From Tools to Teammates: Designing Human-AI Collaboration in Innovation Work

Listen to a review of this article:

Abstract: Human-AI collaboration (HAIC) has moved from research curiosity to strategic priority for organizations pursuing faster, more inventive, and more reliable innovation outcomes. Many implementations underperform, however, because leaders treat artificial intelligence as a generic productivity tool rather than as a designed teammate whose role, capabilities, and trust requirements must match the task. This article synthesizes recent scholarship and practitioner experience into a structured playbook for executives, design leaders, and human resources partners. It argues that the value of HAIC depends on three deliberate design choices: who initiates the collaboration, how broad the AI's knowledge scope must be, and whether the cognitive mode is analytical or synthetic. Drawing on engineering design, aerospace, industrial product development, hospitality, and mental health contexts, the discussion translates research findings into operating practices, governance structures, and capability investments. The contribution is practical: a clearer way to decide what kind of AI teammate to build, deploy, and trust for any given problem.

The conversation about artificial intelligence at work has shifted in a meaningful way. Five years ago, most boardroom discussions centered on automation and headcount. Today, the more interesting question is collaboration—how humans and AI can produce results together that neither could achieve alone. Deloitte (2020) labeled this evolution a movement from substitution through augmentation to collaboration, the stage at which "superteams" of humans and machines jointly innovate. Peeters et al. (2021) reached a similar conclusion from the academic side, describing hybrid collective intelligence as the most promising framing for AI's contribution to society.

Why now? Three forces converge. First, the underlying technology has matured rapidly. Large language models such as GPT-4 demonstrate striking generalization across domains (Bubeck et al., 2023), and generative systems including latent diffusion models (Rombach et al., 2022) and Point-E (Nichol et al., 2022) now produce images and three-dimensional shapes from text prompts, with direct relevance to industrial and engineering design (Liu & Hu, 2023). Second, the nature of work problems has grown more knowledge-intensive and interdisciplinary, often outstripping what any individual or any narrow AI can handle (Memmert & Bittner, 2022). Third, evidence is accumulating that AI's effect on team performance is not automatic. AI helps in some studies (Song, Gyory, et al., 2022; Filippi, 2023) and hurts in others (Zhang et al., 2021; Chong et al., 2022). The difference is rarely the model itself. It is how the collaboration was designed.

This article translates that research stream into a working playbook. Practitioners often ask the wrong first question—"Which AI should we buy?"—when the better question is, "What role do we need this AI to play, and what does that role demand of capability, interaction, and trust?" Building on the role-classification scheme proposed by Song, Zhu, and Luo (2024), and drawing from foundational frameworks by Dellermann et al. (2019), Dubey et al. (2020), and Seeber et al. (2020), the discussion offers a structured way to make those design choices and then organize people, processes, and governance around them.

The Human-AI Collaboration Landscape

Defining Human-AI Collaboration in Knowledge Work

Human-AI collaboration (HAIC) refers to working arrangements in which humans and AI agents jointly contribute to a task with complementary strengths, exchanging information, decisions, or artifacts. The literature uses several near-synonyms—hybrid intelligence (Dellermann et al., 2019), hybrid human-AI teaming (Caldwell et al., 2022), and machines as teammates (Seeber et al., 2020). Each emphasizes a slightly different angle, but the shared idea is that AI is no longer a back-office utility. It participates.

A few terms are worth defining clearly because they get used loosely in practice:

• Augmentation means AI extends what a worker can do without replacing the worker's judgment (Deloitte, 2020).

• Collaboration implies bidirectional exchange—AI contributes outputs, humans contribute direction and judgment, and both adapt.

• Hybrid intelligence refers to the system-level capacity that emerges when human and machine intelligence are combined deliberately (Dellermann et al., 2019).

These distinctions matter because they imply different design requirements. An augmenting AI needs to be highly directable—it should follow human instruction reliably. A collaborating AI needs additional properties: it must sense context, share awareness, and sometimes initiate (Dubey et al., 2020; Seeber et al., 2020).

