instead of meta-design, we have applied the framework in a
of designers aiming at designing complete solutions number of different application areas, including
for users at design time, underdesign aims to provide three briefly mentioned here:
social and technical instruments for the owners of Social creativity. Complex design problems
problems to create the solutions themselves at use require more knowledge than any single person can
time. Within the overall approach of meta-design, possess, and the knowledge relevant to a problem is
underdesign is a defining activity to create design often distributed among stakeholders from different
spaces for others. perspectives and backgrounds. The solution of com-
Meta-design extends the traditional notion of sys- plex design problems requires social creativity in
tem development to include users in an ongoing which all stakeholders reach a shared understanding
process as co-designers, not only at design time but by contributing their different points of view and
throughout the entire existence of the system. A nec- knowledge. We have applied the meta-design
essary, although not sufficient, condition for meta- approach in the creation of augmented reality envi-
design is that software systems include advanced ronments in urban planning [1]. The tools them-
features permitting users to create complex cus- selves are not solutions to any particular problem,
tomizations and extensions. Rather than presenting but provide the socio-technical environment for
users with closed systems, meta-design provides them stakeholders to become informed participants. The
with opportunities, tools, and social structures to immediate and visual feedback facilitates the cre-
extend the system to fit their needs. Meta-design ation of a shared understanding leading to new
shares some important insights, new ideas, and
objectives with user-cen- new artifacts as a result
Seeded Evolved Reseeded
tered and participatory Information Information Information of collaboration.
Space Space Space
design, but it transcends Open source. Open
these objectives by source development is an
changing the processes by Users activity in which a com-
which systems and con- Seeding Reseeding
munity of software devel-
tent are designed. Meta- Evolutionary Growth opers collaboratively
design shifts control from constructs systems to
designers to users and Developers Users Developers Users help solve problems of
empowers users to create shared interest and for
and contribute their own Figure 3. The seeding, mutual benefit. The original designers of an open
evolutionary growth, and
visions and objectives. reseeding process model. source system do not provide a complete solution
Meta-design promotes that addresses all problems of potential users; they
“designing the design process” to a first-class activity, provide a seed that can be evolved by users at use
so that creating the technical and social conditions time. The ability to change source code, the techno-
for broad participation in design activities becomes logical means of sharing changes over the Internet,
as important as creating the artifact itself. It creates and the spontaneous social support among commu-
the enabling conditions for collaborative design in nity members are the enabling conditions for collab-
which all participants, not just skilled computer pro- orative construction of software. Software is changed
fessionals, incrementally acquire ownership of prob- from a fixed entity produced and controlled by a
lems and contribute actively to their solutions. closed group of designers to an open effort that
To support meta-design, we have developed the allows a community to design collaboratively follow-
seeding, evolutionary growth, and reseeding (SER) ing the framework provided by the SER process
process model. As illustrated in Figure 3, SER is a model. The success of open source systems exempli-
descriptive and prescriptive model for large evolving fies meta-design by openly embracing users as co-
systems and information repositories, postulating designers by releasing incomplete code; actively
that systems that evolve over a sustained time span soliciting and incorporating user contributions;
must continually alternate between periods of activ- strategically sharing the control over original design-
ity, unplanned evolution, and periods of deliberate ers and users by granting users direct access to source
(re)structuring and enhancement. The SER model code; aggressively promoting mutual learning among
encourages designers to conceptualize their activity as community members through mailing lists; and
meta-design, thereby supporting users as designers deliberately fostering a reward and recognition struc-
rather than restricting them to passive consumers. ture that motivates active participation by explicitly
To demonstrate the broad applicability and power acknowledging and promoting contributors [10].

