Source-linked AI summary
The Psychological Costs of Artificial Intelligence Adoption in Software Engineering
Adam Alami, Elda Paja, Abhishek Tiwari
TL;DR
AI adoption in software engineering is commonly evaluated through productivity and technical capability, but its psychological costs in organizational practice remain insufficiently explained. Through a qualitative case study of adoption in a large software development company, this paper identifies five costs and examines how practitioners respond. It frames AI adoption as a human transition involving accountability, competence, agency, identity, and meaning.
Problem
Existing research explains AI uptake, technostress, and well-being separately, but insufficiently explains what software practitioners incur when organizational AI adoption is enacted through software engineering practice.
Method
The study uses an in-depth case study of organizational AI adoption, drawing on qualitative data from a large software development company one year into adoption.
Results
The study identifies five psychological costs—uncertainty distress, accountability anxiety, cognitive load intensification, craft identity disruption, and meaning and satisfaction erosion—shaped jointly by AI’s disruptive characteristics and organizational conditions.
Takeaways & Limitations
AI adoption in software engineering should be understood as a human transition and evaluated for its effects on practitioners’ agency, competence, professional identity, and responsible engineering judgment.
Takeaways & Limitations
The findings are situated in SoftHouse, so the prominence of individual costs and observed role patterns may not transfer unchanged to other organizations.
Abstract
from arXiv · showhide
Artificial intelligence (AI) is increasingly used to augment software engineering (SE) workflows. While code generation remains the main use case, organizations are actively seeking AI integration in other practices such as test cases generation and code reviews. Organizational AI adoption strategies seem to focus on tangible outcomes such as productivity. However, AI is a disruptive force, introduced into settings where role identity, team norms, and the sources of job satisfaction were well established before the recent advances in generative AI. Historically, technological disruptions have caused psychological and social strains in workplaces, ranging from anxiety and eroded meaning to deskilling and disrupted professional identities. The assumption that AI for SE is cost-free may not be accurate. Therefore, in this study we sought to understand the psychological costs software professionals experience during organizational AI adoption. We carried out a case study in a large software development services company, one year after the company launched its AI adoption. We collected qualitative data through meetings and semi-structured interviews (N = 21). We found that software professionals experience accountability anxiety, craft identity disruption, meaning and satisfaction erosion, cognitive and workload intensification, and uncertainty distress. Practitioners manage these costs through practices that restore control, mitigate them through protective and identity-preserving adaptations, or absorb them, carrying what neither can resolve. We contribute to AI-human collaboration in SE by repositioning AI adoption as a human transition, not only a technological and organizational one.
1 INTRODUCTION
This study examines how organizational AI adoption becomes psychologically costly in software engineering practice, rather than treating adoption only as a technological or productivity transition. A qualitative case study identifies five psychological costs and shows how practitioners manage them while preserving accountability, competence, agency, and meaning.
- Research focus: The study asks what psychological costs software practitioners experience during organizational AI adoption and how they manage them.The evidence is situated in a large Danish software development company one year into adoption, using interviews, member checking, and meetings.
- Adoption context: Organizational AI adoption introduces commercial generative AI tools, coding assistants, agentic tools, and conversational LLMs into software teams’ workflows under active adoption pressure.Internal messaging, AI ambassadors, and an AI maturity model were used to promote and assess uptake.
- Findings: Practitioners experienced five psychological costs: accountability anxiety, craft identity disruption, meaning and satisfaction erosion, cognitive and workload intensification, and uncertainty distress.Adoption pressure and governance ambiguity amplified these costs.
- Responses: Practitioners managed costs through oversight, skill-preservation routines, identity reframing, and resigned adaptation when other responses fell short.These responses restore control, mitigate costs through protective adaptations, or absorb unresolved burdens.
- Findings: Psychological costs emerged as practitioners integrated AI while preserving accountability, competence, professional agency, and meaning, rather than from opposition to AI adoption.Limited adoption mainly reflected compliance constraints, uneven applicability across roles and use cases, and tensions with business-as-usual demands.
- Contribution: The paper repositions AI adoption in software engineering as a human transition in which competence, identity, and practitioners’ relationship to work are renegotiated.Its diagnostic model connects technological characteristics and organizational conditions to changes in accountability, effort, identity, meaning, and practitioner responses.
2 RELATED WORK
Existing research explains AI adoption, technostress, and well-being separately, but insufficiently explains what organizational AI adoption demands from software practitioners. This paper addresses that gap by analyzing psychological costs and how practitioners carry them in SE practice.
- Generative AI redistributes software engineering effort toward prompting, evaluation, correction, integration, and accountability rather than simply automating production.
