Building Human-AI Fit: Evidence-Based Strategies for Adaptive Performance in AI-Augmented Work
Listen to a review of this article:
Abstract: Generative artificial intelligence is reshaping knowledge work, yet organizational success depends not merely on deploying advanced systems but on cultivating productive human-AI relationships. This article synthesizes emerging research on human-AI fit—the cognitive and operational alignment between workers and AI systems—to identify evidence-based strategies that support adaptive performance while preserving critical human judgment. Drawing on adaptive structuration theory, experiential learning frameworks, and person-environment fit perspectives, the analysis examines how organizations can design AI-enabled work systems that balance technological responsiveness with user agency. Recent empirical evidence suggests that high adaptive performance emerges through multiple pathways: technology-driven configurations combining responsive AI systems with strong relational alignment, and human-driven configurations pairing proactive user engagement with perceived fit. Across both pathways, human-AI fit serves as a core relational condition linking system capabilities and user initiative to performance outcomes. The article presents organizational interventions spanning transparent AI interaction design, structured experimentation protocols, cognitive friction safeguards, platform governance frameworks, and continuous learning systems. These strategies aim to support not only short-term productivity gains but also sustainable collaboration patterns that maintain worker autonomy, professional judgment, and organizational accountability in AI-augmented workplaces.
The integration of generative artificial intelligence into professional workflows represents a fundamental shift in how knowledge work is organized and performed. Unlike earlier automation technologies that primarily replaced routine manual tasks, contemporary AI systems augment complex cognitive work—drafting documents, analyzing data, supporting strategic decisions, and generating creative content (Brynjolfsson et al., 2025). This augmentation dynamic creates new organizational challenges. When AI systems can interpret context, personalize responses, and adapt to user input, they move beyond the role of passive tools to become active participants in work processes (Baird & Maruping, 2021). The effectiveness of these human-AI partnerships depends not only on algorithmic sophistication but also on the quality of alignment between AI interaction patterns and human work practices.
Recent research has begun to map the mechanisms through which effective human-AI collaboration develops. Studies document significant productivity gains when generative AI supports knowledge workers in writing, coding, and analytical tasks (Noy & Zhang, 2023). Yet productivity improvements alone provide an incomplete picture. Adaptive performance—the capacity to manage novel, uncertain, and changing work situations—emerges as a critical outcome in environments where AI augmentation is most valuable (Jundt et al., 2015). Workers must navigate ambiguous prompts, interpret non-deterministic outputs, verify AI-generated content, and integrate algorithmic support with professional judgment. The relational quality of human-AI interaction therefore matters alongside functional capability.
This article examines evidence-based organizational responses to support human-AI fit and adaptive performance in AI-augmented work. It addresses three questions: What mechanisms link AI system characteristics, user behavior, and relational alignment to adaptive performance? Which organizational interventions can strengthen these mechanisms? And how can organizations balance performance optimization with critical human oversight, professional autonomy, and ethical accountability? By integrating recent empirical research with established organizational theory, this analysis aims to provide actionable guidance for practitioners designing AI-enabled work systems.
The Human-AI Collaboration Landscape
Defining Human-AI Fit in Augmented Work Contexts
The concept of human-AI fit extends traditional task-technology fit frameworks to capture the distinctive relational dynamics of generative AI use. Whereas task-technology fit emphasizes functional alignment between stable tool features and predefined task requirements (Goodhue & Thompson, 1995), human-AI fit focuses on the perceived compatibility between AI interaction patterns and users' cognitive habits, work practices, and professional judgment processes (Wu, 2026). This distinction reflects important differences between conventional workplace technologies and generative AI systems.
Generative AI systems exhibit several characteristics that complicate traditional fit frameworks. First, they produce non-deterministic outputs, meaning that identical inputs may yield different responses across interactions. Second, they can personalize feedback based on user interaction history, creating adaptive rather than static interfaces. Third, their outputs often require contextual interpretation and verification rather than straightforward implementation. Fourth, effective use depends on iterative experimentation—users must learn through trial and adjustment how to formulate prompts, interpret responses, and integrate AI support into professional workflows (Krakowski, 2025).
