This work-in-progress paper explores how large industrial organizations can enable organizational ambidexterity through embedded innovation programs. Drawing on the case of the AI BooSTcamp at a large multinational microelectronics company, we examine an embedded innovation program designed to support exploration within an exploitation-dominated environment. Refreshed annually by new cohorts of junior participants, the program operates as an emergent and adaptive socio-technical infrastructure, grounded in informal routines, symbolic framing, and distributed autonomy. Based on ongoing qualitative eldwork, we analyze the symbolic and social microfoundations that foster local conditions for exploratory learning through evolving coordination mechanisms and team dynamics. We show how such bottom-up, socially driven structures can sustain exploration over time. Future work can look at the capacity to scale and inuence broader organizational processes.
2 STMicroelectronics
1. Introduction
Organizational ambidexterity (OA), or the capacity to simultaneously exploit existing capabilities
while exploring new opportunities, is widely recognized as critical for rms navigating fast-paced
technological and market changes [1; 2]. Yet, operationalizing OA in large industrial organizations
remains a persistent challenge. The immediacy of operational demands and challenges oen
overshadows exploratory initiatives, leading exploration to lag behind exploitation [
In response, researchers have increasingly examined the microfoundations of ambidexterity, seeking to understand how individual characteristics, team dynamics, and project-level routines contribute to the emergence and maintenance of ambidextrous behavior [6; 7]. This perspective sheds light on how everyday work practices, autonomy, and social processes shape organizational outcomes. Alongside this, complexity theory has emphasized the value of emergent, self-organizing systems that support adaptability through informal coordination and locally situated learning [8; 9].
Despite these developments, the role of symbolic and cultural elements in enabling ambidexterity is underexplored. Existing studies oen focus on structural or digital supports, while overlooking how meaning-making, rituals, and shared values contribute to balancing exploration and exploitation.
This work-in-progress study contributes to this eld through an in-depth investigation of the AI BooSTcamp at a large multinational in the microelectronics sector. Existing project management frameworks, such as Scrum, SAFe, and Design Thinking, have proven insucient for the organization to fully harness the innovative potential of both experienced operational personnel and technologically adept newcomers. Each year, new cohorts of PhD students and interns join the program to identify, co-design, and implement AI projects in collaboration with operational teams who oen remain within the organization for only a few months to a few years. Designed to support innovation in articial intelligence, the AI BooSTcamp operates without relying on formal digital workplace for co-creation. Instead, it functions as a temporary, recongurable system based on symbolic framing, distributed roles, and recurring interpersonal routines. Hence, this study tackles the following research question: How do micro-level structures, routines, and symbolic elements within innovation programs enable OA in large industrial organizations?
To address this question, we draw on qualitative eldwork, including ve exploratory interviews with BooSTcamp participants. We examine how the program creates the conditions for exploratory learning within a highly operational environment. We focus in particular on how autonomy, coaching, and symbolic infrastructure contribute to enabling innovation from within, rather than through top-down initiatives.
We oer two main contributions. First, we extend research on the microfoundations of OA by showing how temporary, rotating teams, informal mechanisms, and distributed coordination, shaped by a holacratic governance model, enable sustained exploration within an exploitation-dominated industrial context. Second, we contribute to the literature on symbolic infrastructure by analyzing how values such as radical candor and openness to doubt shape participants’ perceptions of agency, legitimacy, and risk-taking. Together, these ndings highlight how embedded, socially-driven innovation systems can operate as complex adaptive systems (CAS), and raise important questions about the conditions under which such models can scale and sustain impact.
2.1. Organizational Ambidexterity and its Microfoundations
OA refers to a rm’s capacity to simultaneously exploit existing capabilities while exploring new
opportunities [10; 11]. It is widely recognized as essential for sustained innovation and long-term
performance in dynamic environments [12; 1]. Early work in this eld distinguished between
structural ambidexterity, where separate units are dedicated to exploration and exploitation, and
contextual ambidexterity, where individuals or teams dynamically shi between both activities within
the same structure [
The microfoundations perspective emphasizes three key domains: (a) individual characteristics,
such as cognitive exibility or motivation [14]; (b) team processes and structures, including diversity,
autonomy, and coordination routines [15; 7]; and (c) cultural and symbolic elements, such as shared
values, rituals, or frames that guide behaviour [16; 17]. They are a multilevel phenomenon, so that an
analysis of each level and their interaction is needed to understand capabilities development.
