Role of Architects in Agile Organizations
Antonio Martini, Lars Pareto, and Jan Bosch
Abstract Agile software development is broadly adopted in industry and works
well for small-scale projects. In the context of large-scale development, however,
there is a need for additional structure in the form of roles and practices, especially
in the area of software architecture. In this chapter, we introduce the CAFFEA
framework that deﬁnes a model for architecture governance. The framework
deﬁnes three roles, i.e., chief architect, governance architect, and team architect,
as well as a set of practices and responsibilities assigned to these roles. The
CAFFEA framework has been developed and validated in close collaboration
with several companies.
Large software industries strive to make their development processes fast and more
responsive, minimizing the time between the identiﬁcation of a customer need and
the delivery of a solution. Support for such goals is given by Agile software
development (ASD), which has been proven successful in small software projects
in the last decade .
Some frameworks including sets of practices have been proposed for
implementing Agile principles of which few are suitable for large projects (e.g.,
Scrum). However, a major gap in such frameworks is the lack of activities to
enhance agility in the task of developing and maintaining a reference architecture,
necessary for the development of portfolios of products sharing large amount of
software (e.g., platforms) [1,2]. This lack might lead to an underdeveloped or
quickly eroded architecture, which is needed to coordinate large software projects,
to enhance future development (future features), and to assure cost reduction
(through software reuse). The key need, for the companies, is to reach the ability
of Agile architecting and therefore to redeﬁne the role of the software architects in
an Agile organization, which means understanding which activities related to
A. Martini (*) • L. Pareto • J. Bosch
Chalmers University of Technology, Gothenburg University, Gothenburg, Sweden
©Springer International Publishing Switzerland 2014
J. Bosch (ed.), Continuous Software Engineering,
software architecture are necessary to be carried out and by whom. We have
therefore redeﬁned the key roles for architects, with the responsibilities of archi-
tecture management that were missed in large Agile organizations: chief architect,
governance architect, and team architect.
• Chief architects, responsible for the whole overall portfolio architecture, which
might include more products and more than one system.
• Governance architects, responsible for areas of the architecture, related to single
products or systems or sub-systems, but not related to only one team.
• Team architects, the usually most experienced developer in a team who have the
most knowledge about the architecture and support/lead the team on such area.
Different levels in the hierarchy are usually connected with the level of abstrac-
tion of the software architecture and design for which the architects are
Such generic architecture roles might change from company to company, but
important questions are: What are they actually doing? What should they do? How
should they spend their time?
There are many activities suggested by the research community that need to be
done to develop and maintain a sound software architecture [3–12]. However, little
is said about who is responsible for which activities in a large Agile software
development organization. This gap is especially important where ASD is
employed, and the organization tends to avoid big time investments in upfront
design in favor of a shorter time to market.
In our previous studies, we have shown that the Agile teams need to communi-
cate and coordinate, and the need for strategic input is of utmost importance [13,
14]. Also, in  the authors introduce Agile in an SPL setting but report an open
the management of architecture evolution and refactoring without sacriﬁcing the principles
of agility and self-managed teams.
Lefﬁngwell  introduced the concept of architecture runway as an
existing or planned infrastructure sufﬁcient to allow incorporation of current and anticipate
requirements without excessive refactoring.
We will use such concept when we will explain the need for a Runway Team,
which in  has been identiﬁed as a Prototyping Team. Lefﬁngwell also mentions
the possibility, for large systems, to have architects outside of the teams. However,
no speciﬁc roles and activities have been mentioned for the architects: on the
contrary, Lefﬁngwell advocates the current impossibility of placing traditional
architect roles (e.g., system architects) present in many organizations (especially
in products with embedded software) into Agile frameworks. We have therefore
collected roles with speciﬁc responsibilities in a framework called CAFFEA (con-
tinuous architecture framework for embedded software and Agile).
