Research gaps

Where the evidence base is still thin or missing. Each entry names what exists so far and what kind of evidence is still needed.

• Equilibrium labor demand. 2025-marguerit-augmenting-automating-labor now gives the wiki outcome evidence on new work, employment, and wages by automation versus augmentation exposure, but the wiki is still thin on firm-level mechanisms showing how organizations translate exposure into hiring, role redesign, skill demand, and substitution decisions.
• Worker experience & motivation under AI-mediated work is organized through ai-mediated-work-experience as an integrative parent domain, with narrower pages separating resources, strain, load, identity, learning, skill erosion, job crafting, teamwork, and network effects. 2026-edwards-developer-productivity adds a senior-developer account of autonomy–competence tension and raised productivity expectations, but the wiki still needs longitudinal and field-intervention evidence that measures multiple experience dimensions together and tests whether redesigned control rights, support routines, career paths, and digital-boundary practices improve the overall configuration rather than one outcome in isolation.
• AI as a job resource has been promoted into ai-as-a-job-resource through 2026-cogenli-generative-ai-conditional-job-resource and 2026-vuong-genai-job-performance. The wiki now distinguishes cognitive, structural/strategic, and psychosocial support functions from observed JD-R outcomes, but it needs multi-user, multi-occupation, longitudinal, and intervention evidence testing when verification, correction, traceability, and accountability work leave AI assistance as a net resource rather than a demand shift.
• Foundational I-O constructs remain mostly deferred to the triage-index shelf (“AI-relevant first, foundation later”), but competency-modeling and complex-collaborative-problem-solving have been promoted because they are directly adjacent to job analysis, skills architecture, AI-era work redesign, and the boundary between describing current work and assessing future-relevant capability.
• Human capital resource measurement has been promoted into human-capital-resource-measurement through 2023-zhang-human-capital-resources, but the wiki still needs evidence on how organizations should measure human capability, AI capability, and combined workflow capability as domain coverage converges.
• AI-enabled skills inference and skills architecture now has a first implementation case in 2025-tona-deployment-ai-infer-employee-skills, and 2025-teyssier-roberge-21st-century-skills-overlap adds a construct-validity warning about building evergreen skill architectures from overlapping 21st-century-skills labels. The wiki still needs independent validation evidence on inferred skills, employee reactions, fairness, downstream reskilling and mobility outcomes, and whether these systems improve workforce planning beyond making skills data more visible.
• Measurement of AI workforce impact has been promoted into ai-workforce-impact-measurement through 2025-tomlinson-working-with-ai, 2026-massenkoff-labor-market-impacts-ai adds observed-exposure evidence linking theoretical capability, Claude work-use traces, BLS projections, unemployment, and young-worker job starts, 2026-richmond-ai-jobs-transition-framework adds a transition-archetype frame combining exposure, human necessity, demand elasticity, and ChatGPT usage, and 2026-putka-indexing-ai-impact-onet proposes a job-analysis-centered 16-index architecture. The wiki still needs field evidence testing whether this richer measurement architecture predicts realized productivity, quality, job redesign, wages, staffing decisions, skill development, and worker experience better than simpler exposure or applicability measures.
• Worker-desired agency in AI-agent work is now represented through 2025-shao-future-work-ai-agents, but still needs live implementation evidence on whether agency matching improves adoption, performance, motivation, or trust.
• AI-supported knowledge synthesis has been promoted into ai-supported-knowledge-synthesis through 2025-yun-generative-ai-knowledge-work, but the wiki still needs field evidence on whether synthesis interfaces improve decision quality, learning, and team alignment over time.
• AI-mediated learning and cognitive debt has been split: ai-mediated-learning tracks whether AI-supported episodes build durable competence, while cognitive-debt now tracks delayed costs to attention, transfer, ownership, recovery, and judgment through 2025-kosmyna-brain-chatgpt-cognitive-debt, 2026-ruttenberg-cognitive-debt-ai-research, and 2026-ruttenberg-neurodivergent-expertise-ai-work. The wiki still needs workplace-longitudinal evidence on whether AI-supported learning designs preserve transfer, debugging skill, professional judgment, and expertise maintenance over time.
• Cognitive sustainability has been promoted into cognitive-sustainability through 2025-kabashkin-cognitive-atrophy-paradox, which supplies a conceptual, simulation-based Cognitive Sustainability Index linking autonomy, reflection, creativity, delegation, and reliance. The wiki now needs independent psychometric validation, longitudinal workplace evidence, and participatory intervention studies testing whether cognitive-sustainability measures predict learning, judgment, recovery, skill maintenance, and system performance without becoming worker surveillance.
• Human-AI co-creation modes has been promoted into human-ai-co-creation-modes through 2025-randazzo-cyborgs-centaurs-self-automators, but the wiki still needs longitudinal field evidence on whether Cyborg/Centaur/Self-Automator patterns persist across tasks, can be deliberately designed, and predict durable productivity, AI fluency, domain expertise, or skill erosion.