Developing microfoundations is not without challenges [
Despite growing interest in these domains, several authors highlight gaps in the literature. For
instance, [
While much of the OA literature has focused on formal structures and role-based mechanisms, an
emerging body of work grounded in complexity theory oers a dynamic and processual perspective.
Organizations are increasingly seen as CAS, composed of interdependent agents whose local
interactions can generate nonlinear and emergent outcomes [20; 8]. From this viewpoint,
ambidexterity is not achieved through static design or strict dual structures, but through enabling
conditions that support continuous interaction, experimentation, and adaptation. CAS may benet
from developing adaptive capacities, which describes the ability of an organization, to respond to
change, shocks, or uncertainty [
Scholars working within this tradition emphasize that relational dynamics, symbolic framing, and
exible coordination mechanisms play a critical role in enabling ambidexterity. Cultural elements
such as shared values, language, and rituals help legitimize uncertainty and experimentation [22; 23],
while psychological safety and interpersonal trust are seen as foundational for risk-taking and
learning [
This study takes place within a large microelectronics company that has been operating for over 35 years. While the organization is multi-national and employs over 40,000 people worldwide, the study was performed at one of its locations in France with about 5,000 employees. The organization currently faces several challenges: (1) an aging workforce with expertise that is scarce in the job market, specically in articial intelligence; (2) strong pressure for productivity gains; (3) a highly competitive global market putting constant pressure on time-to-market. Leveraging AI to develop new microchips and products, or to improve existing oerings, has emerged as a possible response to these challenges. However, due to a strong historical focus on operational excellence and limited internal AI experience, using AI for innovation has oen been looked upon with caution or outright skepticism. When the Research & Development (R&D) department attempts to push innovation toward operational teams, it can also encounter signicant resistance, as is common in organizations lacking ambidexterity.
To address these issues, the AI BooSTcamp was launched within the R&D department on one of the French sites. It is a dedicated program designed to foster AI-driven innovation while remaining closely aligned with the organization’s operational context. The program’s founder explicitly aimed to create the conditions for OA, enabling both exploration and exploitation to occur simultaneously, with the goal of generating AI innovations that can be continuously injected into operations.
The AI BooSTcamp manager works in close collaboration with operational teams to identify promising AI-related opportunities across business units. Rather than imposing innovation from above, this approach encourages the co-construction of projects between juniors and local teams, helping to ensure contextual relevance and reduce resistance to change by embedding innovation within existing workows and needs. Once a project is selected, the manager assembles temporary project teams composed of PhD students and interns. PhD students are fully embedded in the organization and typically oversee several projects over a multi-year period. They work alongside master’s-level interns, who join the AI BooSTcamp between January and August and are replaced annually. This structure is designed to infuse the organization with the fresh perspectives and technical expertise of younger generations, and to enable a form of innovation “injection” into operational departments, which act as internal clients.
The AI BooSTcamp founder has built the program around several guiding principles as follows.
First, the program draws on holacratic principles. Holacracy [
Second, it promotes radical candor [
Third, the program emphasizes intergenerational collaboration by bringing together the up-todate technical skills and fresh perspectives of juniors with the domain expertise and contextual knowledge of more senior employees. This cross-generational synergy should enable mutual learning and help bridge the gap between exploratory innovation and operational realities. Such intergenerational collaboration has been shown to enhance team functioning by combining diverse experiences and perspectives, fostering both task eectiveness and relational integration.