State of Practice: From Substitution to Collaboration

Adoption is uneven. In engineering and design contexts, AI assistants have demonstrated benefits across multiple stages of work. Studies have shown improvements in design performance at the individual and team levels (Song, Soria Zurita, et al., 2022; Song, Gyory, et al., 2022), gains in analytic and decision-making confidence (Chong et al., 2023), enhanced creativity (Song et al., 2021), and improved coordination among team members (Gyory et al., 2021). Tools such as Daphne, a virtual assistant for designing Earth-observation spacecraft missions, have demonstrated that domain-specialized AI agents can compress what used to be weeks of trade-space exploration (Viros-i-Martin & Selva, 2019).

The picture is not uniformly bright. Zhang et al. (2021) reported a cautionary case in which AI assistance hindered already high-performing design teams, a result later echoed in studies of confidence dynamics where over-reliance on AI eroded human judgment (Chong et al., 2022). Generative AI in education and creative work shows similar mixed signals: helpful when scaffolded with role clarity and feedback (Tan et al., 2023; Filippi, 2023), problematic when introduced without it.

Three drivers explain the variance:

• Task fit. Some problems benefit from AI's pattern recognition, others from its generative breadth, others from neither.

• Role ambiguity. Workers do not know whether the AI is a tutor, a peer, an oracle, or a tool, and they calibrate trust poorly as a result (Papachristos et al., 2021).

• Trust calibration. Without transparency and interpretability, users either over-trust or under-trust AI outputs (Schelble et al., 2022).

The practical implication is that HAIC outcomes depend less on the choice of model and more on the design of the collaboration—roles, capabilities, interactions, and trust enablers.

Organizational and Individual Consequences of Poorly Designed HAIC

Organizational Performance Impacts

When HAIC is well-designed, organizations report meaningful gains in productivity, quality, and time-to-insight. Song, Soria Zurita, and colleagues (2022) showed that AI assistance helped design teams cope with rising problem complexity in drone design tasks, with measurable improvements in solution quality. Filippi (2023) reported that ChatGPT supported novelty and usefulness in concept generation for product design. In aerospace, Daphne reduced the cognitive load of mission designers by handling repetitive trade studies, freeing engineers to focus on architectural decisions (Viros-i-Martin & Selva, 2019). Reviews of generative models in mechanical design point to similar opportunities in topology optimization and shape synthesis (Regenwetter et al., 2022).

Poorly designed HAIC produces a different set of outcomes:

• Performance regressions in capable teams. Zhang et al. (2021) found that introducing AI advice into already strong design teams pulled their performance down rather than up.

• Erosion of designer learning. When AI provides finished solutions instead of guidance, novices may absorb less domain knowledge over time (Viros-i-Martin & Selva, 2019).

• Confidence miscalibration. Chong et al. (2022, 2023) demonstrated that human confidence in AI evolves over a project's life and can drift away from actual AI accuracy, leading to misuse of advice.

Quantified effects are still emerging in the academic literature, and practitioners should treat eye-catching productivity statistics from vendor reports with caution. The robust pattern is qualitative: HAIC amplifies whatever design choices preceded it. Good design produces compounding gains. Poor design produces compounding drift.

Individual Wellbeing and Stakeholder Impacts

The human side of HAIC matters as much as the system side. Vorobeva et al. (2023) showed that the way an AI service is framed—as augmenting versus replacing human workers—shapes customer acceptance, enjoyment, and ease of use. Workers themselves tend to engage more readily with AI presented as an assistant or partner than with AI presented as a substitute.

Stakeholder impacts vary by context:

• Designers and engineers report higher engagement when AI handles tedious analysis and leaves synthesis to them, but report frustration and disengagement when AI generates outputs they cannot interpret or modify (Ross et al., 2021).

• End users and customers react to AI's framing and visible role. Augmentation framings produce higher acceptance than substitution framings in service contexts (Vorobeva et al., 2023).

• Vulnerable populations require particular care. Shao (2023) cautioned that AI empathy in mental health contexts can backfire when designs fail to generalize across cultural and demographic groups.

