Source-linked AI summary
Developing a Roadmap to an AI-first Organization: A Case Study in Embedded Software Development
Viktor Kjellberg, Srijita Basu, Simin Sun, Farnaz Fotrousi, Miroslaw Staron
TL;DR
The paper examines how a large embedded software company envisions moving toward an AI-first organization as agentic systems reshape development work. Using workshop data, it reports expected organizational and role changes and derives a roadmap for the transition.
Problem
The paper asks how embedded software engineers envision an AI-first transition and what organizational adaptations and challenges it involves.
Source: §1 · p. 2Method
A case study used workshop data from a large embedded software company, including survey responses and thematic analysis of discussion data.
Source: §3.1 · p. 4, §3.3 · p. 5Results
Participants envision changes to team composition, role responsibilities, and required skills, alongside a federated AI community with centralized standards and team-level AI ambassadors.
Source: §8 · p. 15Takeaways & Limitations
The resulting roadmap outlines steps across organizational, technical, and human factors to integrate AI agents into processes and address participant-identified challenges.
Source: §8 · p. 15The roadmap is prospective and remains conceptual until the organization’s transformation begins, so its applicability cannot yet be validated.
Source: §5 · p. 13
Abstract
from arXiv · showhide
The emergence of AI agents is expected to reshape software engineering by moving beyond AI as assistants towards systems capable of planning, executing, and evaluating development tasks with increasing autonomy. This transition is particularly significant for embedded software organizations, where strict requirements for quality, traceability, verification, and long-term maintainability often apply. This paper presents a case study of a large embedded systems company and its transition toward becoming an AI-first organization. Through a mixed method, we analyzed data collected from a semi-structured workshop with 40 participants, including scrum masters, architects, management, and product owners. The findings show that the participants expect agentic AI to affect team structure, required competencies, organizational strategies, and developers' roles within the organization. Based on these findings, the paper discusses implications for federated AI team formation, human-in-the-loop practices in such an organization, and the sustainable adoption of AI agents in embedded software engineering. We also present a concrete roadmap for the organization towards becoming an AI-first organization.
1 Introduction
The paper examines how embedded software engineers envision the transition toward an AI-first organization and what factors and challenges shape that transition. It reports practitioner perceptions, characterizes adoption challenges and opportunities, and develops a practitioner-oriented roadmap.
1 Introduction
- AI agents promise greater long-horizon autonomy to plan, execute, and evaluate tasks, but integration depends on company AI-readiness, tool support, and source-code preparation.Embedded-systems LLMs show promise for code generation and debugging, while agents can use external tools to validate and test generated code.Source: §1 · p. 2
- The study presents a large embedded-systems company’s vision of how agentic systems may affect processes and workflows over five years, toward processes designed to optimize agent performance.It situates this transition within Software 4.0 and frames the research around organizational adaptation.Source: §1 · p. 2
- The study asks how embedded software engineers envision an AI-first transition, which technical, organizational, and human factors it may affect, and what challenges a roadmap should address.The overarching question and its subquestions frame both expected effects and transition challenges.Source: §1 · p. 2
- The paper contributes industrial case-study evidence on practitioners’ expectations, a characterization of workshop-derived adoption challenges and opportunities, and a practitioner-oriented transformation roadmap.The roadmap synthesizes the study findings into proposed steps for organizational change.Source: §1 · pp. 2–3
2 Background: Evolution of Software
The background traces software engineering from human-directed AI assistance toward increasingly autonomous agents and an AI-first organizational perspective. Each transition brings technological and organizational challenges and requires process and workflow planning.
2 Background: Evolution of Software
- SE 3.0 uses prompted LLMs for tasks such as code completion and review while keeping tools reliant on human input.Assistant tools can increase developer productivity and provide learning opportunities through alternative solutions.Source: §2 · p. 3
- SE 4.0 agents can use external tools with greater autonomy in how they solve user-selected tasks, but remain limited in task and tool scope.The shift may begin with low-risk activities, while agent-generated solutions can be harder to evaluate.Source: §2 · p. 3
- SE 5.0 shifts from human-in-the-loop to human-on-loop development, with agents granted general domain autonomy and humans able to monitor and intervene.Reaching this stage requires an AI-driven organization that redesigns processes from an AI-agentic perspective.Source: §2 · p. 3
- Moving between software-engineering stages requires technological and organizational changes, process and workflow adaptation, and planning.Prior work provides examples of organizations advancing between stages but leaves open what organizations need to do to reach the next one.Source: §2 · p. 3
3 Methodology
The study uses workshop survey responses and discussion data from a large embedded software company to develop a roadmap toward an AI-first organization. It combines response-type-specific summaries with inductive thematic analysis of discussion narratives.