40 A. Martini et al.
4.2 Research Design
We planned a multiple-case embedded case study involving seven sites in ﬁve large
software development companies. For conﬁdentiality reasons, we will call the
companies A, B, C, D, and E. The main rationale for selecting the cases was that
they needed to be developing software product lines and they had adopted or at least
been in transition to ASD.
Companies A–D had extensive in-house embedded software development, while
company E was developing general-purpose software. The choice for including
company E was to compare the results with non-embedded software development.
We involved three different units within the same company, C, and we will refer to
them as C
, and C
. We used this approach in order to assess the variance within
the same company. The companies studied were to have some years of experience
of ASD. The companies chosen were situated in the same geographical area
(Scandinavia) but were active on different international markets.
4.2.1 Cases Description
Company A is involved in the automotive industry. Part of the development is
carried out by suppliers and some by in-house teams following Scrum. The sur-
rounding organization follows a stage-gate release model for product development.
Business is driven by products for mass customization. The speciﬁc unit studied
provides a software platform for different products.
Company B is a manufacturer of recording devices. Teams work in parallel in
projects: some of the projects are more hardware oriented, while others are related
to the implementation of features developed on top of a speciﬁc Linux distribution.
The software involves in-house development with the integration of a substantial
amount of open-source components. Despite the Agile setup of the organization,
the iterations are quite long compared to the other companies involved in the study.
Company C is a manufacturer of telecommunication system product lines. Their
customers receive a platform and pay to unlock new features. The organization is
split into different units and then into cross-functional teams, most of which have
feature development roles. Most of the teams use their preferred variant of ASD
(often Scrum). Features were developed on top of a reference architecture, and the
main process consisted of a pre-study followed by few (ca. 3) sprint iterations. The
embedded cases studied slightly differed: C3 involved globally distributed teams
(Europe) while the other unit (C1 and C2) teams were co-located in the same city.
Company D is a manufacturer of a product line of devices for the control of
urban infrastructure. The organization is divided into teams working in parallel.
The organization has also adopted principles of software product line engineering,
such as the employment of a reference architecture.
4 Role of Architects in Agile Organizations 41
Company E is a company developing software for calculating optimized solu-
tions. The software is not deployed in embedded systems. The company has
employed ASD with teams working in parallel. The product is structured in a
platform entirely developed by E and a layer of customizable assets for the
customers to conﬁgure. E supports also a set of APIs for allowing development
on top of their software.
All the companies have one or more product lines and have adopted a
component-based software architecture, where they reuse a substantial part of the
system, such as several components or a shared platform. The language that is
mainly used is C and C++, with some parts of the system developed in Java and
Python. Company A uses a domain-speciﬁc language (DSL) to generate C code,
while company E uses a DSL for specifying rules to be converted into libraries. The
development at C
involves extensive XML.
4.2.2 Data Collection and Analysis
First we conducted a literature review to ﬁnd the main state-of-the-art architecture
activities that need to be done to develop and maintain a sound software architec-
ture [3–12]. During the interactive workshops, we mapped such activities into roles
with the help of the informants. With such process we found which activities were
already covered by the Agile process employed at the companies and which ones
were not. The workshops were recorded and transcribed. The analysis was done
following an approach based on grounded theory , mixing inductive and
deductive techniques and using a tool for qualitative analysis, to keep track of the
links between the codes and the quotations they were grounded to. The overall
process is shown in Fig. 4.1.
Architecture ac vi es
Decisions and changes
Monitoring the current
Fig. 4.1 CAFFEA framework and its components. The arrows show the analysis steps leading to
42 A. Martini et al.
4.3 Roles and Activities
4.3.1 Chief Architect
The main role of the Chief Architect (CA) is to take high-level decisions and to
drive and support governance architects and teams in order to reach strategic goals
such as the development and maintenance of a sound architecture that would
support the business goals of the organization. In many cases such goals include
the support for multiple products or product lines, which assures multiple incoming
to the company. All of the cases that we have analyzed have such portfolio of
multiple customers. The main activities for the CA are described in the following,
grouped by the main areas.