To understand how these principles are enacted and experienced in practice, and how they
contribute to the microfoundations of ambidexterity, we conducted a rst round of qualitative
interviews. We followed an inductive approach and performed thematic analysis to analyze data [
4.1. Holacracy in action: The Power of Informal Routines
The AI BooSTcamp is structured around a set of recurring rituals designed to foster coordination, visibility, and legitimacy for exploratory work. These include a Monday morning coee, where AI topics are discussed informally; two weekly meetings, which serve as coordination checkpoints for sharing progress within projects and occasionally a manager; monthly innovation reviews with the AI BooSTcamp members and internal clients, which expose projects to a broader internal audience; and a yearly innovation board, where selected outcomes are showcased to senior leadership. These rituals form the symbolic backbone of the program and are intended to promote communication, transparency, and a sense of shared purpose. They are a must-have in holacratic organizations.
Despite the formal meeting points described above, interviews suggest that the most authentic moments of exploratory behavior and learning happen in the informal spaces between these scheduled events. Peer-to-peer feedback oen occurs spontaneously, enabled by a strong sense of proximity and psychological safety among participants. BooSTcamp members share an open-plan workspace, with no cubicles or enclosed oces, and regularly socialize outside of work hours. This environment facilitates candid exchanges, quick alignment, and emotional support — forms of interaction that participants describe as more meaningful and generative than formal meetings.
As such, rituals like the weekly coordination meetings can be experienced by participants as either performative or constraining. This can lead the group to want to reorganize the holacracy itself. One participant recalled a collective eort to negotiate a reduction in the number of weekly meetings. This request was ultimately refused: “We had tried to negotiate for what they call ‘free weeks’ every other week — when we felt we didn’t have much to report, just a short note at the end of the week to say, ‘I’m working on this, I’ll have results next week.’ We all agreed on this. But it was rejected because [the managers] felt those meetings had a really positive impact. From our side, we felt exactly the opposite.” This quote highlights a critical tension: while the symbolic infrastructure emphasizes atness, selforganization within the holacratic circles, radical candor and autonomy, participants do not always feel empowered to reshape the very rituals intended to support them. The minimal structure required from a holacratic system does not fully prevent the perception of a traditional hierarchy. For instance, in the absence of job titles, the AI BooSTcamp members refer to each other according to their contractual status, with some being very short-term, like interns, others in a mid-term perspective, like the PhDs, and nally the permanent sta. The holacratic circles are active mostly between the non-permanent sta thanks to the informal social routines and sense of proximity exposed above. Holacracy also involves structure, whose value is not always perceived by short-term sta. 4.2.
While respondents do describe traditional hierarchies, with interns reporting to PhD students and then to the AI BooSTcamp management, they also describe how they learn from one another on a project, as well as across projects. An intern emphasizes the importance of this approach in giving them experience. Another participant notes that it is not always possible to learn what the others are doing, but still it is possible to get an idea of dierent projects and help sometimes. This mode of functioning, with knowledge circulating within a project, and across projects, in circular manners, is typical of a holacratic organization. It leads interns to give feedback to each other, sometimes it also leads them to take on new roles in coaching or helping others. An intern notes that the most important and unexpected learning outcome from his internship is communication. A participant describes the AI BooSTcamp as a “training center”, where so skills can be learned, and experience can be obtained by seeing several projects. Several other participants consider communication, to identify and solve issues, as crucial in the AI BooSTcamp. Despite their early-career status, participants who do their internship do not passively absorb feedback. While they value guidance from PhD students or managers with recognized AI expertise, they tend to discount input from those perceived as lacking technical depth, who are typically outside of the holacratic circles.
4.3.
Radical Candor, AI BooSTcamp Values, and Holacracy Lead to Strong So Skills Development
Participants describe decisions on projects as possibly made either by the intern alone, or with the PhD student, or with the permanent sta during meetings – albeit less frequently. This conviction of one’s ability to decide and solve issues strengthens a high sense of autonomy, as well as perseverance. Several interviewees described polite forms of resistance, in which they appear to follow advice while quietly continuing with their preferred direction: “They can tell us to do something, and so we’ll kind of do it in the background — but we’ll still pursue our idea. Later we’ll come up with some results to make them happy, so they’ll leave us alone and we can keep working on what we think is a better direction.” This form of subtle resistance does not indicate disengagement but reects a growing sense of technical judgment and ownership. Perseverance is oen described as a good quality for researchers.