The common thread is that HAIC is a sociotechnical system. Outcomes depend on how the technology is positioned, how roles are made legible, and how trust is earned over time.

Evidence-Based Organizational Responses

Table 1: Human-AI Collaboration Roles and Design Frameworks

AI Role Name	Initiation Mode	Intelligence Scope	Cognitive Mode	Core Capabilities	Key Interaction Attributes	Industry Context	Outcome (Inferred)
Ideation Partner	AI-prompter or Human-prompter	General	Synthesis-oriented	Generation, Cross-modal (text-to-image/3D)	Directivity, Adaptability, Awareness sharing	Industrial Product Development, Creative Work	Enhanced creativity and novelty in concept generation; potential disruption if poorly integrated.
Process Facilitator	AI-prompter	General	Synthesis-oriented	Reasoning, Action (Coordination)	Sensing, Predictability, Awareness sharing	Collaborative Team Environments	Improved team coordination and management of complex interdisciplinary problems.
Domain Analyst	Human-prompter	Specialized	Analysis-oriented	Recognition, Prediction, Reasoning	Directability, Situational Awareness, Interpretability	Aerospace, Engineering Design	Improved design performance and efficiency in complex tasks like trade-space exploration.
Technical Analyst (e.g., Surrogate Modeler)	Human-prompter	Specialized	Analysis-oriented	Prediction (Simulation-grounded)	Directability, Speed, Accuracy	Automotive Design, Mechanical Engineering	Dramatic compression of analysis loops and time-to-insight in early-stage design.
Decision Recommender	Human-prompter	Specialized	Analysis-oriented	Prediction, Reasoning	Transparency, Reliability, Surfacing uncertainty	Defense, Aerospace, Engineering Management	Gains in decision-making confidence; risk of over-reliance or trust miscalibration.
Empathic Partner / Assistant	AI-prompter	General or Specialized	Synthesis-oriented	Generation, Artificial Empathy	Sensing, Adaptability, Empathy	Mental Health, Hospitality, Service Contexts	Higher customer acceptance and engagement; risk of cultural backfire if empathy fails to generalize.

What follows are five practices, each grounded in published research, that organizations can implement to design HAIC deliberately. They are not mutually exclusive. Most mature deployments use all five.

Role Clarity Through Structured Classification

Without an explicit role definition for the AI, teams improvise inconsistently. Earlier classifications offered useful starting points: Bruemmer et al. (2002) ranged AI from tool to subordinate to peer to leader; Dubey et al. (2020) identified four roles—task-oriented assistant, teamwork facilitator, human-like associate, and collective moderator; Bittner et al. (2019) distinguished facilitator, peer, and expert; and Siemon (2022) elaborated roles such as coordinating leader, creative idea generator, perfectionist, and practical doer. Song, Zhu, and Luo (2024) recently proposed a unifying scheme that classifies AI roles along three dimensions: initiation spectrum (human-prompter versus AI-prompter), intelligence scope (specialized versus general), and cognitive mode (analysis-oriented versus synthesis-oriented), yielding eight combinable role types.

Effective approaches in practice:

• Name the role explicitly before deployment. "Domain analyst," "ideation partner," "decision recommender," and "process facilitator" each imply different capabilities and interactions.

• Document who initiates. Decide whether the human prompts the AI or whether the AI proactively surfaces signals.

• Match scope to task. A specialized AI for stress analysis is different from a general AI for brainstorming.

• Refresh the role definition as the task evolves. Roles in design rarely stay constant from concept to detailed engineering.

Carnegie Mellon University researchers, working with the Pennsylvania State University Applied Research Lab, ran a multi-year program studying AI-assisted design teams in drone design tasks. Their consistent finding was that performance depended on the fit between the AI's role—primarily a recommender and analyst in their setup—and the design phase the team was navigating (Song, Soria Zurita, et al., 2022; Song, Gyory, et al., 2022). When teams understood whether the AI was advising on component selection or evaluating tradeoffs, they used it well. When the role was ambiguous, even capable teams underused or misused the assistance.