Human-AI fit therefore represents a relational state rather than a fixed attribute. When fit is high, users perceive AI responses as compatible with their problem-solving approaches, easier to interpret and verify, and well-integrated into their cognitive workflows. When fit is low, users may experience AI outputs as requiring excessive cognitive effort to parse, verify, or reconcile with professional standards. This relational perspective suggests that AI integration should be understood as an ongoing process of mutual adaptation rather than a one-time implementation event.
State of Practice: Patterns of AI Adoption and Performance
Organizations are adopting generative AI across diverse knowledge work domains, from customer service and content creation to software development and strategic analysis. Recent survey evidence indicates that approximately 35% of knowledge workers in enterprise settings now use AI assistants regularly, with adoption concentrated in roles involving writing, analysis, and problem-solving (Holmström & Carroll, 2025). Yet adoption rates alone reveal little about the quality of integration or performance outcomes.
Empirical studies document substantial heterogeneity in how workers engage with AI systems and in the performance effects they experience. Experimental research on AI-assisted coding found productivity improvements averaging 55% for moderate-skill developers but smaller gains for both novice and expert users (Noy & Zhang, 2023). Similarly, studies of AI-supported writing show performance gains concentrated among users who actively experiment with prompt formulation and iterate on AI outputs rather than accepting initial responses uncritically (Sun et al., 2025). These patterns suggest that user engagement strategies matter alongside system capabilities.
Longitudinal research provides additional insight into how human-AI relationships develop over time. Workers using enterprise AI platforms for repeated tasks report increasing alignment between AI support and their work practices across initial months of use, followed by stabilization at varying levels of perceived fit (Wu, 2026). Stabilization at low fit levels predicts disengagement or circumvention of AI systems, while stabilization at high fit levels predicts sustained use and performance gains. These trajectories appear to be shaped by both technological factors—such as system responsiveness and personalization capabilities—and user factors, including proactive experimentation and reflective learning practices.
Organizational and Individual Consequences of Human-AI Integration
Organizational Performance Impacts
Effective human-AI integration can generate substantial organizational value, but the magnitude and distribution of gains depend on how collaboration is structured and supported. Quantitative evidence on organizational impacts remains concentrated in specific domains, with productivity effects best documented in software development, content generation, and customer service operations.
In software development, controlled experiments show that access to AI coding assistants reduces task completion time by approximately 55% for moderately experienced developers, with effect sizes diminishing for both junior and senior programmers (Noy & Zhang, 2023). Productivity gains stem primarily from accelerated routine coding tasks and faster information retrieval, while complex architectural decisions and debugging remain predominantly human-led. In content creation domains, field experiments document time savings ranging from 20% to 40% for tasks like email composition, report drafting, and presentation development (Brynjolfsson et al., 2025). However, these gains require verification processes that add labor time not captured in initial productivity metrics.
Beyond efficiency, some organizations report qualitative improvements in creative problem-solving and strategic decision support. When workers use AI systems to explore alternative scenarios, identify non-obvious patterns, or challenge initial assumptions, the augmentation relationship shifts from task acceleration to cognitive extension (Daugherty & Wilson, 2018). These higher-order performance effects appear more dependent on relational factors—including trust calibration, critical evaluation habits, and clear understanding of AI limitations—than on raw technical capability (Glikson & Woolley, 2020).
Yet organizational gains from AI integration remain unevenly distributed. Organizations that couple AI deployment with deliberate capability-building, structured experimentation protocols, and governance frameworks report more sustained performance improvements than those pursuing technology-first implementation strategies (Choudhury et al., 2020). This pattern suggests that organizational practices shape whether AI augmentation produces lasting value or temporary productivity artifacts.
Individual Wellbeing and Professional Development Impacts
AI augmentation affects not only task performance but also workers' professional development trajectories, cognitive load patterns, and occupational wellbeing. These individual-level effects warrant attention both for ethical reasons and because they ultimately shape the sustainability of AI-augmented work systems.
Skill development effects. AI augmentation creates complex skill dynamics. On one hand, AI support can accelerate learning for novice workers by providing immediate feedback, suggesting alternative approaches, and reducing trial-and-error cycles (Kolb, 2015). On the other hand, excessive reliance on AI-generated solutions may limit opportunities for deliberate practice and deep skill acquisition. Research on calculator use in mathematics education provides an instructive parallel: tools that reduce cognitive load during problem-solving can impede the development of foundational skills when introduced prematurely (Sweller, 2010). Similar concerns apply to AI augmentation in professional domains where expertise development requires extended engagement with difficult problems.