18.104.22.168 Risk Management
The CA should drive the activities connected to risk management related to product
and architecture evolution. Such an activity is done usually on a 2–3-year scope.
The CA is usually not directly involved in the detailed development; however, in
order to take feasibility decisions, the CA needs to elicit the information about the
current status of the system from the governance architects and team architects by
the institution of a virtual architecture team. The presence of CAs in estimation is
usual, but the involvement of the GA and TA is not often applied, although many
informants reported that such roles are important for understanding the feasibility
of deliveries that require commitment. The virtual team should also involve Product
Owners since risks in terms of cost related to the architecture need to be matched
with the business model and goals of the company.
22.214.171.124 Managing Decisions and Changes
The CA takes decisions that affect more products, for example, tools and frame-
works (e.g., libraries) that need to be used cross-company, or high-level design
decisions (e.g., architectural styles, archetypes). Such decisions cannot be taken by
development teams alone but need to be taken by architects with high experience
and overall picture of the system.
Decisions on the adoption of frameworks, tools, and high-level design need to be
followed by decisions to support them by educating the development teams. Such
decisions involve the budget, so they need to be taken by (or in combination with)
4 Role of Architects in Agile Organizations 43
126.96.36.199 Pattern Distillation
The chief architect should participate in the learning activity with the governance
architects (described later) on distilling patterns. Such retrospective activity is not
done per feature, but it involves the architectural analysis of the system for a whole
product or product line. It’s not a frequent activity, but it’s usually performed before
high-level patterns are chosen and product-wise (and therefore cost-impacting)
decisions need to be taken, for example, the starting of a complete new product
or its inclusion in the organization portfolio.
188.8.131.52 Providing Architecture Documentation (Communication
One of the main tasks for the CA should be to communicate the overall architecture
and the purpose of it. The best way, according to the informants, is to broadcast
such information in a visual way and in person. The mentioned means for this
activity are videos, “road shows,” and plenary sessions with all the consumers of the
The CA needs to “order” education for the consumers after a decision is made to
change infrastructures, high-level design patterns, or processes. It’s not likely that
the CA alone is able to take care of such education. For what concerns the
architecture, the CA should delegate most of the documentation creation and
education to the GAs.
184.108.40.206 Receiving Input About the Current Status of the System
According to the informants, the current communication practices lack good mech-
anisms for providing input to the CAs. In fact, CAs cannot go around the organi-
zation and talk to every consumer of the architecture. Input is fundamental for the
CA in order to take informed decision and to be aware of bad decisions taken
previously that need to be addressed (e.g., about tool chains not working as
expected). Some of the means for such communication input are:
• Plenary sessions
44 A. Martini et al.
4.3.2 Governance Architect
The Governance Architect (GA) is an intermediate role in the organization and the
key for scalability of the architecture development and maintenance in a large Agile
setting. Such role should function as a link among several teams that need to be
coordinated and supported when developing features within a (evolving) system
architecture. The main activities for the CA are described in the following, grouped
by the main areas.
220.127.116.11 Inter-features Architecting (Architecture Decision)
The GA should be responsible for decisions about inter-feature architecting. The
development of certain features might require adding design patterns or making
sure that the architecture is optimal not just for one feature but for more, for
example, to support features planned for the near future. This is something that
requires a wider view of the system architecture than the view focused on a single
feature typical of the Agile team. These decisions are usually taken considering also
high-level architecture. An example of such architecture is the production of a
meta-model for the understanding of the system components and their connectors
Besides the technical integration of the features, the GA needs to monitor and
drive an architecture that would minimize or make clear the inter-team effects or
the development of each team on the other. For this purpose, there is also the need
to deﬁne the boundaries for the teams, i.e., the critical points in the architecture in
which the risk is high for the team to interact with the others (see also documen-
tation [*]). GAs need to take architecture decisions involving at least 2 or more
teams, although the more teams are involved, the more coordination effort needs to
be spent on such activities.