Capability Matching to Task Context

Once a role is named, the next question is what the AI must actually be able to do. Dellermann et al. (2019) classified AI capabilities as recognition, prediction, reasoning, and action; Song, Zhu, and Luo (2024) adapted the last category to generation for design contexts. Different combinations of these capabilities suit different stages of innovation work.

Effective approaches:

• Recognition and prediction dominate the discovery and analysis phases. Natural language processing of customer reviews, patent landscapes, and technical literature illustrates the value here (Siddharth et al., 2022).

• Reasoning matters when the AI must apply learned rules to novel situations. Reasoning is the least mature of the four capabilities, though large language models are advancing it rapidly (Bubeck et al., 2023).

• Generation dominates synthesis phases—concept generation, shape synthesis, and topology optimization (Regenwetter et al., 2022; Zhu & Luo, 2023a; Zhu et al., 2023).

• Cross-modal capability (text-to-image, text-to-3D) becomes relevant when teams ideate at an abstract level and want concrete artifacts in return (Rombach et al., 2022; Nichol et al., 2022).

Toyota Research Institute teams have applied surrogate modeling approaches that pair depth and normal renderings with neural networks to predict car drag coefficients, dramatically compressing the analysis loop in early-stage automotive design (Song, Yuan, et al., 2023). The capability mix here is deliberately narrow—prediction grounded in domain-specific simulation data—because the role is technical analyst, not ideation partner. The lesson generalizes: do not over-buy capability. A specialized analyst does not need general world knowledge, and adding it can introduce new failure modes.

Designing Interactive Attributes for Fluent Teaming

Capability is necessary but not sufficient. AI also has to interact well. Drawing on Dubey et al. (2020) and Seeber et al. (2020), interactive attributes worth designing for include sensing, predictability, directivity, directability, adaptability, and awareness sharing. The right mix depends primarily on who initiates.

Effective approaches:

• When humans initiate, prioritize directability—the AI should be guidable, responsive to refinement, and willing to surface uncertainty.

• When AI initiates, prioritize sensing, predictability, directivity, adaptability, and awareness sharing so it can read context and share what it sees.

• Invest in situation awareness on both sides. Endsley (2023) and Jiang et al. (2023) argued that mutual situation awareness is the cornerstone of effective human-AI teams.

• Engineer adaptability into the AI itself. Wang et al. (2023) demonstrated neuro-inspired adaptability mechanisms for continual learning, which matter for AI agents that need to stay useful as design contexts evolve.

Industrial design groups working with Stable Diffusion across early-, mid-, and late-stage workflows reported that the most useful interaction patterns were those in which designers could iterate quickly with text prompts (high directability) and the system surfaced unexpected stylistic alternatives (mild directivity) (Liu & Hu, 2023). Where the system produced finished-looking outputs without giving designers handles to refine them, designers either accepted them uncritically or rejected them outright—neither a healthy collaboration pattern.

Trust Enablers: Transparency, Interpretability, Empathy, Reliability, Ethicality

Trust is not a vibe; it is a designed property. Schelble et al. (2022) showed empirically that ethical AI behavior, trust repair after errors, and team performance interact tightly. The trust enablers worth engineering include transparency (Vössing et al., 2022), empathy (Srinivasan & González, 2022; Zhu & Luo, 2023b), reliability (Inel et al., 2023), interpretability (Ross et al., 2021), and ethicality (Schelble et al., 2022).

Effective approaches:

• Be transparent about limits. Tell users what the AI was trained on and where it is likely to fail (Vössing et al., 2022).

• Make outputs interpretable. Even partial explanations help users calibrate trust (Ross et al., 2021).

• Track reliability across conditions and surface that history to users (Inel et al., 2023).

• Approach empathy carefully. Shao (2023) warned that AI empathy can backfire if designs fail to generalize across populations; Zhu and Luo (2023b) proposed a framework for artificial empathy specifically tied to human-centered design.

• Codify ethicality. Privacy, fairness, and accountability rules should be explicit in operating procedures, not left to engineers' improvisation.