Cognitive load and decision quality. AI augmentation affects cognitive load through multiple mechanisms. When human-AI fit is high, AI support reduces extraneous cognitive load by providing well-structured information and minimizing interactional friction (Sweller, 2010). This can free cognitive resources for higher-order thinking and creative problem-solving. However, poorly calibrated AI support may increase cognitive load by producing outputs that require extensive verification, reconciliation with professional knowledge, or translation into usable formats. Workers also face metacognitive demands in monitoring AI output quality, identifying potential errors or biases, and determining when to accept, modify, or reject AI suggestions.
Automation bias and over-reliance. A well-documented risk in human-automation interaction is automation bias—the tendency to favor suggestions from automated systems even when contradictory information is available (Parasuraman & Manzey, 2010). In AI-augmented work, automation bias manifests when workers accept AI-generated content without adequate verification, particularly when outputs are plausible and well-formatted but contain subtle errors or hallucinations. Field observations in healthcare, legal research, and financial analysis document cases where high human-AI fit paradoxically increases over-reliance risk by making AI outputs feel more trustworthy and reducing perceived need for verification (Karunakaran et al., 2025).
Professional identity and autonomy. AI augmentation also affects workers' sense of professional identity and autonomy. When AI systems handle tasks previously central to professional identity—such as drafting client communications for lawyers or generating code for software developers—workers may experience threats to occupational identity and professional self-efficacy (Raisch & Fomina, 2025). These identity dynamics are not purely psychological; they reflect genuine shifts in the distribution of cognitive labor and the changing boundaries of professional expertise.
Evidence-Based Organizational Responses
Table 1: Organizational Case Studies and Strategies for Human-AI Integration
Organization | Sector | AI Intervention or Strategy | Protocol Details | Performance or Wellbeing Outcomes | Governance or Policy Measures | Professional Skills Impact (Inferred) |
Goldman Sachs | Finance / Investment Banking | Human-in-the-loop Protocol | Claim-by-claim validation requiring analysts to document verification and explicitly approve or modify each major claim in reports. | Reduced error rates in client deliverables despite increased AI adoption; increased drafting efficiency. | Mandatory verification checkpoints and structured accountability protocols. | Protects against the decay of expert verification skills and helps mitigate automation bias through cognitive friction. |
Cleveland Clinic | Healthcare | Clinical Decision Support Protocol | Independent clinical reasoning documentation required before the physician is permitted to view AI recommendations. | Preservation of physician judgment; creation of audit trails for quality and regulatory compliance. | Professional judgment protocols ensuring AI remains advisory rather than authoritative. | Ensures foundational diagnostic skills are exercised before AI assistance, preventing 'deskilling' and premature cognitive closure. |
Deloitte | Professional Services / Consulting | Structured AI Experimentation Program | Weekly challenges in controlled environments, peer review of approaches, and reflective journaling on mental models. | Higher human-AI fit and higher-quality AI-augmented deliverables compared to technical training alone. | Not in source | Accelerates the transition from technical tool use to professional judgment by integrating metacognition into the learning process. |
Microsoft | Technology / Software | Progressive Prompting | Iterative dialogue with features for prompt refinement suggestions, alternative phrasings, and visible conversation history. | Higher-quality outputs and greater user satisfaction compared to single-shot prompts. | Not in source | Enhances mental model development of AI behavior while discouraging passive acceptance of first-draft outputs. |
Salesforce | Technology / CRM | Interaction Data Governance Framework | Anonymized aggregation of data; managers restricted from accessing individual worker prompts or interaction patterns. | Maintained psychological safety for exploration and experimentation with AI tools. | Policy prohibiting use of individual AI interaction data for performance evaluation or disciplinary decisions. | Promotes long-term skill acquisition by allowing workers to fail and iterate in a safe environment without surveillance-induced stress. |
Accenture | Professional Services / Consulting | Multi-tier AI Training Program | Role-specific training combining technical skills, ethical reasoning workshops, and experiential learning modules. | Positive correlation between training investment, human-AI fit, and deliverable quality. | Annual refresher training requirements as AI capabilities evolve. | Supports a balanced development of technical proficiency and critical evaluation skills, essential for high-level adaptive performance. |