Participants also recognize that autonomy is not static, but rather develops through experience, maturity, and project progression. As one stated, “At the beginning, it was really hierarchical, but I think that depends on maturity. At my stage now, it’s much atter. I’m the one proposing things. Sometimes I even push back on management’s ideas because they aren’t technically feasible.” This highlights ways in which changing roles and radical candor can be expressed.
Participants also demonstrate they are not afraid of failure and that it is part of the process. “We know we will hit the wall sometimes, and it’s okay. It’s better to do it at the beginning though, so I can produce something in the last leg of my internship.” Some also highlight that they are encouraged to challenge everything. Acting in this way may be dependent on one’s personality. A participant notes “We are told to doubt everything and I do it, but it is in my nature”. Personality plays an important role, as another participant mentions how important it is to be curious to learn from the others and move forward, and that it is in his personality.
This evolving relationship to autonomy and one’s growing ability to explore is shaped not only by individual condence but also by the maturity of the project itself, from ‘sensing’ to ‘seizing’ and ‘transforming’ phases. Participants must work with the legacy of previous interns, including unnished ideas and abandoned paths. As one explained, “ We actually documented more the paths we abandoned than the ones we developed. Because you have to justify why you’re not going down a certain road.” Exploration is embedded in a collective process of historical layering, where value lies as much in learning what does not work as in delivering a nished output.
4.4.
Injecting Innovation Requires Building Bridges Outside of the Holacracy The AI BooSTcamp exists somewhat apart from the rest of the company. The monthly meetings allow exchanging knowledge across projects with the aim of fostering situational awareness and enabling expertise location. Despite the program's internal coherence and strong exploratory capacity, interviews highlight a limited organizational reach. Participants report few sustained interactions with other teams beyond project handovers. One participant lamented, “Even if we do monthly meetings, they’re not really meant to explain things. It’s more like marketing, to sell the projects, but we don’t really go into the technical details.” Another explained the lack of uptake more bluntly: “In my case, they just take it as a black box, so there’s no investment in understanding how it works to reuse it... they almost never come back to me, so I know they use it very little.” These observations point to the challenges of transferring innovation from temporary, exploratory units into stable organizational routines. Knowledge does not travel automatically, and symbolic visibility through rituals is not enough to ensure adoption. Without stronger relational bridges, innovation risks remaining siloed within the exploratory space. However, this insight may stem from the limited perspective of interns whose short-term presence in the organization may prevent them from grasping how injection works.
The AI BooSTcamp exists somewhat apart from the rest of the company. The monthly meetings allow exchanging knowledge across projects with the aim of fostering situational awareness and enabling expertise location. Despite the program's internal coherence and strong exploratory capacity, interviews highlight a limited organizational reach. Participants report few sustained interactions with other teams beyond project handovers. One participant lamented, “Even if we do monthly meetings, they’re not really meant to explain things. It’s more like marketing, to sell the projects, but we don’t really go into the technical details.” Another explained the lack of uptake more bluntly: “In my case, they just take it as a black box, so there’s no investment in understanding how it works to reuse it... they almost never come back to me, so I know they use it very little.” These observations point to the challenges of transferring innovation from temporary, exploratory units into stable organizational routines. Knowledge does not travel automatically, and symbolic visibility through rituals is not enough to ensure adoption. Without stronger relational bridges, innovation risks remaining siloed within the exploratory space. However, this insight may stem from the limited perspective of interns whose short-term presence in the organization may prevent them from grasping how injection works.
This study sheds light on how large industrial organizations can operationalize organizational ambidexterity (OA) by embedding exploratory innovation within operational contexts. The AI BooSTcamp oers an illustrative case of a temporary, non-technocentric infrastructure that enables innovation. Our study explores the microfoundations enabling exploration, summarized in Table 1.
Micro Level
Mechanism
Empirical Illustration
Role in Sustaining Exploration Individual Team
Reasoned perseverance
Evolving autonomy
Personal alignment values
Peer-to-peer coaching feedback
&
Emotional support & candid exchange
“We kind of do what they say… but keep working on what we think is a better direction.”
Ensures continuity of intent and supports local agency in CAS.
