Steven Watson | EdgeLab | Research and Scholarship | 19 September 2026
Updated 21 September 2026 following a response from co-author Joana Carneiro Pinto. The discussion below now explains the table-label error and hypothesis typo she identified, and clarifies the intended meaning of “long-term”. The authors intend to request a formal journal correction.
What kind of organisational support helps when workers cannot see where AI leaves their careers? Reassurance, a helpful manager or a training session may reduce an immediate fear of job loss. They do not necessarily answer a longer question: what place will a person’s judgement, experience and developing expertise have after work is reorganised around AI?
A study of insecurity and support
João Matias and Joana Carneiro Pinto investigate this distinction in a study of 106 professionals in Portugal whose organisations had introduced AI. The authors ask whether AI-related insecurity is associated with perceived job security, career competence and career satisfaction, and whether perceived organisational support weakens those relationships.
The study is cross-sectional: all the main measures were collected in one survey. Participants were recruited through purposive and convenience sampling from professional networks. Nearly 70 per cent were women, about 86 per cent had a university degree, and most worked in large private organisations. Education and research formed the largest sector. Sixty-eight participants also answered at least one of two open questions about organisational help and AI’s effect on daily life.
The statistical results show associations rather than changes through time. Greater AI-related insecurity was strongly associated with lower perceived job security and weakly associated with lower career satisfaction. Its association with self-reported career competence was not statistically significant. Organisational support was positively associated with all three outcomes. In the moderation analysis, support weakened the relationship between AI insecurity and job security, but did not significantly change the relationship between insecurity and career satisfaction.
The open responses help explain why. Participants described fears of replacement and professional devaluation, unclear organisational communication and a need for practical upskilling. A smaller group described AI as useful augmentation. The authors call the overall pattern “asymmetric buffering”: support may help with the proximate fear of losing a job while leaving uncertainty about a career’s longer direction unresolved.
What the paper gets right
The paper’s most useful move is to refuse a single category of employee wellbeing. Feeling secure in one’s present post, feeling capable of developing and feeling satisfied with a career are related but different. An organisation can offer emotional support while withholding the information, participation or opportunities that would let employees form a credible account of their future.
The authors also resist treating resistance to AI as a simple skills deficit. Their qualitative material points towards strategic ambiguity and exclusion from decisions, not merely reluctance to learn a tool. Their practical recommendations—clearer communication, participation in procurement choices, complementary skill development and psychological-safety assessment—follow sensibly from that framing, although the study does not test those interventions.
What the evidence cannot yet establish
In her response to this review on 21 September 2026, Joana Carneiro Pinto clarified that “long-term” refers to career satisfaction as a longer-horizon career outcome, not to evidence from following people over time. That distinction is helpful, and I have revised my earlier wording accordingly. Career satisfaction was measured once: the study cannot establish change over time or causal relationships. A non-significant interaction in a sample of 106 also does not establish that support fails to protect satisfaction. It shows that this study did not detect the proposed buffering relationship.
All central variables are self-reports collected together. The authors acknowledge common-method bias and reverse causality, but the usual statistical check cannot remove them. People who feel secure may judge their organisation as more supportive; a generally positive or negative work experience may shape several answers at once. The highly educated, single-country convenience sample also limits transfer to workers with less organisational protection.
The apparent mismatch between the correlation table and surrounding prose has now received an author clarification. Responding on 21 September 2026, Pinto explained that the coefficients are correct, but the displayed row labels were reordered while the numerical matrix retained the sequence stated in the table title: AI-related insecurity, Career Competency, Career Satisfaction, Organizational Support, and Job Security. She confirmed that the surrounding text and regression analyses report the coefficients correctly. She also identified a related Results-section typo: the non-significant association between AI-related insecurity and Career Competency should refer to H3, not H1. The authors say they will contact the journal to request a formal correction of both points. This explains the discrepancy flagged in the original review; it is an author clarification, not confirmation that a journal correction has been published.
A plausible alternative reading is that organisational support remains important for career satisfaction, but operates directly rather than by changing the effect of AI insecurity. The paper itself reports a sizeable positive association between support and satisfaction. Longitudinal data are needed to distinguish protection, selection and changing perceptions.
An AE interpretation: support must alter what happens next
The following is my interpretation, not the authors’ theoretical claim. Autopoietic ecology asks whether an activity renews the capacities needed for later activity. In this case, supportive communication is not enough if AI adoption still reallocates work, discretion and opportunity without employee influence. A conversation may reduce anxiety while leaving the next procurement, redesign or promotion decision untouched.
EdgeLab’s research programme distinguishes recognising a burden from allowing it to change a later allocation. Workers may report uncertainty and receive wellbeing support, yet the burden of adapting can still be transferred to them individually. The organisation continues to gain productivity while each worker must somehow preserve their own employability.
The AE Concept Guide therefore suggests a practical test: what decision changes after insecurity becomes visible? Evidence of effective support would include paid learning time, influence over task design, protected routes into new roles and explicit limits on displacement—not only higher scores on perceived support. The paper does not prove this AE account, but it identifies the gap such research should follow.
What should be tested next?
A stronger study would follow employees before, during and after a specific AI introduction. It would record what technology changed, who influenced implementation, which resources were allocated and how roles, pay, learning time and career movement changed. Comparing generic reassurance with participatory job redesign and protected development opportunities would test whether support becomes consequential.
Matias and Pinto provide a valuable warning: an employee can feel cared for while remaining unable to see a viable professional future. The next step is to move from perceived support to observable organisational commitments—and to ask whether workers’ accounts of uncertainty can alter the arrangements producing it.
Reference
Matias, J., and Pinto, J. C. (2026). “The rise of AI in the workplace: why perceived organizational support fails to protect long-term career satisfaction.” Frontiers in Artificial Intelligence, 9, 1891537. Published 18 September 2026. https://doi.org/10.3389/frai.2026.1891537.
Leave a Reply