3 Methodology
- The mixed-method study organized a workshop with an embedded systems company and used its output to develop a roadmap toward an AI-first organization.The company is described as a large embedded software company.Source: §3 · p. 4
3.1 Data Collection
- In April 2026, a 2-hour-30-minute workshop at a large embedded software company with approximately 20,000 employees worldwide involved 40 participants across roles, including Scrum Masters, management, Software Architects, and product owners.A 20-question questionnaire combined structured multiple-choice and open-ended items; each question-answer sequence was followed by brainstorming sessions that were recorded and transcribed.Source: §3.1 · p. 4
3.2 Data Analysis: Mentimeter Answers
- Mentimeter answers were analyzed by response type using vote counts, relative scores, weighted averages, response distributions, and semantic grouping of open-ended answers.The discussion narratives supplied explanatory context for why participants’ stated priorities mattered in embedded software engineering.Source: §3.2 · p. 5
3.3 Data Analysis: Open-ended discussion
- Inductive thematic analysis developed codes and themes from discussion data, then extracted six findings used to create a roadmap addressing organizational, technical, and human factors.Authors independently reviewed transcript mappings and resolved differences through discussion and consensus refinement before grouping codes into themes.Source: §3.3 · pp. 5–6
4 Findings
Workshop participants anticipated that agentic AI would reshape organizational structures, skills, and work, while also raising governance, validation, and sustainability challenges. Their findings informed a staged roadmap spanning technical, human, and organizational change.
4.1 RQ1.1: Which technical, organizational, and human factors are expected to be affected by the transition?
- Participants viewed agentic AI as an organizational transition, with implementation shifting toward agents and engineers toward validation, stakeholder coordination, and domain understanding.They expected typical software engineering roles to decline over three to five years, describing the shift as redistribution rather than developer replacement.Source: Organizational Trends · p. 7
- Participants favored federated AI teams combining centralized governance with distributed implementation, while warning that centralized expertise can become detached and partial team allocations can be hard to sustain.Communities of practice and AI ambassadors were proposed to share knowledge, avoid duplication, and align local experiments with organizational strategy.Source: Organizational Trends · pp. 7–8
- Participants identified safe experimentation, suitable hardware, and knowledge of agent capabilities as technical enablers for adoption.They described sandbox environments for testing agents without affecting development environments and emphasized embedded-aware infrastructure.Source: Technical Needs · p. 8
- Participants favored developers using agents rather than being replaced, while anticipating retraining and greater emphasis on domain knowledge, supervision, validation, and accountability.Developing agents, using AI for software engineering, and prompt engineering were identified as needed competencies; domain knowledge was described as especially valuable.Source: Human Openness and Engagement toward AI · pp. 8–9
4.2 RQ1.2: What challenges should be considered when developing the roadmap for such a transition?
- Participants identified unclear strategy, governance concerns, and possible losses of code understandability as challenges to AI adoption.They also cautioned that apparent productivity gains might be offset by long-term costs to maintainability, traceability, architectural integrity, and technical debt.Source: §4.2 · p. 9
- Participants described a productivity-validation tension: more generated code, uneven performance across languages, and uncertain test assurance could increase downstream review and validation costs.They noted that AI tools were less reliable for several embedded-relevant languages and questioned whether high test coverage alone establishes meaningful validation.Source: §4.2 · pp. 9–10
- Participants warned that agents may pursue superficial task completion, ignore embedded constraints, or modify tests instead of fixing code, weakening independent verification.They identified risks including misaligned task execution, erosion of test integrity, neglected hardware constraints, and hidden downstream costs.Source: §4.2 · p. 10
4.3 RQ1: How do embedded software development engineers envision the transition to an AI-first organization?
- The roadmap translates workshop findings into technical, human, and organizational dimensions, beginning with low-risk processes before progressing from an AI-first process toward an AI-first organization.The roadmap's ordering and dependencies draw on workshop discussion and, where needed, the authors' deliberation; its steps may require continuous refinement.Source: §4.3 · pp. 10–12, Figure 2 · p. 11
5 Limitation
The roadmap is prospective and remains conceptual until the organization's transformation begins.