Mental health chatbot providers such as Woebot Health have learned, sometimes painfully, that empathic AI in clinical-adjacent contexts demands more than warm phrasing. Shao (2023) documented how attempts at artificial empathy can misfire across cultural lines, eroding the very trust they were meant to build. The mature response is to pair AI empathy designs with continuous evaluation and to keep humans in the loop for emotionally significant interactions.

Adaptive Governance and Operating Models

Even well-designed AI agents drift in real settings. Data shifts, user expectations evolve, and edge cases accumulate. Caldwell et al. (2022) proposed an agile research framework for hybrid human-AI teaming organized around trust, transparency, and transferability—principles that translate directly into governance practice.

Effective approaches:

• Establish review cadences tied to model performance and user feedback.

• Maintain a roles-and-rules registry that documents what each AI agent is authorized to do, what data it sees, and who is accountable when it fails.

• Invest in trust repair playbooks. When AI errs, the speed and clarity of the response shape long-term adoption (Schelble et al., 2022).

• Use cross-functional councils combining engineering, design, legal, ethics, and HR to make deployment decisions.

• Monitor confidence dynamics. Chong et al. (2022, 2023) showed that user confidence in AI shifts over time and may not track accuracy; governance should detect that drift.

Consulting organizations advising defense and aerospace clients have adopted governance patterns that explicitly separate the design of the AI artifact from the design of the institutional rules around it—a separation Seeber et al. (2020) recommended in their original research agenda. The result is that role definitions, escalation rules, and audit practices evolve in step with model upgrades rather than being patched in afterward.

Building Long-Term HAIC Capability

Discrete deployments matter, but they decay if the underlying organizational capability is not built. Three pillars sustain HAIC over time.

Multidisciplinary Development Teams

The most consistent recommendation across the HAIC literature is to broaden who designs AI. Engineers alone cannot build empathic, ethical, well-roled AI agents. Zhu and Luo (2023b) argued for incorporating perspectives from cognitive science and design research into artificial-empathy work. Memmert and Bittner (2022) and Caldwell et al. (2022) emphasized the sociotechnical nature of HAIC—the team designing the AI must reflect the team that will use it.

Practical moves:

• Embed psychologists, anthropologists, and designers in AI development teams, not as reviewers but as co-creators.

• Run structured interviews and surveys with frontline users—designers, analysts, customer-facing staff—before specifying AI role and capabilities.

• Treat governance, legal, and ethics partners as design contributors rather than gatekeepers.

• Pilot in cross-disciplinary contexts where the limitations of single-discipline framings will surface quickly.

Continuous Learning Systems for Humans and AI

HAIC implies mutual learning. The AI learns from new data and feedback; humans learn how to use the AI well. Dellermann et al. (2019) made this loop a central feature of their hybrid-intelligence taxonomy.

Practical moves:

• Build feedback channels that capture not only AI errors but also instances where users overrode correct AI advice or accepted incorrect advice (Chong et al., 2022).

• Train users on the AI's role, capability boundaries, and confidence indicators—do not assume they will infer these.

• Invest in adaptable AI. Continual-learning approaches are advancing rapidly (Wang et al., 2023), and their incorporation reduces the risk of model staleness.

• Treat designer learning as a deliverable. Viros-i-Martin and Selva (2019) showed that AI assistance can help or hurt designer learning depending on how it is framed; explicit attention to skill development is required.

Ethical Stewardship and Responsible Deployment

Long-term legitimacy depends on how organizations handle the harder questions: bias, accountability, and the social effects of AI deployment. Schelble et al. (2022) treated ethicality as a measurable team-level property, and Srinivasan and González (2022) connected empathy and accountability tightly. Inel et al. (2023) emphasized data-collection reliability as a foundation for responsible AI.

Practical moves:

• Audit training data and surface known biases to users (Inel et al., 2023).

• Define accountability ahead of failure—who answers for AI errors, and how is the answer communicated?

• Establish red-team and ethics review functions with authority to delay or stop deployment.

• Communicate framing clearly to internal and external stakeholders, since framing shapes acceptance (Vorobeva et al., 2023).

