Keywords

1 Introduction

Today’s fast-changing business environment forces companies to react quickly to stay competitive [34]. In our digitalized world, software development is particularly impacted by this need for agility [6]. While traditional development methods increasingly reach their limits [6], agile practices and frameworks, like Scrum [26], started spreading widely due to their potential benefits regarding responsiveness and resilience to change [2]. Motivated by the success of agile methods in small projects, organizations have begun to use them in a larger context [2, 3], including the joint development of software by multiple agile teams and the adoption of agile methods and principles across the organization [4].

The complexity of scaling agile methods and resulting changes in the organizational structure, e.g., toward cross-functional teams, lead to challenges [2, 3]. Besides efficiently coordinating multiple teams, organizations struggle with undiscovered dependencies, redundant work, silo thinking, lacking collaboration, and inconsistent agile practices [2, 3]. Also, knowledge and experience regarding agile are often lacking [2, 3]. To avoid the mentioned problems, concepts for cross-organizational communication and collaboration [2, 4], continuous improvement [2], and alignment [3] are crucial. Several scaling agile frameworks recommend Communities of Practice (CoPs) [5, 12, 14, 24], groups of people with a common interest who regularly exchange ideas, share experiences and knowledge, and learn from each other [35]. Despite extensive research in other contexts, limited empirical insights about CoPs exist in scaled agile settings. By now, researchers have mainly studied individual communities in-depth or analyzed communities in one organization, not providing a broad overview of communities established in practice. To fill this need for further research [3, 20, 21, 33], we conducted a large expert interview study investigating why organizations applying agile methods at scale establish CoPs and what characteristics these communities have. We defined the following research questions (RQs) for our study:

RQ1: What are reasons for establishing CoPs in large-scale agile environments?

RQ2: What CoPs exist in organizations applying agile methods at scale?

2 Background and Related Work

CoPs are “groups of people who share a concern, a set of problems, or a passion about a topic, and who deepen their knowledge and expertise by interacting on an ongoing basis” [35]. Those groups can exist in various contexts (e.g., organizations, governments, and education) [35]. A CoP has three crucial characteristics [35]: (1) Domain, the shared interest of its members; (2) Community, the interaction between the CoP members in the form of collaboration, support, and knowledge sharing; and (3) Practice, a shared set of experiences, knowledge, and approaches the CoP creates. According to Wenger et al. [35], CoPs can be distinguished from other group structures, like formal departments, operational teams, project teams, and informal networks, based on their purpose, members, boundaries, motivation, and lifespan. The authors argue that CoPs are intended “to create, expand, and exchange knowledge, and to develop individual capabilities.” Moreover, people join the community based on their “expertise or passion for a topic”, its boundaries are fuzzy, and members stay together due to “passion, commitment, and identification with the group and its expertise.” Finally, a CoP “evolves and ends organically”, if no longer needed. In contrast, other group structures differ in at least one of the mentioned characteristics.

Multiple researchers found that CoPs can vary in different aspects. For example, Wenger et al. [35] state that communities can differ in their scope, which can be within a business unit, across business units, and organizational boundaries [35]. Moreover, how official the CoPs are can vary from unrecognized to institutionalized, influencing the degree of organizational steering and support [35]. According to Wenger et al. [35], CoPs also differ in their size, lifetime, member distribution, homogeneity, and initiation (top-down vs. bottom-up) [35]. Jassbi et al. [9] extend these aspects, e.g., by member selection (open vs. closed), enrollment (voluntary vs. mandatory), and organizational support (high vs. low).

According to Wenger et al. [35], CoPs can have short- and long-term values for their members and the organization. Members can get help with challenges, access to expertise, feel a sense of belonging, develop personally, and create a network [35]. Likewise, Fontaine and Millen [7] report a potential to improve the working environment, job satisfaction, and collaboration. Organizations can benefit from problem-solving, time savings, knowledge sharing, identified synergies across units, and reused resources [35]. In the long term, CoPs can positively influence innovation, strategic capabilities, or new strategies [35]. According to Lesser and Storck [15] and Fontaine and Millen [7], the communities can help respond faster and improve operational efficiency.