The GA needs to take product-oriented decisions, but also to monitor and
support the team in following such decisions. With the focus of the team on the
features, someone needs to be able to “protect” the whole product, for example, in
terms of quality or pattern distillation (see CA [*]). In order to have such a mind-set
and information, there is the need to maintain a strong and iterative communication
link with the CA.
18.104.22.168 Architecting for Testability
An important point made by the informants is the need to architect to achieve
testability. The continuous integration team, responsible of the integration at (sub-)
system level, is a stakeholder of the team. Decisions about patterns and measure-
ment mechanisms to be embedded in the code for testing and quality management
need to be taken by the GAs.
4 Role of Architects in Agile Organizations 45
22.214.171.124 Risk Management
One of the main needs in a large organization is to balance the prioritization of
short-term and long-term goals. Such risk management activity is done on different
levels, and the GA should be a connector between the levels by participating in the
relative risk management activities. The GAs need to plan long-term product
development and architecture evolution with CAs and Product management
(High-level Product Owners), while they also need to be involved in the prioriti-
zation of features and architecture improvements at lower levels (closer to design
and implementation) with the TA and, if present, a lower-level Product Owner. In
two of the cases studied, belonging to the same large organization but being quite
disconnected sites, there was the presence of a so-called Operational Product
Owner, who was responsible for the feature prioritization directly with the team.
126.96.36.199 Controlling Erosion
The GAs are the main drivers for monitoring and reacting to architecting erosion.
Such activity is quite complex, and the shift to Agile has also led to a shift from a
document-centric architecture monitor to the more iterative and in-person partici-
pation of the architects. Controlling erosion needs also the support within the team
by the TAs, but the information provided by the team lead to decisions that have to
be taken with large perspective (see [*] decisions). On the other hand, the CA
cannot monitor the whole system to understand what is happening (see [*] CA
Consequently, the GAs are needed to lead architecture erosion control by
coordinating sub-activities, such as:
• Retrospective sessions: the topic of the retrospective sessions (e.g., postmortem
analysis) suggested by Agile practices is team performance. From such activity,
the team learns how to be more efﬁcient. However, speciﬁc sessions or
sub-sessions need to be organized and led by the GAs, in order to focus also
the retrospective towards architecture. Such sessions might be organized at the
• Code reviews: a usual practice is to peer-review the code. However, such
practice seems to be quite a costly one and is not always possible because of
the time constraints due to the low number of GAs present in the organization
compared to the amount of code and the time invested in the other activities
listed also in this section. For this reason, the GA should be able to order
automation for such code-reviews.
• Order automation: to facilitate the monitor of erosion, the inconsistencies with
respect to the desired architecture or architecture principles and requirements
46 A. Martini et al.
need to be visualized by analyzing the system. The GAs should be able to
“order” (allocate the budget and resources) automation to an infrastructure
team (or third party vendor) that could put in place a tool for visualizing and
monitoring some erosion.
188.8.131.52 Architecture Education for the Teams (Communication
Although the chief architect is responsible for the production and communication of
high-level architecture guidelines and decisions, the governance architect should be
responsible for the capillary spreading of knowledge to all the teams. In such role,
the governance architect, in the organization, needs to have enough overview and
architecture knowledge to express the important concepts.
The main means for communicating architecture and educating the teams is
maintaining architecture documentation (inter-team, inter-feature): although the
transition to ASD has questioned the need for the maintenance of superﬂuous
architecture documentation, part of it is critical and needs to be very well created
and kept updated. Such documentation is the one concerning architectural patterns
and requirements that are shared by more than one team. As an example, there
might be patterns that involve temporal requirements (e.g., involving the access to a
centralized database) that need to be expressed (and veriﬁed) in order to keep the
teams from hindering each other. An important requirement for the architecture
documentation is that it should be well navigable by the team members, meaning
that the link between the documents and the actual source code should be well
184.108.40.206 Knowledge of the Status of the System (Communication Input)
Another important point in architecture communication is that the architect should
be able to have knowledge of the status of (part of) the system. The main problem
that would arise without having such knowledge would be to have architects who
deﬁne and evolve a system architecture that is completely different from what is
actually implemented. Such situation obviously hinders the decisions when risk
analysis is done for the augmentation of the products, e.g., the implementation of
new features. It is of utmost importance, therefore, that the architects continuously
and iteratively check the actual status of the system. For doing so, there might be
tools, but, according also to the Agile view of the importance of face-to-face
communication, architects should organize and participate in events including
experienced members of the teams. Such events are in line with the continuous
learning principle and with the practice, suggested in some Agile frameworks, of
conducting retrospective. In order for the architecture to be correctly communicated
to the governance architect, architecture retrospectives (i.e., retrospective sessions
4 Role of Architects in Agile Organizations 47
focused on reﬂecting on the current status of the architecture) have been found to be
a good means.