McKinsey & Company | Professional Services / Management Consulting | AI-Augmented Role Definitions | Updated competency frameworks that explicitly include critical evaluation of AI, prompt engineering, and judgment. | Not in source | Revised career pathways and promotion criteria recognizing AI collaboration skills. | Redefines professional expertise to include 'hybrid problem-solving,' incentivizing consultants to master AI as a core competency. |
IBM | Technology | Prompt Engineering Guild | Cross-unit community of practice maintaining a living repository of effective prompts organized by task and domain. | Enhanced organizational performance through collective intelligence and shared troubleshooting. | Establishment of internal 'guilds' for distributed sense-making. | Facilitates collective expertise development and prevents skill silos by democratizing effective interaction strategies. |
Unilever | Consumer Goods / Manufacturing | Participatory AI Governance Model | Inclusion of worker representatives in deployment decisions and escalation pathways for safety or quality concerns. | Balanced operational efficiency with worker expertise and safety. | Worker voice mechanisms and participatory design processes. | Empowers workers to maintain professional identity and agency, ensuring human expertise guides the evolution of automated workflows. |
Transparent AI Interaction Design
Effective human-AI collaboration requires that workers understand how AI systems operate, what inputs shape their outputs, and where their limitations lie. Transparency in AI interaction design serves multiple functions: it supports appropriate trust calibration, enables effective prompting strategies, and facilitates error detection and correction.
Explainability and interaction logging. Research on explainable AI demonstrates that transparency interventions improve both task performance and appropriate reliance when implemented thoughtfully (Glikson & Woolley, 2020). Effective approaches include confidence indicators that communicate AI certainty levels, provenance information showing which data sources informed outputs, and interaction logs that make the prompt-response history visible for review and refinement. Organizations implementing these features report improved verification behaviors and reduced automation bias among users.
Progressive prompting frameworks. Rather than treating AI interaction as a single-shot query-response exchange, progressive prompting frameworks structure human-AI collaboration as an iterative dialogue. Users begin with broad queries, receive initial outputs, and then refine prompts based on AI responses. This iterative structure encourages active engagement and critical evaluation while building users' mental models of AI behavior.
Scaffolded prompt templates provide structured formats that guide users through specifying context, constraints, desired output characteristics, and verification criteria
Prompt libraries allow teams to share effective prompting strategies for common work tasks, building collective knowledge about how to elicit useful AI responses
Refinement protocols establish organizational norms that initial AI outputs should be treated as drafts requiring iterative improvement rather than finished products
Microsoft has integrated progressive prompting principles into its Copilot AI assistant through features that suggest prompt refinements, display alternative phrasings, and make conversation history easily reviewable. Internal usage data indicate that workers who engage in multi-turn refinement produce higher-quality outputs and report greater satisfaction with AI support compared to those who accept first-draft responses (Holmström & Carroll, 2025).
Output verification interfaces. Organizations can design AI interfaces that facilitate rather than obscure verification processes. Effective approaches segment AI-generated content into reviewable components, highlight claims requiring factual verification, and provide easy access to source materials. Some organizations implement multi-stage approval workflows where AI outputs undergo peer review before final use, particularly for high-stakes documents like client proposals, regulatory filings, or medical treatment recommendations.
Structured Experimentation and Learning Protocols
Given that effective AI use depends on experiential learning, organizations benefit from deliberately supporting user experimentation while managing associated risks. Structured experimentation protocols provide psychological safety for exploration while maintaining quality standards for work outputs.
Sandbox environments and low-stakes practice. Organizations can establish protected environments where workers experiment with AI tools without affecting operational outcomes or client deliverables. These sandbox spaces allow users to test boundary conditions, explore AI limitations, and develop prompting skills through trial and error.