• AI-enabled job crafting has been promoted into ai-enabled-job-crafting through 2026-vuong-genai-job-performance, but the wiki still needs behavioral or field evidence on what workers actually change in tasks, collaboration, and performance criteria when they use GenAI to craft their jobs.
• Cognitive load in AI-assisted work has been promoted into cognitive-load-in-ai-assisted-work through 2026-lepine-precision-proactivity, but the wiki still needs validated load measures across other domains and evidence on which interface interventions reduce extraneous load without suppressing useful AI help.
• AI-use image concerns has been promoted into ai-use-image-concerns through 2026-almog-barriers-ai-adoption, but the wiki still needs field evidence from richer professional work and organizational designs that change whether AI use is read as good judgment, weak judgment, or compliance.
• Novice risk work / reverse mentoring for AI has been promoted into novice-risk-work through 2025-kellogg-novice-risk-work, but the wiki still needs evidence on which literacy, governance, and deployer-level supports let novice users teach seniors without over-relying on project-level routine changes.
• Agentic delegation has been promoted into agentic-delegation through 2026-zhu-choose-your-agent, and 2026-johnston-shift-agentic-ai-codex now adds large-scale observed-use evidence on delegated Codex work, concurrency, runtime, and reusable skills, but the wiki still needs outcome evidence on when autonomous execution improves team or organizational productivity, quality, learning, and accountability.
• Domain expertise in agentic work has been promoted into domain-expertise-in-agentic-work through 2026-hitzig-agentic-coding-expertise, but the wiki still needs field evidence on whether transcript-coded expertise predicts durable output quality, whether novices can develop expertise while delegating implementation, and how organizations should credential domain experts who now direct technical agent work.
• AI-mediated choice and identity has been promoted into ai-mediated-choice-and-identity through 2025-matz-basic-effect, but the wiki still needs workplace evidence on whether delegated AI choices narrow exploration, identity expression, developmental variety, or role agency in real workflows.
• Moral boundaries of AI automation has been promoted into moral-boundaries-of-ai-automation through 2025-friis-performance-principle, but the wiki still needs field evidence on whether public moral repugnance translates into employer behavior, customer acceptance, professional norms, regulation, or durable protection from substitution.
• Human-AI task allocation is proposed for promotion through 2026-zhang-fatigue-aware-learning-defer, but the wiki still needs field evidence on allocation policies that use real human-state signals such as fatigue, workload, skill, confidence, or accountability rather than simulated benchmark states.
• Responsible AI deployment has been promoted into responsible-ai-deployment through 2025-landers-ethical-use-ai-iop, but the wiki still needs empirical evidence on governance arrangements, stakeholder participation, monitoring routines, and reversal mechanisms that prevent worker-facing AI from becoming unchecked efficiency infrastructure.
• Expert authority in AI-mediated decisions has been promoted into expert-authority-in-ai-decisions through 2025-mayer-let-me-explain, which shows that experts may retain responsibility for client-facing explanation after predictive AI takes over core decisions. The wiki still needs comparative and outcome evidence on whether explanation access, discretionary authority, appeal routes, and client relationships affect understanding, procedural justice, substantive decision quality, professional identity, and the risk that experts become legitimacy buffers for opaque systems.
• Agent benchmark validity has been promoted into ai-agent-benchmark-validity through 2026-wang-agent-development-real-world-work, but the wiki still needs field evidence that more representative benchmarks lead to better deployment, redesign, or governance decisions.
• AI-mediated teamwork has been promoted into ai-mediated-teamwork through 2026-dellacqua-cybernetic-teammate, but the wiki still needs evidence on coordination, trust, conflict, mentoring, implementation, and whether organizations actually redesign team boundaries after AI can reproduce some bounded team benefits.
• AI-mediated organizational networks has been promoted into ai-mediated-organizational-networks through 2026-buechsenschuss-genai-organizational-networks, 2025-hautamaki-fully-leveraging-ai-b2b-sales adds qualitative managerial mechanisms for cross-functional AI knowledge transfer, and 2007-serenko-organizational-size-knowledge-flow now gives a pre-AI baseline for why size should degrade knowledge flow. The wiki still needs digital-trace, multi-organization, and longer-term evidence on whether GenAI weakens organizational-size penalties, whether centrality persists, and when improved reach becomes overload for key actors.
• Behavioral human-centered AI has been promoted into behavioral-human-centered-ai through 2025-schweitzer-leading-ai-adoption, and employee-change-support now gives the wiki a foundational change-support mechanism through 2018-fugate-change-compliance-championing, but the wiki still needs AI-specific field evidence on whether behaviorally informed change management improves trust calibration, worker outcomes, governance quality, championing, or realized AI value.
• Organizational learning capability has been promoted into organizational-learning-capability through 1997-goh-benchmarking-learning-capability, but the wiki still needs AI-specific longitudinal evidence on whether learning-capability conditions predict which experiments become redesigned workflows, shared practices, and measurable organizational outcomes.
• Cognitive surrender is proposed for promotion through 2026-shaw-thinking-fast-slow-artificial, but the wiki still needs workplace evidence on whether AI answer adoption undercuts judgment in richer expert tasks with accountability, domain context, and independent verification channels.