4.3.3 Team Architect
The Team Architect (TA) is the actual executor of the architecture in the FT. The
role is important for the capillary spreading of architecture knowledge and for
collecting input from the FT. However, this role is not a dedicated role in the team
but rather a set of responsibilities that need to be mapped to a member of the team,
who might change from time to time.
220.127.116.11 Risk Management
As mentioned in the previous sections, the TA should participate in risk manage-
ment activities with the CA, GA, and P in order to represent the interest of the teams
in feasibility discussions.
18.104.22.168 Managing Decisions and Changes
The TA leads detailed architecture decisions on a detailed design level.
22.214.171.124 Providing Architecture Documentation (Communication
The TA takes input from the GA and should be able to communicate the architec-
ture to support the FT education packages created by the GAs.
126.96.36.199 Monitoring the Current Status of the System (Communication
The TA is responsible for collecting FT’s input, to be communicated to the
GA. This way, the TA lifts decisions for architecture evolution and decisions that
might affect other FTs.
4.3.4 Gap in the Current Practices
The main groups of practices that have been recognized as currently missing are:
48 A. Martini et al.
• Architecture risk management (prioritization of short-term and long-term tasks)
• Architecture decision and changes
• Communication of architecture, composed by two-way directions:
• Providing architecture directions
• Monitoring the current status of the system
The combination of the previous components leads to the identiﬁcation of a
major gap in the current organizations, the lack of architecture technical debt
management. Such phenomenon is recently being studied from different angles
[18,19] and is concerned with the organizations taking risk-informed architecture
decisions about which architecture changes, such as refactoring or evolution (con-
sidered the actual debt if not done) need to be conducted for having an acceptable
ratio of investment/effort. The lack of architecture technical debt management
might quickly lead the companies to crisis points where adding new business
value to the software product lines (new features or new products) incur in major
efforts, paralyzing the long-term responsiveness .
Agile software development is broadly adopted in industry and works well for
small-scale projects. In the context of large-scale development, however,
there is a need for additional structure in the form of roles and practices,
especially in the area of software architecture.
In this chapter, we introduced the CAFFEA framework that deﬁnes a
model for architecture governance. The framework deﬁnes three roles, i.e.,
chief architect, governance architect, and team architect, as well as a set of
practices and responsibilities assigned to these roles. For the chief architect,
these practices include risk management, managing decisions and changes,
pattern distillation, providing architecture documentation, and receiving
input about the current status of the system. The governance architect is
concerned with inter-features architecting (architecture decisions),
architecting for testability, risk management, controlling erosion, architecture
education for the teams, and maintaining knowledge of the status of the
system. Finally, the team architect focuses on risk management, managing
decisions and changes, providing architecture documentation, and monitoring
the current status of the system.
The CAFFEA framework has been developed and validated in close
collaboration with several companies. Currently, most of the companies in
the Software Center have adopted or are in the process of adopting the
Although the framework has brought signiﬁcant beneﬁt for the companies,
there still are some open items, as presented earlier in the chapter, and in
future work, we aim to address those as well as validate the framework with
other companies in other industries.