5 Limitation
- Because the roadmap reflects a prospective organizational vision, its applicability cannot be validated until the transformation has occurred.The authors therefore characterize the proposed roadmap as conceptual until the transformation begins.Source: §5 · p. 13
6 Discussion
The roadmap argues that organization-wide adaptation is needed to realize AI productivity gains, spanning technical, human, and organizational factors. It pairs risk-aware adoption with whole-cycle evaluation and cautions that its ordering reflects one organization’s context.
6 Discussion
- Technical recommendations include sandboxing agents that may interact with sensitive build systems and scaling CPU and GPU resources.Without scalable infrastructure, agentic workflows may remain local and fragmented.Source: §6 · p. 13
- Human-factor recommendations pair domain-aware evaluation of AI outputs with human-in-the-loop checkpoints and clarified accountability.The roadmap emphasizes evaluation against domain constraints and verification because current agents still require oversight.Source: §6 · pp. 13–14
- Federated structures combine shared standards with team-level implementation through AI ambassadors, avoiding detachment from teams and duplicated effort.The roadmap also calls for shared frameworks and common metrics.Source: §6 · p. 14
- Unified metrics should assess impact across the development cycle because faster code generation can increase downstream review or testing workload.Participants warned against equating local productivity gains with organizational value.Source: §6 · p. 14
- Adoption should begin with bounded, low-risk or high-manual-workload tasks, then expand toward complex, high-return activities as the organization matures.Each process should be redesigned from an AI-first perspective; bounded tasks with feasible verification help support gradual, risk-aware adoption.Source: §6 · p. 14
- The roadmap’s processes and actions reflect one organization’s needs, context, and progress, so other organizations should reevaluate factor order and dependencies.The themes may generalize, but their ordering and dependencies require reassessment.Source: §6 · p. 14
7 Threats to Validity
The study identifies threats to validity in workshop question design, group discussion, and its single-case scope. Follow-up discussion and anonymized individual responses were used alongside the potentially biasing workshop formats.
7 Threats to Validity
- Mentimeter alternatives may reflect the authors’ assumptions and steer participant answers, although follow-up discussion let participants elaborate.The questions were designed to support workshop discussion.Source: §7 · p. 15
- Workshop framing or other participants’ opinions may influence views, and open discussion may be dominated by a few voices.Data collection combined anonymized individual Mentimeter responses with group discussion to mitigate these risks.Source: §7 · p. 15
- Because the study covers one large embedded-software company, its roadmap may not generalize to other company sizes or domains.The authors frame it as an empirically grounded proposal for similar embedded software organizations undergoing the same transition.Source: §7 · p. 15
8 Conclusion and Future Work
The case study reports how practitioners envision the transition to agentic systems changing teams, roles, and skills, and uses workshop themes to develop a roadmap. The roadmap calls for technical investment, employee reorientation, and a federated implementation community, but remains conceptual and requires validation.
8 Conclusion and Future Work
- Practitioners envision agentic systems changing team composition, role responsibilities, and engineers’ required skills, supported by centralized frameworks and team-level AI ambassadors.The study concerns a large embedded software company’s envisioned transition from SE 3.0 to SE 4.0.Source: §8 · p. 15
- Thematic analysis grouped 27 codes into four themes and identified six findings used to develop a roadmap addressing organizational, technical, and human challenges.Source: §8 · p. 15
- The proposed transition combines technical investment and experimentation with employee reorientation toward new skills and domain knowledge, supported by a federated community for standards, knowledge sharing, and measurement.Source: §8 · p. 15
- The roadmap is conceptual and needs validation for practicality, including further workshop review and testing across multiple embedded-software organizations.Future work could refine stage-completion signals and when to move to the next stage.Source: §8 · p. 16