- AI adoption research increasingly treats uptake as shaped by usefulness, trust, facilitating conditions, organizational support, and the surrounding work system.
- The paper extends prior work by examining what organizational AI adoption demands from practitioners, why those demands arise in SE, and how practitioners carry them.
- Technostress research frames AI as both resource and demand, with strain emerging through interactions among demands, resources, and appraisal.
- Well-being research extends beyond strain to satisfaction, engagement, autonomy, competence, relatedness, professional growth, and meaningful work.
3 METHODS
The study uses a holistic single-case design to examine AI adoption as it unfolds in a large Danish software development company. The case provides access to practitioners’ lived experiences during the transition.
- The researchers conducted a holistic single case study of SoftHouse, a large Danish software development company approximately one year into organization-wide AI adoption.
- SoftHouse develops and maintains software for critical and highly regulated domains, where reliability, security, and compliance are institutionalized engineering obligations.
- The company is organized into domain-specific business units and projects, with about 1,200 employees and software practitioners comprising most of its workforce.
- The case entered an organization with mature Scrum practices and long-established norms for planning, delivery, and collaboration.
- SoftHouse’s first enterprise rollout, GitHub Copilot, was abandoned after three months because of reported quality issues and overall dissatisfaction.
AI Adoption and Its Context.
SoftHouse supported AI integration through an organization-wide program that encouraged experimentation and knowledge sharing rather than prescribing a uniform workflow. Adoption remained bounded by evolving governance and project-specific constraints.
- SoftHouse established an enterprise AI program with a director and project-level AI ambassadors to promote adoption and knowledge sharing.
- AI use was governed by evolving rules for data classification, online access, and credential delegation, while permitted uses varied across projects.
- One year after launch, the initiative had not achieved its original target of 50% adoption among software engineers.
- The experiment-and-share strategy encouraged practitioners and projects to experiment within compliance boundaries, build competence, and share lessons locally and enterprise-wide.
Experiment-and-Share as an AI Adoption Strategy.
SoftHouse’s experiment-and-share strategy produced uneven adoption and knowledge circulation because compliance, applicability, workload, and social barriers constrained what practitioners could use or disclose. The analysis developed an emergent model linking adoption conditions, psychological costs, and coping responses.
- Adoption constraints: Compliance requirements varied by domain, customer, and product, making uniform AI prescriptions impractical and restricting use in sensitive settings.
- Adoption constraints: External pressure for speed conflicted with compliance and security obligations, creating competing expectations around AI adoption.
- Adoption outcomes: Some practitioners reached agentic adoption, while others were limited to locally hosted models and encountered failed experiments because of limited computation power.
- Knowledge sharing: Knowledge sharing worked when locally scaffolded, but lessons did not always transfer across roles and some practitioners withheld experiences because of judgment or governance concerns.
- Data collection and analysis: The study combined stakeholder meetings, semi-structured interviews, inductive coding, pattern-code development, saturation assessment, and member checking.
- Study boundaries: The confidential case setting required reduced contextual description and anonymized or generalized quotations and identifiers.
- Analytical model: The resulting conceptual model connects conditions producing or intensifying psychological costs with practitioners’ responses to manage, mitigate, or absorb them.
4 FINDINGS
Organizational AI adoption reshapes practitioners’ work, responsibilities, identities, and expectations, producing psychological costs that adoption conditions can amplify. Practitioners face persistent uncertainty, accountability anxiety, and concerns about maintaining professional standards amid rapidly changing AI capabilities.
- AI adoption reshaped how practitioners perceived their work, professional identity, responsibilities, and future in software engineering.
- Uncertainty Distress: Rapidly evolving AI and unstable practices made knowledge perishable, leaving practitioners uncertain about the lasting value of their skills and professional development.
- Uncertainty Distress: Uncertainty distress arose from rapid AI evolution and was amplified by limited organizational knowledge, epistemic uncertainty, and perceived adoption pressure.
- Accountability Anxiety: Opaque AI outputs increased accountability anxiety because practitioners remained responsible for verifying quality and explaining decisions without clear governance guardrails.
- Accountability Anxiety: Strong organizational commitments to quality and compliance made practitioners concerned that pressure to use insufficiently reliable AI tools could compromise standards for which they were responsible.
5 DISCUSSION AND IMPLICATIONS
The discussion presents an explanatory and diagnostic model linking AI characteristics and organizational conditions to psychological costs in software engineering, while extending COR and SDT interpretations of these costs. It identifies actionable organizational targets but emphasizes boundaries related to valued activities, assurance capacity, learning conditions, and role-specific applicability.