Scaled agile frameworks, such as SAFe [24], Spotify [12], LeSS [14], or Disciplined Agile [5] recommend organizations applying agile methods at scale to establish CoPs. According to the frameworks, the communities are relevant to connect experts with specific capabilities, typically distributed over cross-functional teams, to allow them to exchange and evolve shared practices [5, 24]. CoPs can help foster self-organization [14], continuous learning [5, 24], and leverage all experiences and knowledge existing within an organization [24].

Multiple researchers have studied CoPs in large-scale agile software development. Several of these authors investigated individual communities in-depth. For example, Silva and Doss [28] report on their experience establishing a company-wide Agile Coach CoP, which helped encourage learning, mutual support, and agile methods. Sporsem et al. [31] investigated a CoP that supported internal start-ups through experience and knowledge sharing, problem-solving, and the creation of shared experiences. Detofeno et al. [1] report on the establishment of a CoP within an organization that successfully contributed to technical debt management. Moe et al. [17] and Paasivaara and Lassenius [22] studied a corporate-level CoP at Ericsson, focusing on their decision-making power. Moe et al. [17] conclude that CoPs can serve as a bottom-up alignment mechanism between teams and empower them to make decisions.

Other authors investigated the establishment of CoPs within a single organization. Kähkönen [10] studied different CoPs within Nokia and emphasized their importance to succeed in agile multi-team settings. Korbel [13] reports on his experience of establishing different CoPs within an agile program to align distributed teams and individuals and foster continuous improvement. The teams were highly involved in the creation, proposing the CoPs’ themes such as Scrum Mastering, DevOps, and UX. A few researchers provided an overview of what different CoPs they found within the organization they investigated. For example, Šmite et al. [30] investigated corporate-level communities at Ericsson, implemented to support the agile transformation and promote a participative culture. The CoPs, consisting of unit representatives, were responsible for aligning all distributed units, promoting autonomy, and sharing knowledge. According to the authors, the CoPs created value for the organization, its units, and its members. The authors analyzed the CoPs regarding their mission and scope, authority, membership characteristics, scope, and meetings. Šmite et al. [29] studied multiple CoPs at Spotify (called “Guilds”), intended to improve decision-making, enhance support, and promote collaboration and cross-organizational knowledge sharing. The authors analyzed aspects like community members, scope, distribution, meetings, size, values, challenges, and focus (e.g., learning, onboarding and support, alignment, strategy development). Moreover, Paasivaara and Lassenius [20, 21] studied CoPs at a unit of Ericsson with members located at multiple sites. The authors identified knowledge sharing and learning, coordination, technical work, and organizational development as purposes of the studied CoPs.

Uludağ et al. [32] identified Architecture CoPs applying collaborative decision-making in two organizations as part of their research on enterprise architecture in scaled agile contexts. Finally, Monte et al. [18] take an overarching view by investigating how CoPs are used to support agile teams through a literature review. The authors identified knowledge sharing, coordination, decentralized decision making, and fostering agile at scale as applications.

Almost all mentioned research endeavors studied individual or multiple CoPs within a single organization, hereby providing initial insights on why and what CoPs are established in large-scale agile software development. Still, no broad empirical overview across organizations exists.