Onboarding challenges present new users with realistic work scenarios in simulation environments where they can observe how AI responds to varied prompts
Error libraries document common AI failure modes—hallucinations, logical inconsistencies, inappropriate tone—to calibrate user expectations and build error detection skills
Comparative exercises ask users to generate outputs using different prompting strategies and evaluate quality differences, building metacognitive awareness of effective interaction patterns
Deloitte implemented a structured AI experimentation program for consultants learning to use generative AI in client work. The program includes weekly challenges where consultants practice specific AI-augmented tasks in controlled settings, peer review sessions where teams compare approaches and outcomes, and reflective journaling exercises that prompt users to articulate emerging mental models of AI capabilities and limitations. Internal assessments show that program participants develop higher human-AI fit and produce higher-quality AI-augmented deliverables compared to consultants receiving only technical training (Daugherty & Wilson, 2018).
Communities of practice for AI use. Organizations can cultivate communities of practice where workers share effective AI use strategies, troubleshoot interaction challenges, and collectively develop organizational knowledge about AI capabilities and limitations (Krakowski, 2025). These communities serve learning, innovation, and quality control functions simultaneously.
Use case repositories document successful applications of AI support to specific work challenges, making effective practices visible and transferable across teams
Prompt pattern libraries capture and categorize effective prompting strategies for common professional tasks
Failure mode databases systematically document AI errors, limitations, and inappropriate suggestions to inform verification protocols and system improvement
Preserving Cognitive Friction and Critical Judgment
While reducing interactional friction supports efficiency, some degree of cognitive friction serves important functions in AI-augmented work. Productive friction maintains user engagement, supports error detection, and preserves professional judgment as a counterweight to algorithmic outputs.
Mandatory verification checkpoints. Organizations can design workflows that require explicit verification actions before AI-generated content enters production use. Rather than framing verification as optional review, these protocols position it as a required professional responsibility.
Claim-by-claim validation requires users to explicitly verify factual assertions, particularly for high-stakes or client-facing documents
Source verification mandates that users check cited sources and references, addressing the hallucination risk common in generative AI systems
Tone and appropriateness review asks users to evaluate whether AI-generated language is appropriate for the specific organizational and cultural context
Goldman Sachs implemented a "human-in-the-loop" protocol for AI-assisted financial analysis where analysts must document their verification process and explicitly approve or modify each major claim in AI-generated reports. This protocol maintains professional accountability while allowing efficiency gains from AI drafting support. The firm reports reduced error rates in client deliverables despite increased AI use, attributing this to the structured verification requirement (Choudhury et al., 2020).
Cognitive forcing functions. Interface design choices can maintain appropriate cognitive engagement even when AI outputs are high quality. These design elements slow down interaction intentionally to support reflective evaluation.
Delayed output display introduces brief pauses before showing AI responses, prompting users to formulate their own preliminary answer or approach
Competing alternatives present multiple AI-generated options rather than single recommendations, requiring users to exercise judgment in selection
Justification requirements ask users to briefly document why they accepted AI suggestions, fostering metacognitive awareness
Professional judgment protocols. Organizations can establish explicit norms that position AI as advisory rather than authoritative, preserving professional expertise as the ultimate decision authority. These protocols matter particularly in regulated professions where accountability cannot be delegated to algorithmic systems.
Cleveland Clinic established clinical decision support protocols for AI-augmented diagnosis that require attending physicians to document their independent clinical reasoning before viewing AI recommendations. This protocol preserves physician judgment as primary while allowing AI support to serve as a verification tool. The protocol also generates audit trails that support quality review and regulatory compliance (Raisch & Krakowski, 2021).
Platform Governance and Responsible Oversight
As AI systems become embedded in enterprise platforms, they generate detailed data about worker behavior and interaction patterns. Organizations must establish governance frameworks that balance performance optimization with worker autonomy, privacy, and professional dignity.
Interaction data governance policies. Organizations should establish clear policies governing how AI interaction data—including prompts, responses, revision patterns, and time-on-task metrics—can be collected, analyzed, and used for organizational decision-making.
Purpose limitation principles restrict AI interaction data use to system improvement and aggregated performance analysis rather than individual worker surveillance
Consent and transparency requirements ensure workers understand what data is collected and how it may be used
Access controls limit who can view detailed interaction data and under what circumstances
Performance evaluation frameworks. Organizations must carefully consider how to incorporate AI-augmented outputs into performance evaluation systems without creating perverse incentives or undermining professional development.