- Contribution: The model links AI characteristics and organizational conditions to changes in accountability, effort, identity, and meaning, alongside practitioners’ responses.It distinguishes conditions that produce costs from those that intensify them, supporting diagnosis of where burdens originate.
- Practical implications: Governance ambiguity, burdensome verification, and disrupted identity or meaning indicate different intervention targets in guidance, assurance practices, work design, and opportunities for valued expertise.The diagnostic framework translates explanatory pathways into questions for adoption strategy and change management, though its effectiveness requires evaluation.
- Theoretical implications: COR explains these psychological costs as experiences arising when valued resources are threatened, depleted, or require sustained investment to protect.The model further identifies agency displacement and absorption as conditions or responses relevant to AI-mediated software work.
- Theoretical implications: AI adoption threatens agency, competence, and professional identity, with accountability anxiety, craft identity disruption, meaning erosion, cognitive load, skill-atrophy fears, and uncertainty distress illustrating these losses.The findings connect opaque reasoning to reduced ability to understand and justify work, delegated generative activity to alienation, and rapid evolution to unstable expertise.
- Theoretical implications: SDT interprets the costs as frustration of autonomy and competence through diminished authorship, accountability without explainability, perishable expertise, and skill-atrophy concerns.Observed responses include manual coding rituals, staged AI delegation, and identity reframing.
- Boundary conditions: The findings’ transferability is bounded by practitioners’ relationships to delegated activities, their assurance capacity, adoption expectations and learning conditions, and the applicability of organizational experiments to specific roles.The authors state that these boundary effects are informed by the case and have not been independently established.
6 RESEARCH TRUSTWORTHINESS
The study established trustworthiness through meaning saturation and member checking, while limiting claims about saturation and role-specific patterns. Role-based magnitude patterns are therefore descriptive and exploratory rather than conclusive.
- Meaning saturation: Meaning saturation was assessed by comparing new interview excerpts and pattern codes with evolving themes until later interviews added no conceptual properties or relationships.The analysis remained open to further recruitment if conceptual gaps had emerged.
- Scope of claims: Because saturation was assessed at the sample level, role-specific patterns—especially for roles represented by three participants—remain exploratory rather than conclusive.The authors do not claim meaning saturation within individual roles.
- Member checking: Member checking involved sharing interpretations with participants and seeking their feedback to establish reliability.
7 LIMITATIONS AND TRADE-OFFS
The study’s limitations concern its organizational setting, role distribution, cross-sectional timing, participant experience, and lack of gender-based analysis. These boundaries constrain how broadly and longitudinally the findings can be interpreted.
- The findings are situated in SoftHouse’s regulated domains, strong quality culture, mature practices, and experiment-and-share strategy, so observed cost patterns may not transfer unchanged.The authors distinguish AI-related sources from organizational amplifiers and frame transferability as an analytical assessment.
- The sample was skewed toward software engineers, potentially obscuring richer or role-specific findings for managers, architects, and testers.The authors identify verification tax as one role-specific finding that may be affected by this imbalance.
- Data captured a single snapshot one year into adoption, leaving longer-term trajectories of psychological resource depletion, recovery, and adaptation unresolved.The authors specifically identify uncertainty about later burnout, attrition, and evolving responses as AI adoption changes.
- Nineteen of 21 participants had more than ten years of experience, which may have foregrounded disrupted expertise and craft-identity costs.Less-experienced practitioners may experience AI adoption differently while their skills and professional identities develop.
- The study did not conduct gender-based analysis because women were underrepresented in the sample.The authors propose gender-based designs for future research on AI adoption’s impact on software professionals.
8 CONCLUSION
The conclusion argues that organizational AI adoption in software engineering should be understood through practitioners’ psychological experience, not productivity and technical capability alone. It identifies five psychological costs and two concepts explaining how those costs vary across roles and tasks.
- The study identifies five psychological costs of organizational AI adoption: uncertainty distress, accountability anxiety, cognitive load intensification, craft identity disruption, and meaning and satisfaction erosion.
- Agency displacement explains how the extent to which AI assumes expertise-exercising activities shapes psychological costs across software engineering roles.
- Verification tax captures the additional effort involved when practitioners verify AI-generated software, despite reduced generation effort.
- AI adoption should be evaluated by its effects on practitioners’ agency, competence, professional identity, and responsible engineering judgment alongside productivity and technical augmentation.
GENERATIVE-AI USE
The paper reports using Paperpal for grammar, proofreading, and fluency improvements, while the first author independently produced and reviewed the draft.
- Paperpal was used solely for grammar, proofreading, and fluency improvements, with the first author verifying and accepting or rejecting suggestions.