Employee engagement

Employee engagement is a useful case for the wiki because it shows how quickly a practitioner-facing people construct can become operationally important before its nomological network, measurement model, and causal claims are clean.

2014-briner-employee-engagement-evidence-based argues that the popular HR concept of employee engagement suffers from five linked problems: no agreed definition, weak or overlapping measures, limited distinctiveness from older constructs, little public causal evidence on performance effects or interventions, and frequent over-claiming by advocates. The important move is not “engagement does not matter”; it is that the field often acts as if it knows what engagement is and what it causes before the construct and evidence base can bear that weight.

For AI workforce transformation, this matters as a warning. If organizations start measuring “AI engagement,” “AI readiness,” or “AI adoption mindset” with similarly loose construct boundaries, people analytics will reproduce the same problem under newer labels.

2025-brynjolfsson-generative-ai-at-work gives a concrete AI-era worker-outcome case adjacent to engagement but not reducible to it. In a customer-support rollout, AI assistance increased productivity and also improved parts of the interaction climate: customers were more polite and less likely to ask to speak to a manager. The source is useful because it keeps worker experience tied to observable work events rather than a broad, loosely bounded engagement label.

2026-reich-work-design-multidimensional-ai-threat extends the warning to AI adoption constructs. Its use of autonomy, skill variety, task significance, feedback, and differentiated AI-threat dimensions is useful precisely because it avoids treating “AI engagement” or “AI mindset” as a single broad attitude. The construct lesson is to decompose AI-mediated worker experience into job resources, identity threats, perceived work changes, and actual use behaviors.

2026-vuong-genai-job-performance is useful because it studies academic work engagement as a JD-R mediator rather than treating engagement as a loose practitioner label. GenAI usage predicted work engagement (β = 0.478, p = 0.000), and work engagement partially mediated the GenAI usage–job performance relationship (indirect β = 0.176, p = 0.000), while job crafting operated as a separate mediator. The construct-validity lesson is to keep motivational state, proactive work redesign, tool usage, and performance distinct even when the empirical model links them.

Connections

• Relates to construct-validity because engagement illustrates what happens when a construct’s definition, measures, and claimed outcomes drift apart.
• Relates to evidence-based-management because Briner’s critique is fundamentally about matching practical claims to the quality and relevance of available evidence.
• Relates indirectly to ai-adoption: adoption constructs will need sharper definition than engagement often had if they are to guide interventions rather than decorate dashboards.
• Relates to ai-mediated-work-experience because AI-era worker outcomes should be separated into autonomy, competence, identity, workload, learning, and relational experience rather than collapsed into engagement.
• Relates to human-ai-collaboration because AI-assisted support changed the emotional and communicative texture of the work, not just the volume of completed cases.
• Relates to ai-enabled-job-crafting because proactive work reshaping and work engagement may both mediate AI-performance claims, but they are different constructs with different intervention implications.

Contradictions & open questions

• Briner’s critique targets the popular HR practitioner concept of employee engagement, not the more specific academic construct of work engagement. The wiki should keep those separated rather than treating all engagement evidence as interchangeable.
• The source is from 2014. Later evidence reviews may strengthen, weaken, or refine the critique; this page should not freeze the engagement evidence base at Briner’s assessment.
• 2025-brynjolfsson-generative-ai-at-work measures specific interaction outcomes, not engagement as a construct. It should inform worker-experience claims without being treated as evidence that AI improves engagement broadly.
• 2026-reich-work-design-multidimensional-ai-threat measures adoption and depth of use, not engagement or well-being. It supports construct decomposition, not a claim that AI adoption improves employee experience.
• 2026-vuong-genai-job-performance strengthens the academic work-engagement side of the page, but its cross-sectional self-report design should not be read as causal evidence that GenAI increases engagement or performance.