3 Methodology

Study Design. We conducted semi-structured expert interviews to answer our RQs [8, 19, 27]. To ensure rigorous data collection, we respected the guidelines of Myers and Newman [19]. The chosen research design relies on qualitative data collection and is suitable as establishing CoPs in large-scale agile contexts is a problem in practice [27]. This study combines exploratory, descriptive, and explanatory elements. Table 1 presents an overview of the study participants. In two rounds, we interviewed 39 experts from 18 organizations. We combined convenience and purposive sampling [11]. Potential interview partners were contacted directly, e.g., via e-mail, and we shared our research endeavor with existing contacts and networks. Still, the selection was intentional to interview experts who work in scaled agile organizations or contexts [4], having CoPs established. We only interviewed experts involved in the communities as leads, members, or stakeholders (e.g., managers supporting CoPs). Some experts are involved in multiple CoPs in different roles. The interviewees have various job roles and industry backgrounds, leading to a “variety of voices” covering multiple viewpoints [19].

Table 1. Interview partners
Full size table

Data Collection. We conducted the first interview round between February and May 2023 (23 experts, 13 organizations) and the second (16 experts, five organizations) between November 2023 and January 2024. The majority of interviews had a duration of 40–60 minutes. Each interview started with an introduction to ensure a shared understanding of relevant concepts. All interviews had the same outline, with slight changes after the first round, to improve the data collection based on the experience gained and to collect data about challenges and best practices. We divided the interviews into three consecutive phases: (1) questions about the interviewees’ backgrounds (e.g., experience in large-scale agile settings*), (2) questions about CoPs within their organization, (3) questions on how research can support practice in establishing CoPs in the first round and regarding challenges in the second round. The questions of the last two sections were open, allowing our interview partners to share their thoughts in detail. We conducted all interviews using videoconferencing tools. All interviews, besides one, were recorded and subsequently transcribed. Besides the interviews, we included relevant data sources (e.g., websites) shared by the interviewees to facilitate the data triangulation.

Data Analysis. The results presented are a subset of the collected data. We plan to communicate our other insights in later publications. We coded and analyzed the collected data based on the guidelines by Miles et al. [16] and Saldaña [25], with a two-cycle approach. First, we used inductive coding and summarized meaningful data fragments with descriptive codes. Then, these codes were assigned to second-cycle codes of two abstraction levels: (1) a set of high-level codes developed deductively based on the interview design and relevant concepts (2) lower-level codes developed inductively while analyzing the first-cycle codes, reflecting discovered concepts and patterns. In case of uncertainties or missing information, we contacted the interviewees. We included all group structures fulfilling the CoP definition [35], independent of their naming.

4 Results

In the following, we present the results of our interview study.

4.1 Context

Table 2 provides an overview of the experts’ development organizations, which do not necessarily equal the whole IT department of the experts’ companies.

4.2 Reasons for Establishing CoPs

We identified various reasons for establishing CoPs, summarized in Table 3, sorted by frequency. In the following, we present the most common reasons.

Promoting Knowledge Sharing. All CoPs we studied were established to foster knowledge sharing regarding different topics or a specific craft. The communities are intended to break down knowledge silos and allow benefiting from the knowledge and experience of others. For example, E15 explains that the Agile Coach CoP within FoodCo was founded as the Agile Coaches “need some way of distributing experiences, [so] not everybody needs to learn from the same mistakes by themselves. But we want to create a shared pool of knowledge.”

Promoting Collaboration. Another common reason for establishing communities is to promote collaboration between employees and different organizational areas. E18 highlights that the communities are “not only [for] exchanging ideas but really getting support from others and working on topics together.” The members support each other, build knowledge, and collaborate to create results and solutions. For example, within ConsultProj’s Scrum Master CoP, the members collaboratively develop “common values, [...] plan suggestions, improvements that might be a bit more sound than if you just asked single individuals” (E6).

Table 2. Development organization of the interview partners
Full size table

Fostering People Development. Many organizations establish CoPs to foster the development of their employees. CoP members can learn and improve their skills together, and new employees can benefit from the knowledge of experienced members. E11 highlights that with CoPs “people can learn faster, [...] become better in their job because they know the right people to talk to, [...] to learn from.”