“At rst it was very hierarchical… now I propose things and even push back.”
Builds initiative and judgment, aligning with contextual ambidexterity and adaptive capacity. with
“We’re told to doubt everything —and I do, but that’s also in my nature.”
Reinforces norms of critical inquiry and symbolic framing essential for exploration.
“We do learn from others; they Fosters mutual learning and emergent show us what they're doing, and as coordination in self-organizing systems a result, we learn about AI”
“We stay among ourselves …, we make friends, [...] there's a very good sense of cohesion. It's a pleasure to
Maintains psychological safety—key for adaptive capacity and risk-taking.
Structure
So hierarchy & role uidity
Symbolic rituals lightweight routines
Legacy continuity mechanisms
Selective boundary bridging & go to the bar.”
“At rst, you try to understand how it works [...] as time goes on, you take on more or less responsibility.”
“We have what we call weekly meetings every week and monthly meetings, which are called Innovation Review, and social time aerwards”
Enables distributed agency and uid role reconguration, supporting exploratory behavior.
Uses minimal structure and loose coupling to enable emergence and protect exploration from exploitation “It’s all these little things, but no Supports continuity and informal one codes the same way [...] some knowledge transfer across cycles. people are more or less rigorous in how they describe what they do.”
“…monthly meetings, they’re not meant to be explanatory.
They’re like marketing meetings […] we don’t go into the technical details”
Protects exploration from exploitation constraints; exemplies loose coupling typical of OA in CAS
These three levels of microfoundations unfold within a holacratic environment. Holacracy is not
a xed structure, but a shaping force, i.e., an intentional design choice that interacts with evolving
practices [
5.1.
Holacracy as Governance Backbone for Dynamic Microfoundations
The holacracy implemented in the AI BooSTcamp reects the core features described in the
literature [
Newcomers or those perceived as less technically expert may struggle to assert agency, leading to
uneven experiences of radical candor and shared authority. This friction between aspirational
values and lived practice echoes tensions in symbolic infrastructure and cultural control [
5.2.
The AI BooSTcamp is underpinned by a set of symbolic elements—rituals, mantras, and shared
values—that shape how participants perceive legitimacy, agency, and risk. Values such as radical
candor, openness to doubt, and the right to fail are promoted explicitly and are embedded in the
program’s recurring routines, including Monday coees, innovation reviews, and informal exchanges.
These rituals create a shared sense of purpose and act as cultural anchors in an otherwise uid
structure, enabling a form of control based on norms and meaning rather than hierarchy. Yet these
symbolic elements do not operate uniformly. While some participants experience them as
empowering, others—especially younger interns—report a perceived gap between espoused values
and actual responsiveness. They are encouraged to speak up, but do not always feel their voices carry
weight in decision-making. This disconnect signals a tension between cultural ideals and structural
realities, echoing the observation that symbolic infrastructures oen coexist with tacit constraints. In
the AI BooSTcamp, where teams rotate annually and roles are temporary, this tension may be
amplied by the liminal nature of participants’ status: neither fully integrated into the organization
nor fully external to it [
This means that the symbolic elements must be continuously reinterpreted and renegotiated. As each cohort cycles through, it inherits values that are translated, sometimes reinforced, sometimes challenged. This dynamic reinforces the need for active maintenance of the symbolic layer, especially when formal structures are minimal. In this sense, symbolic elements become a governance mechanism for ambidexterity.
Importantly, the relative absence of technological tools in participants’ discourse reinforces the symbolic distinctiveness of the AI BooSTcamp. While oriented toward AI innovation, the program is not dened by AI methods, platforms, or toolkits. Instead, participants emphasize the values, interactions, and rituals that give exploratory work its meaning. This absence reects a deeper logic of non-technocentric innovation, where exploration is enabled by human dynamics. 5.3.
