Conceptualising Human–AI Collaboration Competency for hybrid intelligence: towards learner agency

Synthesis: Graduate competency frameworks acknowledge that students must learn to work with AI, but they rarely specify what collaboration itself requires, and they tend to leave learner Learner Agency implicit. Dang, Hong, Doyle, and Nguyen propose the Human-AI Collaboration Competency (HACC) framework, which models human-AI collaboration as a dynamic configuration of three components — domain competency, AI competency, and metacognitive competency — and the intersections between them, including the ones it labels Working with AI and Agentic Learning with AI. Where TPACK assigns the integrative work to teachers' pedagogy, HACC assigns it to the learner's own metacognitive regulation, which the authors treat as the mechanism through which Learner Agency is exercised. To probe the framework, they coded interviews with 24 higher education educators in Aotearoa New Zealand into seven HACC dimensions and modeled the connections among them with Epistemic Network Analysis. Educators recognized every dimension but weighted domain and AI talk most heavily, and role level shifted how strongly they tied AI to metacognition and professional development.

Key Findings

• HACC defines human-AI collaboration competency as a dynamic configuration of domain, AI, and metacognitive competency rather than a list of AI skills bolted onto an existing curriculum.
• Learner agency is deliberately excluded as a standalone component; it is the emergent quality learners express when they coordinate the three competencies, supported by metacognition.
• Its seven dimensions are the three components plus four intersections — Working with AI, Agentic Learning with AI, Agentic Personal and Professional Development, and HACC itself.
• Twenty-four educators (9 early-career, 8 senior, 7 principal) across New Zealand disciplines recognized all seven dimensions, with domain (12–19%) and AI (16–22%) talk taking the largest share of 1,650 coded units.
• Metacognition was more prominent among senior educators, who connected it to domain, AI, and learning with AI, matching the claim that students' persistent difficulties are regulatory rather than technical.
• Principal educators' networks differed significantly from senior educators (U = 10.00, p = .04) and early-career educators (U = 7.00, p = .01) on the first ENA axis, centering on AI with links to HACC and professional development.

The HACC Framework

The framework describes three components. Domain competency is disciplinary knowledge, conceptual understanding, and reasoning strategies — the epistemic anchor for judging whether AI output is accurate and appropriate. AI competency covers operational fluency with AI tools plus an understanding of how systems generate output, their data dependencies, limits, and ethical concerns such as bias and attribution. Metacognitive competency is the capacity to plan, monitor, and evaluate one's own cognition, extended here to managing AI's involvement: deciding when to consult AI, how to frame interactions, how to judge responses, and when to revise or override them. The authors then name the intersections: Working with AI (domain plus AI), Agentic Learning with AI (AI plus metacognition), Agentic Personal and Professional Development (domain plus metacognition), and HACC (all three). Their central move is to argue that the integrative role pedagogy plays in TPACK passes to the learner's own metacognition. Learner agency is not a separate node but the quality students express when they enact these competencies.

What Educators Recognized

Interview transcripts were segmented into 6,822 utterances and reduced to 1,650 substantively relevant units; two coders independently coded 20% of the data with acceptable agreement (κ = 0.73, individual codes 0.65–0.82). To answer the first research question, the authors compared the share of coded units each dimension drew across role levels. Every HACC dimension appeared in educators' talk. Domain-related talk (12–19% of coded units) and AI-related talk (16–22%) accounted for the largest shares, framing discipline knowledge and skills as the backdrop for reasoning about human-AI collaboration. Metacognitive and integrative HACC-level talk appeared more situationally and rose among senior educators. The authors read the pattern as a shared set of competency considerations across role levels, with systematic differences in what educators foregrounded.

Role Level and the Metacognitive Turn

The second research question asked how educators connect the dimensions, which Epistemic Network Analysis addressed using a four-utterance moving window over the coded units. Early-career educators' networks clustered around Domain, Working with AI, and AI, with weaker links to professional development and metacognition — an emphasis on disciplinary preparation and AI use for work tasks. Senior educators concentrated edge weight on metacognition, connected to domain, AI, and learning with AI, though the early-career versus senior comparison was not statistically significant (p = .07 on both axes) and is treated as descriptive. Principal educators differed significantly from both other groups on the first axis and showed the strongest co-occurrences around AI with links to HACC, professional development, and metacognition — a shift toward holistic graduate preparation. The authors read this as a consequence of expanding curricular responsibility.

What this means for practice

• Instructors. Keep disciplinary judgment in the loop: design tasks in which students evaluate AI output against disciplinary standards, since educators treated domain competency as the anchor without which AI use becomes shallow.
• Instructors. Make regulatory work visible and consequential through process portfolios, reflective AI-use logs, annotated decision trails, and learning analytics that trace revision, rather than grading the output alone.
• Curriculum designers. Resist isolating AI in a standalone module; map where each course contributes domain, AI, or metacognitive competency and where the intersections are actually reached, because principal educators saw siloed programs as the problem.
• Administrators. Recognize that students' AI choices are made inside assessment rules and integrity policies; clarify levels of acceptable use, because framing AI mainly as misconduct skews learners' judgment.

Limitations

• HACC does not directly measure learner agency; it names competencies that may support it, so the framework's central claim is conceptual and future work must show how they translate into observable agentic processes.
• The empirical component is a preliminary, educator-based evaluation: it shows how 24 staff recognized and connected HACC dimensions, not evidence of students' demonstrated competencies or developmental trajectories.
• Participants came from a single country and the ENA rested on a modest number of participant-level units, so the principal-educator differences and the smaller senior-educator shifts indicate emphasis rather than definitive practice; discipline-level comparison was out of scope.

Citation

Dang, B., Hong, Y., Doyle, C., & Nguyen, A. (2026). Conceptualising Human–AI Collaboration Competency for hybrid intelligence: towards learner agency through a metacognitive perspective. Studies in Higher Education.