Improving Efficiency. Another motive for establishing CoPs is to improve efficiency. The communities can help identify synergies, solve common challenges, and reuse, e.g., solutions. At SoftwareCo2, for instance, a technology-specific CoP helps developers to be aware of what “other teams are doing with this technology so that whether we can reuse it or we can benefit from this” (E5).

Fostering Networking. Another purpose of CoPs is to promote networking, as they can help their members meet people interested in a specific topic or with a particular role working in other areas. For example, E16 explains that the Scrum Master CoP within SoftwareCo2’s commerce area helps “to meet each other and to know what kind of Scrum Masters [...] we have in our organization.”

Distributing Information. Another common reason for establishing CoPs is distributing information to a target group. E14 highlights that the communities can help “to get information from other areas [...] so that you can also create a bigger picture, a clearer picture about the things which are going on in the company.” Within SoftwareCo2, e.g., a CoP for Scrum Masters in commerce helps communicate information relevant to all employees in this role. According to E16, the “CoP is a very crucial [...] place where we can communicate changes.”

Fostering Empowerment. Many CoPs are established to empower employees to act self-organized, take responsibility, make decisions, and influence actively. For example, E22 argues that the communities help “avoiding discussions between people who are not really involved in the topics [...], have to feel the pain about their decision.” According to E11, communities are empowering employees as they “give them the platform also to involve themselves.”

Aligning Across the Organization. Within several organizations, CoPs were established to align teams, areas, and specific roles across the organization. For example, for the ConsultProj, communities are needed “to get people aligned” as the project is “very large”, spanning “different geographical areas, [..] time zones, [...] [and] various companies” (E6). Likewise, E7 highlights that “people with a certain task, [...] should align horizontally over the organization.’’

Supporting the Agile Transformation. Some CoPs we investigated were established to support transitioning to agile. The communities can help to spread and create knowledge and experience regarding agile practices. Moreover, E29 highlights CoPs themselves “bring agility into the organization.” Within SoftwareCo2, e.g., a central Scrum Master CoP was formed as “there was the need then to offer a capability to exchange among each other because Scrum master is really a special role compared to a traditional project manager” (E3).

4.3 Characteristics of the Established CoPs

In the following, we present the characteristics of the CoPs we investigated, without considering how they are organized internally (e.g., meeting frequency).

Table 3. Reasons for establishing CoPs
Full size table

Initiation. In most organizations we investigated, CoPs initiated top-down by a management decision, and CoPs created bottom-up by employees exist. In a few cases, we only found communities initiated bottom-up (CarCo2, TeleCo1) or top-down (ConsultCo1, ConsultProj, HealthCo). The driver for initiating CoPs can change over time. Initially, in FoodCo, communities for agile topics and roles were started by individuals who saw a need. Then, management recognized the value of CoPs and initiated them for non-agile themes like security. In ConsultCo2 and ElectroCo, management initiated the first CoPs, and employees later followed. Overall, slightly more than half of the CoPs we found were top-down-initiated. Most communities created to empower employees, establish safety, and improve offerings were initiated top-down. CoPs for new trends were mainly initiated bottom-up. Employees of one participating firm initiated the cross-company CoPs.

Target Group. Most organizations we studied have CoPs for specific roles (role-based) and independent of any role (topic-based). A few organizations have mainly one type (e.g., SoftwareCo1, ConsultCo1, ConsultProj, TeleCo1 &2). Role-based CoPs are often open to others, e.g., for interested employees.