The forward outlook is summarized well by Luo (2023): designing the future of innovation work is not primarily a question of which model to use; it is a question of how human and machine capabilities are structured to compound each other. Organizations that treat HAIC as a design problem will move faster and stumble less than those that treat it as a procurement problem.

Conclusion

The most useful reframing leaders can adopt is this: AI deployments are team designs, not technology purchases. Once that frame is in place, the questions become tractable. What role should the AI play? What capabilities does that role demand—recognition, prediction, reasoning, generation, or some combination? Who initiates the collaboration, and what interactive attributes follow from that choice? What trust enablers are needed to keep the collaboration calibrated over time? What governance, learning systems, and multidisciplinary development practices will keep the answers fresh as conditions change?

The evidence base supports a few sturdy conclusions. AI assistance helps when roles are explicit, capabilities match the task, and trust enablers are engineered intentionally; it hurts or stalls when these are left implicit (Zhang et al., 2021; Chong et al., 2022; Schelble et al., 2022). Generative AI is broadening from late-stage optimization into early-stage ideation and even empathetic understanding (Zhu & Luo, 2023a, 2023b; Zhu et al., 2023; Filippi, 2023), expanding the range of roles AI can play. Mutual situation awareness, adaptability, and continual learning sit at the heart of fluent collaboration (Endsley, 2023; Jiang et al., 2023; Wang et al., 2023).

The actionable takeaways for leaders are concrete. First, name AI roles explicitly before deployment and revisit them as projects evolve. Second, match capability narrowly to the role rather than over-buying generality. Third, design interactive attributes—especially situation awareness and directability—deliberately. Fourth, treat trust enablers (transparency, interpretability, reliability, empathy, ethicality) as engineered properties, not happy accidents. Fifth, build the institutional muscle—multidisciplinary teams, learning loops, ethical stewardship—that lets HAIC compound rather than decay.

The path forward is not a question of replacing human work; it is a question of designing the partnership. Organizations that do this well will produce innovation outcomes their competitors cannot easily match. Those that skip the design step will collect models without realizing the collaboration their leaders thought they were buying.

Research Infographic

The AI Teammate Blueprint Slide Deck

References

1. Bittner, E., Oeste-Reiß, S., & Leimeister, J. M. (2019). Where is the bot in our team? Toward a taxonomy of design option combinations for conversational agents in collaborative work. In Proceedings of the 52nd Hawaii International Conference on System Sciences.

2. Bruemmer, D. J., Marble, J. L., & Dudenhoeffer, D. D. (2002). Mutual initiative in human-machine teams. In Proceedings of the IEEE 7th Conference on Human Factors and Power Plants.

3. Bubeck, S., Chandrasekaran, V., Eldan, R., Gehrke, J., Horvitz, E., Kamar, E., Lee, P., Lee, Y. T., Li, Y., Lundberg, S., Nori, H., Palangi, H., Ribeiro, M. T., & Zhang, Y. (2023). Sparks of artificial general intelligence: Early experiments with GPT-4. arXiv preprint, 2303.12712.

4. Caldwell, S., Sweetser, P., O'Donnell, N., Knight, M. J., Aitchison, M., Gedeon, T., Johnson, D., Brereton, M., Gallagher, M., & Conroy, D. (2022). An agile new research framework for hybrid human-AI teaming: Trust, transparency, and transferability. ACM Transactions on Interactive Intelligent Systems, 12(3), 1–36.

5. Chong, L., Raina, A., Goucher-Lambert, K., Kotovsky, K., & Cagan, J. (2023). The evolution and impact of human confidence in artificial intelligence and in themselves on AI-assisted decision-making in design. Journal of Mechanical Design, 145(3), 031401.

6. Chong, L., Zhang, G., Goucher-Lambert, K., Kotovsky, K., & Cagan, J. (2022). Human confidence in artificial intelligence and in themselves: The evolution and impact of confidence on adoption of AI advice. Computers in Human Behavior, 127, 107018.