4 Role of Architects in Agile Organizations 49
1. Dingsøyr, T., Nerur, S., Balijepally, V., Moe, N.B.: A decade of agile methodologies: towards
explaining agile software development. J. Syst. Softw. 85(6), 1213–1221 (2012)
2. Daneva, M., van der Veen, E., Amrit, C., Ghaisas, S., Sikkel, K., Kumar, R., Ajmeri, N.,
Ramteerthkar, U., Wieringa, R.: Agile requirements prioritization in large-scale outsourced
system projects: an empirical study. J. Syst. Softw. 86(5), 1333–1353 (2013)
3. Kruchten, P.: What do software architects really do? J. Syst. Softw. 81(12), 2413–2416 (2008)
4. Tang, A., Avgeriou, P., Jansen, A., Capilla, R., Ali Babar, M.: A comparative study of
architecture knowledge management tools. J. Syst. Softw. 83(3), 352–370 (2010)
5. Pareto, L., Eriksson, P., Ehnebom, S.: Architectural descriptions as boundary objects in system
and design work. Model Driven Eng. Lang. Syst. 6395, 406–419 (2010)
6. Williams, B.J., Carver, J.C.: Characterizing software architecture changes: a systematic
review. Inf. Softw. Technol. 52(1), 31–51 (2010)
7. de Silva, L., Balasubramaniam, D.: Controlling software architecture erosion: a survey. J. Syst.
Softw. 85(1), 132–151 (2012)
8. Qumer, A.: Deﬁning an integrated agile governance for large agile software development
environments. In: Concas, G., Damiani, E., Scotto, M., Succi, G. (eds.) Agile Processes in
Software Engineering and Extreme Programming, pp. 157–160. Springer, Berlin, Heidelberg
9. Drury, M., Conboy, K., Power, K.: Obstacles to decision making in Agile software develop-
ment teams. J. Syst. Softw. 85(6), 1239–1254 (2012)
10. Zimmermann, O., Miksovic, C., Ku
¨ster, J.M.: Reference architecture, metamodel, and model-
ing principles for architectural knowledge management in information technology services.
J. Syst. Softw. 85(9), 2014–2033 (2012)
11. Unphon, H., Dittrich, Y.: Software architecture awareness in long-term software product
evolution. J. Syst. Softw. 83(11), 2211–2226 (2010)
12. McAvoy, J., Butler, T.: The impact of the Abilene Paradox on double-loop learning in an agile
team. Inf. Softw. Technol. 49(6), 552–563 (2007)
13. Martini, A., Pareto, L., Bosch, J.: Enablers and inhibitors for speed with reuse. In: Proceedings
of the 16th International Software Product Line Conference, vol. 1, pp. 116–125. New York,
14. Martini, A., Pareto, L., Bosch, J.: Communication factors for speed and reuse in large-scale
agile software development. In: Proceedings of the 17th International Software Product Line
Conference, pp. 42–51. New York, USA, (2013)
15. Bosch, J., Bosch-Sijtsema, P.M.: Introducing agile customer-centered development in a legacy
software product line. Softw. Pract. Exp. 41(8), 871–882 (2011)
16. Lefﬁngwell, D.: Scaling Software Agility: Best Practices for Large Enterprises. Pearson
17. Strauss A., Corbin, J.M.: Grounded Theory in Practice. SAGE (1997)
18. Tom, E., Aurum, A., Vidgen, R.: An exploration of technical debt. J. Syst. Softw. 86(6), 1498–
19. Nord, R.L., Ozkaya, I., Kruchten, P., Gonzalez-Rojas, M.: In search of a metric for managing
architectural technical debt. In: 2012 Joint Working IEEE/IFIP Conference on Software
Architecture (WICSA) and European Conference on Software Architecture (ECSA), pp. 91–
20. Martini, A., Bosch, J., Chaudron, M.: Architecture technical debt: understanding causes and a
qualitative model. Presented at 40th Euromicro Conference on Software Engineering and
Advanced Applications, Verona, pp. 85–92 (2014). doi:10.1109/SEAA.2014.65
50 A. Martini et al.