Themes. We identified various themes for which CoPs are established (see Table 4). The interviewees reported at least one community around Agility for almost every organization. Most of these communities are role-based, e.g., for Scrum Masters. Also, the cross-company CoPs are about agility. Another common theme is Architecture. Most Architecture CoPs are role-based. Many organizations also have communities for Software Development, some focusing on a particular technology or method, such as Citizen Development (TransportCo). Moreover, several organizations have, mainly top-down initiated, communities for classical and agile Management and Leadership roles (e.g., Disciplinary Leadership CoP (CarCo1, TeleCo2)) and Security. Most CoPs around Product Management are role-based PO CoPs. In addition, several organizations have mainly top-down initiated communities dealing with Testing, UI/UX, and Quality, e.g., software quality. Some organizations established, mainly topic-based, CoPs for Cloud and Transformation and Change. The latter include, e.g., cultural (CarCo1) and agile transformation (FashionCo). Finally, some communities for Business Analysis and Digitialization and Innovation exist, for example, for digitalizing transportation at TransportCo. Next to the presented themes, we found others implemented by only a few organizations, e.g., AI, or DevOps.

Table 4. CoP themes
Full size table

Openness. In many expert organizations, CoPs are open to everyone. Still, in multiple organizations, at least a few closed communities exist (ConsultProj, ConsultCo2, FashionCo, FoodCo, InsureCo1 &2, RetailCo, SoftwareCo2, TeleCo2, TransportCo). These closed communities are mostly role-based or, if topic-based, initiated top-down. One studied cross-company CoP is closed.

Organizational Acceptance. Top-down-initiated CoPs are, by nature, accepted by management. Also, the bottom-up established communities we investigated are tolerated by the management, either due to official approval or as employees generally have the freedom to establish them. A few CoPs in some companies are even led by managers (e.g., ConsultCo1, ConsultProj, RetailCo, SoftwareCo1 &2, TeleCo2). For example, the PO CoP within RetailCo is led by the Head of Product. We only found two CoPs of which management is unaware: a local Scrum Master CoP at SoftwareCo2 and the cross-company CoP of E37. Still, the companies’ culture gives employees enough freedom to engage in such an exchange. In ConsultCo2 and FoodCo, establishing a CoP is formalized. For example, FoodCo has a “form which needs to be filled [...] to create a CoP” (E15).

Decision-Making Power. In some organizations, all CoPs we investigated can either make decisions affecting their area of expertise or influence them (CarCo1, ConsultCo1 &2, ConsultProj, FoodCo, HealthCo, InsureCo1 &2, SoftwareCo1, TeleCo1 &2). In many other organizations, the decision-making power depends on the community (CarCo2, FashionCo, RetailCo, SoftwareCo2, TransportCo). Overall, most communities we studied have decision-making power or can influence decisions. CoPs established to foster empowerment largely have the power to at least influence decisions. The communities without the power to make decisions were mainly initiated bottom-up. Community decisions are often agreements, best practices, guidelines, or standards. For example, CoPs within ConsultCo1 are “working on a guideline on how certain roles should behave in a project. [...] It’s not forcing any projects to apply this, but it’s definitely something that will [have] an influence” (E18). The cross-company CoPs we studied are not entitled to make any decisions affecting the involved organizations.

Organizational Support. While the time employees spend within CoPs is funded, most communities we studied have no dedicated budget. Only within a few organizations (some) CoPs have a budget or sponsor funding, e.g., external guest speakers (CarCo2, FoodCo, InsureCo1, RetailCo, TeleCo1, TransportCo). A minority of CoPs have full-time leads. Often, Agile Coaches or a team support (ConsultProj, ElectroCo, FoodCo, HealthCo, InsureCo1, RetailCo, TeleCo2) and management promotes (some) CoPs actively (ConsultProj, ConsultCo2, FashionCo, HealthCo, InsureCo1, RetailCo, SoftwareCo1 &2, TeleCo2, TransportCo).

Steering. In many organizations, at least some CoPs we investigated are monitored and steered by management to a certain degree, e.g., regarding content and strategy (CarCo1 &2, ConsultCo1 &2, ConsultProj, FashionCo, FoodCo, InsureCo1, RetailCo, SoftwareCo1 &2, TeleCo2, TransportCo). More top-down than bottom-up initiated communities belong to this group.