Despite the rotating members in the AI BooSTcamp, the program sustains a surprising degree of
knowledge continuity across cohorts. Interns inherit unnished codebases, stories of failed attempts,
and design choices made by their predecessors. According to interns, most of this transfer occurs not
through formal documentation, but through lightweight routines such as informal coaching,
storytelling, and embodied engagement with the technological artifact. Participants must “absorb”
prior knowledge not only in documented code, but through the cultural traces le in rituals,
comments, or social cues. Consequently, these microfoundations foster knowledge absorption [
This mode of learning aligns with the notion of lightweight infrastructure and extends it to project-based, exploratory contexts. Rather than codied procedures, continuity relies on shared space, informal mentorship, and symbolic rituals. These mechanisms act as cultural carriers, embedding knowledge in relationships and routines. The AI BooSTcamp’s reliance on informal mechanisms challenges traditional views of project handover and points to new forms of temporally distributed exploration. It shows how organizational memory is preserved without centralized control, and how design intent can persist even when project ownership rotates yearly. At the same time, documentation is also produced by long term participants of the AI BooSTcamp. This shows how the information system sustaining the AI BooSTcamp is perceived to be socio-centric and informal, despite the creation of technical elements and formal artefacts. The role of the leading circle in the holacracy and its perception of the subcircles need to be investigated to assess the degree of formality and informality in the BooSTcamp.
The structure of the AI Boostcamp enables strong learning, in technical skills and so skills. Over the time of an internship, participants develop key skills for researchers: perseverance, creativity, resilience, acceptance that they will sometimes ‘hit the wall’ and the understanding that this is alright as they will bounce back from it. The ndings also show how participants can focus on exploration and develop their practice of innovation in a sheltered and friendly atmosphere.
5.4.
The AI BooSTcamp operates as a CAS in which exploration emerges from local interactions, bottomup coordination, and distributed judgment. Participants adapt continuously to shiing conditions. Autonomy is negotiated, earned, and recongured over time. Roles are uid, expertise is situational, and leadership is distributed. These are signs of adaptive capacity.
Exploration within the AI BooSTcamp is not orchestrated top-down. It materializes as interns challenge guidance, reinterpret feedback, or quietly persist with alternative ideas. These micro-moves contribute to a broader system of innovation sustained not by structure, but by informal cohesion and shared symbolic frames. The program functions as a modular, non-technocentric exploration infrastructure—loosely coupled to operational systems, yet robust enough to persist across cycles.
By grounding exploration in social interaction and symbolic meaning rather than formal hierarchy or digital tooling, the AI BooSTcamp provides a distinctive model for cultivating innovation in industrial contexts. It supports recent calls [22; 8; 4] to move beyond static typologies of ambidexterity and attend to the evolving, enacted, and emergent nature of exploratory systems. Here, holacracy functions not as a structure to be replicated, but as a scaolding that legitimizes dynamic microfoundations and makes room for technical experimentation to coexist.
In our future work, we will expand the number of interviews to overcome the limitations of our small sample. We expect that more interviews will allows us to uncover structural elements that foster exploitation allowing us to better understand OA. Since the current sample is weighted toward junior participants, future work will expand to include senior and long-term employees in order to capture a more balanced view of OA. Investigating how a non-technocentric CAS can be stabilized or scaled may oer valuable insights for organizations seeking to foster OA.
We aim to contribute to theory by identifying new microfoundations at the individual, team, and structural levels that shed light on the dynamic nature of OA. Our ndings may also oer practical insights for managers seeking to adopt the BooSTcamp approach as an innovation program in their organizations to foster OA.
During the preparation of this paper, the authors used ChatGPT-4o to improve the clarity and grammar of the manuscript. The authors have reviewed and edited the content as needed and take full responsibility of the work.
[14] Birkinshaw, Julian, and Kamini Gupta. "Clarifying the distinctive contribution of ambidexterity to the eld of organization studies." Academy of Management Perspectives 27.4 (2013): 287-298.
[15] Alvesson, Mats, and Dan Kärreman. "Interfaces of control. Technocratic and socio-ideological control in a global management consultancy rm." Accounting, organizations and society 29.3-4 (2004): 423-444.
[16] Witschel, Daliborka, Julian Marius Müller, and Kai-Ingo Voigt. "What takes the wind out of their sails? A micro-foundational perspective of challenges for building dynamic capabilities towards digital business model innovation." Journal of Business Research 75.3 (2023): 345-388.