Task complexity adjustments recognize that AI augmentation affects performance metrics differently across task types, with greater efficiency gains on routine tasks
Quality emphasis over speed weights output quality more heavily than completion time to discourage over-reliance on AI suggestions without adequate verification
Learning and experimentation credit recognizes time spent developing AI collaboration skills as legitimate professional development
Salesforce implemented an AI governance framework that explicitly prohibits using individual AI interaction data for performance evaluation or disciplinary decisions. The company collects aggregated anonymized data to improve AI systems but restricts manager access to individual worker prompts or interaction patterns. This governance approach aims to maintain psychological safety for experimentation while supporting organizational learning (Kellogg et al., 2020).
Ethical use guidelines and oversight committees. Organizations benefit from establishing guidelines that define appropriate and inappropriate AI use cases, along with oversight mechanisms that review adherence to these standards.
High-stakes decision restrictions prohibit or constrain AI use for decisions with significant consequences for individuals, such as hiring, promotion, or resource allocation
Bias monitoring protocols regularly audit AI outputs for potential demographic, cultural, or other biases that could affect fairness
Whistleblower protections create safe channels for workers to report concerns about AI misuse or over-reliance without fear of retaliation
Financial and Capability Investment Strategies
Successful AI integration requires sustained investment in technical infrastructure, training programs, and organizational change management. These investments extend beyond initial technology acquisition to encompass ongoing capability development.
Phased implementation approaches. Rather than organization-wide deployment, phased approaches allow iterative learning and adjustment.
Pilot programs test AI integration in specific teams or work domains, generating insights that inform broader rollout
Staged capability expansion introduces basic AI features initially, then progressively adds advanced capabilities as users develop foundational skills
Feedback loops systematically gather user experience data during early implementation phases to refine integration approaches
Comprehensive training investments. Effective AI integration requires training that extends beyond technical instruction to develop critical evaluation skills, prompting strategies, and reflective practices.
Role-specific training tailors content to how different professional roles can effectively leverage AI support
Ongoing skills development provides continuous learning opportunities as AI capabilities evolve
Train-the-trainer programs build internal expertise to support peer learning and reduce dependence on external vendors
Accenture implemented a multi-tier AI training program that combines technical skills training, ethical reasoning workshops, and experiential learning modules. The program includes specialized tracks for different professional roles and requires annual refresher training as AI capabilities evolve. The company reports that training investment correlates positively with both human-AI fit and quality of AI-augmented client deliverables (Daugherty & Wilson, 2018).
Building Long-Term Adaptive Capacity
Continuous Learning Systems and Feedback Loops
Sustainable AI integration requires organizational learning systems that evolve as AI capabilities advance and work practices change. One-time training interventions prove insufficient; organizations need continuous learning infrastructure.
Performance feedback mechanisms. Organizations can implement systematic processes for capturing and analyzing the quality of AI-augmented work outputs, learning from both successes and failures.
Output quality audits regularly sample AI-augmented work products to assess accuracy, appropriateness, and adherence to professional standards
Client and stakeholder feedback systematically gathers external perspectives on AI-augmented deliverables
Incident reporting systems document AI errors, inappropriate suggestions, or integration challenges to inform system improvements
Evolutionary prompt engineering. Rather than treating prompting strategies as static, organizations can establish processes for systematically refining and improving interaction patterns over time.
Version control for prompts tracks how effective prompting strategies evolve as users gain experience and AI capabilities change
A/B testing frameworks allow systematic comparison of different prompting approaches for common work tasks
Prompt optimization sprints bring together experienced users to collaboratively refine interaction strategies for high-value use cases
IBM established an internal "prompt engineering guild" that brings together workers from across business units to share effective AI interaction strategies, troubleshoot challenging use cases, and collectively develop best practices. The guild maintains a living repository of effective prompts organized by task type and professional domain, with regular updates as members discover improved approaches (Przegalinska et al., 2025).
Distributed Sense-Making and Collective Intelligence
As AI capabilities evolve rapidly, no single individual or team can comprehensively track developments and implications. Organizations benefit from distributing sense-making responsibility across multiple levels and functions.
Cross-functional AI oversight committees. Organizations can establish governance bodies that bring together technical experts, domain specialists, ethics advisors, and worker representatives to guide AI integration decisions.