Participation. In most experts’ organizations, participation in CoPs is voluntary or sometimes expected. Some companies additionally have a few mandatory communities (CarCo2, FashionCo, InsureCo1, SoftwareCo2). These CoPs are officially approved or initiated, and often steered by management. Also, these CoPs mostly have access to funding and decision-making power. The cross-company CoPs are voluntary.

Scope. Within some organizations we only found CoPs spanning across the whole development organization (ConsultCo2, ConsultProj, TeleCo1 &2, SoftwareCo1). These development organizations often have no “team of teams” structure grouping teams, e.g., into ARTs. Other organizations with “teams of teams” or even more complex structures mostly have communities on additional levels. Some organizations have CoPs restricted to a single or a few “team of teams” (CarCo1, FashionCo, FoodCo, HealthCo, InsureCo1 &2, RetailCo, SoftwareCo2, TransportCo), for representatives of each “team of teams”, or for roles with leadership responsibility, e.g., for a “team of teams,” mainly with decision-making power (ConsultCo1, ElectroCo, FoodCo, InsureCo1, RetailCo, SoftwareCo2, TransportCo). For example, InsureCo1 has a community for its Tribe Leads. The most common scope of CoPs we found is the overall development organization. Especially topic-based communities tend to have a broad scope. Moreover, some companies (e.g., CarCo2, TransportCo) have CoPs covering the whole organization, not just IT. The cross-company CoPs go beyond a single organization. Almost all CoPs we investigated have members from different sites in Germany or other countries. The only single-site communities we found were all communities of TeleCo1 and InsureCo2, and two Scrum Master CoPs and a PO CoP at SoftwareCo2.

Size. The size of the CoPs we investigated ranges from less than ten members to several hundred. Naturally, the organization’s size plays a role. For example, CoPs within SoftwareCo1, a small company, could not reach hundreds of members. The topic-based and open communities we studied are often bigger than role-based and closed ones. Likewise, the CoPs with a broader scope are in many cases bigger than ones, e.g., for a single “team of teams”. The CoPs we found intended to be a safe space for employees tend to have less than 50 members.

5 Discussion

In this section, we discuss our key findings and the limitations of our study.

5.1 Key Findings

Our study identified promoting knowledge sharing as the most common reason for establishing CoPs. Likewise, Paasivaara and Lassenius [21] and Šmite et al. [30] found knowledge sharing as at least an implicit goal behind all communities in the scaled agile organizations they studied. As in both studies, the CoPs we investigated have various purposes. The most common reasons are to enhance collaboration, people development, efficiency, and networking. These findings align with the values CoPs can provide according to Wenger et al. [35] and the results of other authors studying communities in scaled agile settings [17, 21, 28,29,30,31]. Moreover, we identified promoting empowerment, alignment and the agile transformation as common motives. Likewise, Moe et al. [17] found that CoPs can empower through community-based decision-making. Other authors report on CoPs intended to align distributed units and teams within scaled agile settings [13, 29, 30] and conclude that CoPs can support the agile transformation [18, 21, 30].

The communities we studied cover a variety of themes not specific to agile methods. Besides CoPs for agile roles or about agility, we identified numerous ones for themes like architecture or security. Likewise, other related studies mainly found CoPs for themes not specific to agile practices [21, 29, 30]. While both topic- and role-based communities were common in our study, some themes (e.g., architecture) are mainly addressed in role-based CoPs, and others (e.g., transformation and change) are the subject of topic-based CoPs.

Our study shows that in most organizations, CoPs are initiated top-down and bottom-up. Communities for several themes, like quality, were primarily initiated by managers. The same applies to CoPs created for specific reasons, like empowerment. CoPs focusing on new trends were mainly created bottom-up. We also found that the drivers for CoPs can change over time, e.g., if management or individuals start recognizing their value. Likewise, Paasivaara and Lassenius [21] report a change in the initiation from managers to those who needed them.