36 September 2004/Vol. 47, No. 9 COMMUNICATIONS OF THE ACM

Open source projects based on meta-design have a must address the challenges of creating new mind-
lower cost for each user because the development cost sets, new sources of creativity, cultural changes, and
is distributed among a large number of participants innovative societies. It has the potential to create a
and individual contributions are shared. culture in which all participants in collaborative
Interactive art. Conceptualized as meta-design, design processes can express themselves and engage in
interactive art [4] focuses on participation and col- personally meaningful activities. c
laboration as forms of co-creation, in which users
become co-developers of artwork. The original seed References
design establishes a context in which users can 1. Arias, E.G., Eden, H., Fischer, G., Gorman, A., and Scharff, E. Tran-
scending the individual human mind: Creating shared understanding
actively produce new content and meaning through a through collaborative design. ACM Trans. on Computer-Human Inter-
process of mutual interaction and evolutionary action 7, 1 (2000). ACM, NY, 84–113.
2. Brancheau, J.C., and Brown, C.V. The management of end user com-
growth. By putting the tools rather than the object of puting: Status and directions. ACM Computing Surveys 25, 4. ACM,
design in the hands of users, interactive art seeds col- NY, (1993), 437–482.
laboration between the participants (both technical 3. Fischer, G. Domain-oriented design environments. Automated Soft-
ware Engineering, 1, 2 (1994), 177–203.
and human) and sees this interaction as the real 4. Fischer, G., and Giaccardi, E. Meta-design: A framework for the
object of creative production. Hence meta-design future of end user development. End User Development: Empowering
People to Flexibly Employ Advanced Information and Communication
creates interactive systems that define the conditions Technology. H. Lieberman, F. Paternò, and V. Wulf, Eds. Kluwer Aca-
for interaction. Meta-design environments not only demic Publishers, Dordrecht, The Netherlands, 2004, in press.
allow users to create content, but also modify the 5. Lieberman, H., Ed. Your Wish Is My Command: Programming By
Example. Morgan Kaufmann, San Francisco, 2001.
behavior and components of the system at use time 6. Mehandjiev, N., Sutcliffe, A.G., and Lee, D. Organisational views of
through interaction (see A-Volve; end user development. End User Development: Empowering People to
Flexibly Employ Advanced Information and Communication Technology.
www.iamas.ac.jp/~christa/). The initial seed is often H. Lieberman, F. Paternò, and V. Wulf, Eds. Kluwer Academic Pub-
developed by a community of artists, and can be lishers, Dordrecht, The Netherlands, 2004, in press.
adjusted and improved according to the talk-back 7. Mumford, E., and Henshall, D. A Participative Approach to Computer
System Design. Associated Business Press, London, 1979.
deriving from the continuing experience of using the 8. National Research Council. Beyond Productivity: Information Technol-
creative environment as in SITO, (www.sito.org), a ogy, Innovation and Creativity. National Academy Press, Washington,
virtual community of artists-participants. Interaction D.C., 2003.
9. Powell, A., and Moore, J.E. The focus of research in end user com-
and evolution occur both at the level of the develop- puting: Where have we come since the 1980s? Journal of End User
ment of materials and at the level of the creation, Computing 14, 1 (2002), 3–22.
10. Ye, Y., and Kishida, K. Toward an understanding of the motivation of
elaboration and completion of collective artworks. open source software developers. In Proceedings of the 25th Interna-
Interactive art emphasizes different objectives com- tional Conference on Software Engineering (Portland OR, 2003) ACM,
pared to traditional design approaches, including cul- NY, 419–429.
tural shifts from following guidelines and rules to
learning from exceptions and negotiations, content G. Fischer (gerhard@cs.colorado.edu) is a professor in the
Department of Computer Science and director of the Center for
to context of design, changing focus from design Lifelong Learning and Design at the University of Colorado at
objects to process, and from working with represen- Boulder.
tation to the act of construction. E. Giaccardi (giaccard@cs.colorado.edu) is a research associate in
the Center for Lifelong Learning and Design in the Department of
Computer Science at the University of Colorado at Boulder.
Conclusion Y. Ye (yuwen@cs.colorado.edu) is a research associate in the Center
To evolve, EUD development needs technologies for Lifelong Learning and Design in the Department of Computer
that foster collaboration between communities of Science at the University of Colorado at Boulder.
end-user designers and users and managers, while A.G. Sutcliffe (a.sutcliffe@co.umist.ac.uk) is a professor of
systems engineering and director of the Centre for Human Computer
increasing motivation and reducing cognitive and Interface Design in the School of Informatics at the University of
organizational costs. Meta-design provides a path- Manchester, U.K.
way to transform development as coding—a dis- N. Mehandjiev (mehandjiev@acm.org) is a senior lecturer in the
crete computing activity—into design of artifacts as School of Informatics at the University of Manchester, U.K.
part of the users’ work (or leisure) practice.
Permission to make digital or hard copies of all or part of this work for personal or
Meta-design puts owners of problems in charge of classroom use is granted without fee provided that copies are not made or distributed
creating open, evolvable systems that address the lim- for profit or commercial advantage and that copies bear this notice and the full cita-
tion on the first page. To copy otherwise, to republish, to post on servers or to redis-
itations associated with closed systems. Open systems tribute to lists, requires prior specific permission and/or a fee.
allow significant modifications when the need arises
and the evolution takes place through modifications
by the owners of problems as a major design activity.
Meta-design is more than a technical problem; it © 2004 ACM 0001-0782/04/0900 $5.00

COMMUNICATIONS OF THE ACM September 2004/Vol. 47, No. 9 37