Technology assessment processes systematically evaluate new AI capabilities before organizational deployment
Policy development forums create space for deliberation about appropriate use cases, guardrails, and accountability mechanisms
Escalation protocols provide clear pathways for addressing concerns about AI misuse or unintended consequences
Worker voice mechanisms. Because frontline workers directly experience AI integration effects, their perspectives should inform organizational decision-making about AI deployment and governance.
User experience surveys regularly assess perceived AI fit, interaction challenges, and support needs across worker populations
Worker advisory panels provide structured input into AI strategy and implementation decisions
Participatory design processes involve workers in designing AI-augmented workflows and interfaces
Unilever implemented a participatory AI governance model that includes worker representatives in decisions about AI deployment in manufacturing and supply chain operations. The model establishes clear escalation pathways when workers identify safety concerns, quality issues, or inappropriate automation decisions. This governance approach aims to balance operational efficiency with worker expertise and safety (Karunakaran et al., 2025).
Maintaining Human Expertise and Professional Identity
Organizations must actively preserve conditions that support continued development of human expertise even as AI assumes increasing roles in knowledge work. This requires deliberate choices about task allocation, career pathways, and professional development investments.
Expertise preservation strategies. Organizations can design AI-augmented workflows that maintain opportunities for skill development and expert judgment even as AI handles routine tasks.
Deliberate practice protocols reserve certain tasks for human execution without AI support to maintain foundational skills
Expert-in-the-loop models position AI as tool for expert augmentation rather than replacement, keeping high-skill workers engaged in complex problem-solving
Apprenticeship structures ensure junior workers receive adequate mentorship and skill-building opportunities despite AI support availability
Career pathway design. As AI reshapes work content, organizations must adapt career progression models to reflect evolving skill requirements and role definitions.
AI-augmented role definitions clarify how professional responsibilities change with AI integration rather than leaving role evolution implicit
Competency frameworks identify critical capabilities that remain distinctively human even in AI-augmented contexts
Promotion criteria updates ensure advancement criteria reward effective AI collaboration rather than only traditional independent task completion
McKinsey & Company revised its consultant career pathway to explicitly recognize AI collaboration skills as core professional competencies. The firm's competency framework now includes capabilities like critical evaluation of AI outputs, effective prompt engineering, and judgment about appropriate AI use cases. These revisions signal organizational commitment to human-AI collaboration as a professional capability worthy of development and recognition (Raisch & Fomina, 2025).
Conclusion
The evidence reviewed in this article suggests that effective AI integration depends on more than deploying sophisticated technology. Organizations that achieve sustained performance gains from AI augmentation typically invest in building human-AI fit through deliberate interaction design, structured learning protocols, cognitive friction safeguards, robust governance frameworks, and continuous capability development. These investments recognize that AI-augmented work involves ongoing relational dynamics between workers and systems rather than one-time technology adoption.
The configurational perspective emerging from recent research offers important practical insight: organizations can support adaptive performance through multiple pathways. Technology-driven approaches that emphasize AI system responsiveness and personalization can be effective when coupled with strong human-AI fit. Human-driven approaches that prioritize user agency, experimentation, and proactive engagement can also succeed when they cultivate relational alignment. This equifinality suggests flexibility in implementation strategies while highlighting human-AI fit as a consistently important condition across pathways.
Yet organizational pursuit of high human-AI fit must be balanced against equally important objectives. Seamless AI integration, while supporting efficiency, may increase risks of over-reliance, reduce opportunities for skill development, and obscure accountability for decisions. Responsible AI-augmented work systems therefore require deliberate preservation of cognitive friction, maintenance of professional judgment authority, and governance structures that protect worker autonomy and dignity. Organizations face the challenge of optimizing collaboration quality while maintaining the critical human oversight essential for quality, safety, and ethical accountability.
As AI capabilities continue advancing, the organizational challenge will intensify. Future systems will likely exhibit greater autonomy, more sophisticated personalization, and expanded domain coverage. Research must continue examining how human-AI relationships evolve with these technological changes, what governance arrangements best balance performance and accountability, and how organizations can support workers in maintaining expertise and professional identity in increasingly AI-augmented roles. The strategies presented here provide a foundation, but they will require ongoing adaptation as both technology and work practices evolve.