Most communities we investigated are open and voluntary. In line with our findings, other studies found mainly open [29] and voluntary communities [21, 30]. According to several authors [21, 35], openness is crucial. Moreover, Wenger et al. [35] see mandatory participation as a weakness, whereas Šmite et al. [30] found that it can ensure all organizational areas are represented.

The CoPs in our study are accepted by management due to top-down initiation, approval, or the overall organizational culture. While the time spent in communities is funded, most lack a dedicated budget, regardless of whether they are initiated top-down. Similarly, in other studies [21, 29, 30], CoPs are officially recognized. Only Šmite et al. [29] report a dedicated CoP budget. Most CoPs we studied, particularly if established to foster empowerment, can make or influence decisions in their areas of expertise. Several authors (e.g., [21, 30]) investigated CoPs entitled to make decisions or give recommendations and concluded that this authority is needed. Our study also showed that management often steers communities, particularly if initiated top-down. Likewise, Wenger et al. [35] connect institutionalization with potential overmanagement.

In our study, CoPs most often span the whole development organization, especially if topic-based. In addition, most organizations with a “team of teams” structure have CoPs on more levels. The scopes we found align with the categories Wenger et al. [35] and the scopes in agile organizations other researchers describe [21, 29, 30]. As in these studies, the CoPs we studied mainly cover different sites. The size of the communities we investigated ranges from less than 10 to several hundred members. Topic-based and open CoPs tend to have more members than role-based and closed ones, potentially due to a broader target group. Same applies to CoPs spanning the whole development organization. A similar pattern can be found when comparing studies of CoPs with different scopes ([29] vs. [30]).

5.2 Limitations

We used the assessment schema of Runeson and Höst [23] to evaluate threats to our study’s validity. We addressed the threat of construct validity by clarifying relevant concepts and potential ambiguities during each interview. To overcome the threat of external validity, also influenced by our sampling strategy, we interviewed 39 experts with different roles from 18 organizations in different industries to increase generalizability. For achieving reliability, two researchers were involved in the study process, and we followed guidelines for the data collection and analysis [16, 19, 25]. Whenever the interpretation of data needed clarification, we contacted the interviewees. We tried to address threats to internal validity by following the same outline to collect data and building on patterns identified between the organizations and communities we studied. Still, quantitative research is needed to validate the causal implications.

6 Conclusion and Future Work

Organizations applying agile methods at scale have to cope with increased complexity and challenges like redundant work, silo thinking, lacking collaboration, and inconsistent agile practices [2, 3]. Existing research provides first insights into how CoPs can address these issues and support organizations applying agile methods at scale [21]. However, a broad overview of why large agile organizations establish CoPs and what these communities look like is missing. To fill this gap, we conducted an expert interview study comprising 39 experts from 18 organizations. We found that next to benefiting from the known advantages of CoPs, organizations applying agile methods at scale establish them to foster empowerment, strengthen alignment, and drive their agile transformation. Moreover, CoPs focus not only on agile but also on classical themes such as architecture. Most communities are open and voluntary but not necessarily established bottom-up. Often, management initiates CoPs, e.g., to empower their employees, giving them the authority to make decisions in their area of expertise. CoPs are either explicitly or implicitly accepted by management, as the organizational culture provides the freedom for such, and often steered, especially if they are top-down initiated. Our study provides insights into CoPs in large-scale agile software development, which can inspire practitioners and allow future research. We aim to publish the remaining evidence on how communities are organized and identified barriers, challenges, and best practices. We also want to use quantitative methods to validate and extend our results. Moreover, building on our findings on why and what CoPs are established in scaled agile settings by enhancing them with additional empirical insights could allow for a well-founded taxonomy of CoPs in this context. Furthermore, we presented why organizations have established CoPs but did not provide insights into their success. Future research could, for example, focus on appropriate indicators or measurement approaches that allow organizations to evaluate the success and value of CoPs within scaled agile settings.