The Psychological Costs of Artificial Intelligence Adoption in Software Engineering
Adam Alami, Elda Paja, Abhishek Tiwari
cs.SE · cs.AI
Quick verdict
Qualitative single-case study claims five psychological costs of AI adoption in SE; well-executed, candid, but single-site volunteer sample limits transfer.
Summary
This paper presents a qualitative single-case study of the psychological costs software practitioners experience during organizational AI adoption, conducted one year after a large Danish software services company ('SoftHouse', ~1,200 employees, regulated domains) launched its adoption initiative. Data comprise 21 semi-structured interviews (~290 transcript pages, ~17 hours 40 minutes of audio), five stakeholder meetings, and member checking with 12 participants, analyzed through Miles et al.'s two-cycle coding with meaning-saturation monitoring. The authors identify five costs — uncertainty distress, accountability anxiety, cognitive and workload intensification, craft identity disruption, and meaning and satisfaction erosion — arising from AI's characteristics (rapid pace, unknown risks, role-altering capability) and amplified by organizational conditions such as maturity-model pressure and governance ambiguity. Practitioners respond by managing (oversight and verification practices), mitigating (risk containment, skill preservation, identity reframing), or absorbing the costs (resigned adaptation), thereby preserving agency, competence, and professional identity. The paper proposes 'agency displacement' and a 'verification tax' as explanatory constructs and applies Conservation of Resources and Self-Determination Theory post hoc. Main limitations: a single regulated case; a cross-sectional snapshot; a volunteer sample recruited via the AI program director, who also participated; and role-level counts built on as few as three participants. The authors acknowledge these and frame conclusions as analytically transferable rather than statistically generalizable.
Review complete
Assessment at a glance
4
risk flags identified
● high 0
● medium 2
● low 2
5 strengths · 5 weaknesses · 5 author questions
AI text signal = Binoculars zero-shot detector score over 20 sampled passages (Fast-DetectGPT co-reported) — absence does not prove human authorship, and a flag does not imply misconduct.
Levels reflect verification complexity — not a quality score. Evolve never rates your paper.
Key claims under review
- Organizational AI adoption imposes five recurring psychological costs on practitioners — uncertainty distress, accountability anxiety, cognitive/workload intensification, craft identity disruption, and meaning/satisfaction erosion — that the adoption strategy neither anticipated nor addressed.
- These costs arise from AI's disruptive characteristics amplified by organizational conditions (maturity-model pressure, governance ambiguity), not from resistance to AI.
- Practitioners respond by managing, mitigating, or absorbing costs in order to preserve agency, competence, and professional identity (Table 8).
- Psychological costs scale with 'agency displacement' — the extent to which AI substitutes the generative activities through which a role expresses expertise.
- AI shifts engineers' effort from production to post-hoc verification — a 'verification tax' that current productivity metrics ignore.
Strengths
Transparent method chain
Corpus is quantified (21 interviews, ~290 transcript pages, 17h40m audio, mean 50 min; Sect. 3.3), the interview guide is documented (Table 3), analysis follows Miles et al. First/Second Cycle coding, and meaning saturation is monitored with explicit disclaimers that saturation was assessed at sample level, not per role (Sect. 6).
Member checking with reported feedback
12 of 21 participants (asterisked in Table 2) reviewed the findings and verbatim agreement quotes are included (e.g., P7: 'I agree largely with what has been said'), which is rarer than it should be in SE qualitative work.
Candor about the case
Boundary conditions and honesty about scope
Actionable synthesis
Weaknesses
Gatekeeper recruitment and dual roles
Role-level fractions on thin cells
Participant attribution inconsistency
Post-hoc theorizing and modest construct novelty
Single regulated case, cross-sectional, senior-skewed
Full Review Report
$1.99 · one-timeUnlock the complete preliminary analysis Evolve produced for this paper.
Report preview
Scope & Claims
The paper asks what psychological costs software practitioners incur during organizational AI adoption and how they manage them (Sect. 1 RQ), defining 'psychological costs' via the APA scope of psychological phenomena plus Lazarus-Folkman appraisal. The setting is 'SoftHouse', an anonymized 1,200-employee Danish software services firm in regulated domains, studied one year into an adoption that rolled out GitHub Copilot (abandoned after three months over 'quality issues'), then Claude Code/Desktop, under an 'experiment-and-share' strategy with AI ambassadors and a 1-5 maturity model (Sect. 3.2). Central claims: (1) five recurring costs — uncertainty distress, accountability anxiety, cognitive/workload intensification, craft identity disruption, meaning/satisfaction erosion; (2) they are amplified by adoption pressure and governance ambiguity, not caused by resistance; (3) practitioners manage, mitigate, or absorb them to protect agency, competence, and professional identity (Fig. 4.1, Table 8); (4) costs scale with 'agency displacement', and verification work constitutes a 'verification tax' (Sect. 5.4).
The remaining sections cover Methodology Assessment, Statistical Evidence, Results & Reproducibility, Limitations & Risks, and Verdict & Recommendations.
- All 5 strengths & 5 weaknesses, analyzed in detail
- 4 risk flags with severity, section & reviewer questions
- Methods & statistical evidence assessment
- 5 questions a peer reviewer would ask
- Full six-dimension evaluation report & recommendations
One-time purchase · this paper only
One-time purchase, this paper only. AI-generated preliminary analysis — not peer review.
Frequently asked questions
What are the five psychological costs of AI adoption identified in this paper?
Uncertainty distress, accountability anxiety, cognitive and workload intensification, craft identity disruption, and meaning-and-satisfaction erosion. The paper says AI's characteristics (fast pace, unknown risks, role-altering capability) give rise to these costs, while organizational conditions — maturity-model pressure and governance ambiguity — amplify them.
How was this study conducted and on whom?
It is a qualitative single-case study of 21 volunteers at a large anonymized Danish software services company (~1,200 staff, regulated domains), one year into its AI adoption. Data came from semi-structured interviews (~290 transcript pages, 17h40m audio), five stakeholder meetings, and member checking with 12 participants, analyzed via two-cycle coding with meaning-saturation monitoring.
What is the 'verification tax' described in the paper?
It is the paper's term for the extra cognitive burden engineers bear when they must verify, justify, and remain accountable for AI-generated code they did not write. The authors define it not as a sixth cost but as the synthesis of accountability anxiety and cognitive load intensification, arguing productivity metrics that count only generation speed ignore it.
Does this paper claim software engineers resist AI?
No. The authors explicitly find the opposite: psychological costs coexist with acceptance, and where adoption stayed limited it reflected compliance constraints, uneven use cases, and business-as-usual tensions rather than opposition to AI.
About this review
This is an AI-generated preliminary review by Evolve.Pub, produced from the openly available text of the paper. It is not peer review and does not represent the authors' views. The canonical record of the paper is at the publisher link above.
- Three '[...sampling gap...]' markers skip portions of the provided text — parts of Related Work §2.3, some methodology subsections, Findings §4.5 and part of §4.6, Table 10's contents, and part of §5.2; judgments touching those sections rest on the surrounding visible text only.
- No inter-coder reliability reporting is visible in the provided text; whether one exists in the skipped sections is unknown.
Go deeper on this paper
Wrote a paper? Stress-test it before reviewers do.
Upload your manuscript for the same structured scrutiny — reasoning, statistics, evidence, figures, and reproducibility.