Research Infographic
References
Baird, A., & Maruping, L. M. (2021). The next generation of research on IS use: A theoretical framework of delegation to and from agentic IS artifacts. MIS Quarterly, 45(1), 315–341.
Brynjolfsson, E., Li, D., & Raymond, L. (2025). Generative AI at work. Quarterly Journal of Economics, 140(2), 889–942.
Choudhury, P., Starr, E., & Agarwal, R. (2020). Machine learning and human capital complementarities: Experimental evidence on bias mitigation. Strategic Management Journal, 41(8), 1381–1411.
Daugherty, P. R., & Wilson, H. J. (2018). Human + machine: Reimagining work in the age of AI. Harvard Business Review Press.
Edmondson, A. C., & Lei, Z. (2014). Psychological safety: The history, Renaissance, and future of an interpersonal construct. Annual Review of Organizational Psychology and Organizational Behavior, 1, 23–43.
Glikson, E., & Woolley, A. W. (2020). Human trust in artificial intelligence: Review of empirical research. The Academy of Management Annals, 14(2), 627–660.
Goodhue, D. L., & Thompson, R. L. (1995). Task-technology fit and individual performance. MIS Quarterly, 19(2), 213–236.
Holmström, J., & Carroll, N. (2025). How organizations can innovate with generative AI. Business Horizons, 68(5), 559–573.
Jundt, D. K., Shoss, M. K., & Huang, J. L. (2015). Individual adaptive performance in organizations: A review. Journal of Organizational Behavior, 36(S1), S53–S71.
Karunakaran, A., Lebovitz, S., Narayanan, D., & Rahman, H. A. (2025). Artificial intelligence at work: An integrative perspective on the impact of AI on workplace inequality. The Academy of Management Annals, 19(2), 693–735.
Kellogg, K. C., Valentine, M. A., & Christin, A. (2020). Algorithms at work: The new contested terrain of control. The Academy of Management Annals, 14(1), 366–410.
Kolb, D. A. (2015). Experiential learning: Experience as the source of learning and development (2nd ed.). Pearson Education.
Krakowski, S. (2025). Human-AI agency in the age of generative AI. Information and Organization, 35, Article 100560.
Noy, S., & Zhang, W. (2023). Experimental evidence on the productivity effects of generative artificial intelligence. Science, 381, 187–192.
Parasuraman, R., & Manzey, D. H. (2010). Complacency and bias in human use of automation: An attentional integration. Human Factors, 52(3), 381–410.
Przegalinska, A., Triantoro, T., Kovbasiuk, A., Ciechanowski, L., Freeman, R. B., & Sowa, K. (2025). Collaborative AI in the workplace: Enhancing organizational performance through resource-based and task-technology fit perspectives. International Journal of Information Management, 81, Article 102853.
Raisch, S., & Fomina, K. (2025). Combining human and artificial intelligence: Hybrid problem-solving in organizations. Academy of Management Review, 50(2), 441–464.
Raisch, S., & Krakowski, S. (2021). Artificial intelligence and management: The automation-augmentation paradox. Academy of Management Review, 46(1), 192–210.
Sun, S., Li, Z. A., Foo, M.-D., Zhou, J., & Lu, J. G. (2025). How and for whom using generative AI affects creativity: A field experiment. Journal of Applied Psychology, 110(12), 1561–1573.
Sweller, J. (2010). Element interactivity and intrinsic, extraneous, and germane cognitive load. Educational Psychology Review, 22(2), 123–138.
Wu, Y. (2026). A three-wave study of human-AI fit and adaptive performance. Technology in Society, 87, Article 103386.
Jonathan H. Westover, PhD is Chief Research Officer (Nexus Institute for Work and AI); Associate Dean and Director of HR Academic Programs (WGU); Professor, Organizational Leadership (UVU); OD/HR/Leadership Consultant (Human Capital Innovations). Read Jonathan Westover's executive profile here.
Suggested Citation: Westover, J. H. (2026). Building Human-AI Fit: Evidence-Based Strategies for Adaptive Performance in AI-Augmented Work. Human Capital Leadership Review, 36(3). doi.org/10.70175/hclreview.2020.36.3.1