7. Dellermann, D., Calma, A., Lipusch, N., Weber, T., Weigel, S., & Ebel, P. (2019). The future of human-AI collaboration: A taxonomy of design knowledge for hybrid intelligence systems. In Proceedings of the 52nd Hawaii International Conference on System Sciences.

8. Deloitte. (2020). 2021 Deloitte global human capital trends: The social enterprise in a world disrupted. Deloitte Insights.

9. Dubey, A., Abhinav, K., Jain, S., Arora, V., & Puttaveerana, A. (2020). HACO: A framework for developing human-AI teaming. In Proceedings of the 13th Innovations in Software Engineering Conference (pp. 1–9).

10. Endsley, M. R. (2023). Supporting human-AI teams: Transparency, explainability, and situation awareness. Computers in Human Behavior, 140, 107574.

11. Filippi, S. (2023). Measuring the impact of ChatGPT on fostering concept generation in innovative product design. Electronics, 12(16), 3535.

12. Gyory, J. T., Song, B., Cagan, J., & McComb, C. (2021). Communication in AI-assisted teams during an interdisciplinary drone design problem. Proceedings of the Design Society, 1, 651–660.

13. Inel, O., Draws, T., & Aroyo, L. (2023). Collect, measure, repeat: Reliability factors for responsible AI data collection. Proceedings of the AAAI Conference on Human Computation and Crowdsourcing, 11(1), 51–64.

14. Jiang, J., Karran, A. J., Coursaris, C. K., Léger, P. M., & Beringer, J. (2023). A situation awareness perspective on human-AI interaction: Tensions and opportunities. International Journal of Human–Computer Interaction, 39(9), 1789–1806.

15. Liu, M., & Hu, Y. (2023). Application potential of stable diffusion in different stages of industrial design. In Artificial Intelligence in HCI (Lecture Notes in Computer Science, Vol. 14050). Springer.

16. Luo, J. (2023). Designing the future of the fourth industrial revolution. Journal of Engineering Design, 34(10), 779–785.

17. Memmert, L., & Bittner, E. (2022). Complex problem solving through human-AI collaboration: Literature review on research contexts. In Proceedings of the 55th Hawaii International Conference on System Sciences.

18. Nichol, A., Jun, H., Dhariwal, P., Mishkin, P., & Chen, M. (2022). Point-E: A system for generating 3D point clouds from complex prompts. arXiv preprint, 2212.08751.

19. Papachristos, E., Skov Johansen, P., Møberg Jacobsen, R., Bjørn Leer Bysted, L., & Skov, M. B. (2021). How do people perceive the role of AI in human-AI collaboration to solve everyday tasks? In Proceedings of the 1st International Conference of the ACM Greek SIGCHI Chapter (pp. 1–6).

20. Peeters, M. M., van Diggelen, J., Van Den Bosch, K., Bronkhorst, A., Neerincx, M. A., Schraagen, J. M., & Raaijmakers, S. (2021). Hybrid collective intelligence in a human–AI society. AI & Society, 36, 217–238.

21. Regenwetter, L., Nobari, A. H., & Ahmed, F. (2022). Deep generative models in engineering design: A review. Journal of Mechanical Design, 144(7), 071704.

22. Rombach, R., Blattmann, A., Lorenz, D., Esser, P., & Ommer, B. (2022). High-resolution image synthesis with latent diffusion models. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (pp. 10684–10695).

23. Ross, A., Chen, N., Hang, E. Z., Glassman, E. L., & Doshi-Velez, F. (2021). Evaluating the interpretability of generative models by interactive reconstruction. In Proceedings of the 2021 CHI Conference on Human Factors in Computing Systems (pp. 1–15).

24. Schelble, B. G., Lopez, J., Textor, C., Zhang, R., McNeese, N. J., Pak, R., & Freeman, G. (2022). Towards ethical AI: Empirically investigating dimensions of AI ethics, trust repair, and performance in human-AI teaming. Human Factors. Advance online publication.

25. Seeber, I., Bittner, E., Briggs, R. O., De Vreede, T., De Vreede, G. J., Elkins, A., Maier, R., Merz, A. B., Oeste-Reiß, S., Randrup, N., & Schwabe, G. (2020). Machines as teammates: A research agenda on AI in team collaboration. Information & Management, 57(2), 103174.

26. Shao, R. (2023). An empathetic AI for mental health intervention: Conceptualizing and examining artificial empathy. In Proceedings of the 2nd Empathy-Centric Design Workshop (pp. 1–6).

27. Siddharth, L., Blessing, L., & Luo, J. (2022). Natural language processing in-and-for design research. Design Science, 8, e21.

28. Siemon, D. (2022). Elaborating team roles for artificial intelligence-based teammates in human-AI collaboration. Group Decision and Negotiation, 31(5), 871–912.

29. Song, B., Gyory, J. T., Zhang, G., Soria Zurita, N. F., Stump, G., Martin, J., Miller, S., Balon, C., Yukish, M., McComb, C., & Cagan, J. (2022). Decoding the agility of artificial intelligence-assisted human design teams. Design Studies, 79, 101094.

30. Song, B., Soria Zurita, N. F., Nolte, H., Singh, H., Cagan, J., & McComb, C. (2021). Addressing challenges to problem complexity: Effectiveness of AI assistance during the design process. In International Design Engineering Technical Conferences and Computers and Information in Engineering Conference.

31. Song, B., Soria Zurita, N. F., Nolte, H., Singh, H., Cagan, J., & McComb, C. (2022). When faced with increasing complexity: The effectiveness of artificial intelligence assistance for drone design. Journal of Mechanical Design, 144(2), 021701.

32. Song, B., Yuan, C., Permenter, F., Arechiga, N., & Ahmed, F. (2023). Surrogate modeling of car drag coefficient with depth and normal renderings. arXiv preprint, 2306.06110.

33. Song, B., Zhu, Q., & Luo, J. (2024). Human-AI collaboration by design. In Proceedings of the International Design Conference – DESIGN 2024 (pp. 2247–2256).

34. Srinivasan, R., & González, B. S. M. (2022). The role of empathy for artificial intelligence accountability. Journal of Responsible Technology, 9, 100021.

35. Tan, S. C., Chen, W., & Chua, B. L. (2023). Leveraging generative artificial intelligence based on large language models for collaborative learning. Learning: Research and Practice, 9(2), 125–134.

36. Viros-i-Martin, A., & Selva, D. (2019). Daphne: A virtual assistant for designing earth observation distributed spacecraft missions. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 13, 30–48.

37. Vorobeva, D., Costa Pinto, D., António, N., & Mattila, A. S. (2023). The augmentation effect of artificial intelligence: Can AI framing shape customer acceptance of AI-based services? Current Issues in Tourism, 1–21.

38. Vössing, M., Kühl, N., Lind, M., & Satzger, G. (2022). Designing transparency for effective human-AI collaboration. Information Systems Frontiers, 24, 877–895.

39. Wang, L., Zhang, X., Li, Q., Zhang, M., Su, H., Zhu, J., & Zhong, Y. (2023). Incorporating neuro-inspired adaptability for continual learning in artificial intelligence. Nature Machine Intelligence, 5, 1356–1368.

40. Zhang, G., Raina, A., Cagan, J., & McComb, C. (2021). A cautionary tale about the impact of AI on human design teams. Design Studies, 72, 100990.

41. Zhu, Q., & Luo, J. (2023a). Generative transformers for design concept generation. Journal of Computing and Information Science in Engineering, 23(4), 041003.

42. Zhu, Q., & Luo, J. (2023b). Toward artificial empathy for human-centered design: A framework. Journal of Mechanical Design, 146(6), 061401.

43. Zhu, Q., Zhang, X., & Luo, J. (2023). Biologically inspired design concept generation using generative pre-trained transformers. Journal of Mechanical Design, 145(4), 041409.

---

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). From Tools to Teammates: A Practitioner's Playbook for Designing Human-AI Collaboration in Innovation Work. Human Capital Leadership Review, 37(1). doi.org/10.70175/hclreview.2020.37.1.4