Capacity and Disposition: A Framework for Understanding GenAI Engagement in Assessment Practice

Synthesis: Miller (2026) argues that Assessment professionals' uneven engagement with Generative AI is explained less by enthusiasm or resistance than by the interaction of two conditions: the protected time they have to learn and test the tool, and the professional stance they bring to it. He defines capacity as time on the calendar and disposition as the beliefs, motivations, and values through which a person reads GenAI as threat, shortcut, tool, partner, or catalyst, with institutional support as the environment that expands or constrains both. The Capacity-Disposition Framework names four engagement conditions — The Calendar Trap, The Margin Lab, The Evidence Gate, and The Practice Commons — each with its own institutional risk and its own support move. The article is conceptual and non-empirical; it Self-Report Measures of students and faculty and from a Spring 2026 pulse survey of 278 assessment professionals, and it ends with a research agenda rather than results.

Key Findings

1. The problem is conditions, not people. Uneven GenAI engagement reflects workload and professional stance working together, so leaders who read low engagement as resistance risk overlooking time scarcity, unclear policy, and unrecognized labor.
2. Capacity is defined as time. Protected time to explore, test, validate, document, and implement is the most immediate condition for responsible engagement, and responsible adoption initially demands more time, not less.
3. Disposition is a stance, not an identity. A preservationist stance begins from continuity and concern about harm; a generative stance begins from experimentation and possibility; the productive stance is critical openness.
4. Four conditions describe work, not personality. The Calendar Trap is low time plus preservationist caution; The Margin Lab is generative curiosity with no protected structure; The Evidence Gate is time plus caution; The Practice Commons is time plus stewardship.
5. Time is assessment infrastructure. Institutions invest in assessment systems and dashboards, but GenAI engagement needs a parallel investment in release time, pilot periods, learning communities, and workload recognition.
6. Institutional support has five layers. Policy clarity, access to approved tools, repeated professional development rather than one-off training, workload recognition, and risk calibration that matches oversight to consequence.
7. Skepticism can become a method. Cautious professionals should be asked to lead validation checks, risk reviews, item-review protocols, and inter-rater comparisons, so their caution becomes an asset instead of delay.

Why generic adoption models do not fit assessment work

The article starts from survey evidence that GenAI has entered the environment where evidence of learning is produced. In a survey of UK full-time undergraduates, the share reporting use of GenAI to help prepare assessed work rose from 53% in 2024 to 88% in 2025, while 92% reported using AI tools in some form (Freeman, 2025). Higher Education of 1,057 faculty respondents reported deep concern about overreliance, Critical Thinking, Academic Integrity, and institutional preparation (Watson & Rainie, 2026). The Spring 2026 GenAI Pulse Survey used a convenience sample of 278 assessment leaders and team members: among respondents to the relevant items, 79% were regular or occasional users, 87% of users reported efficiency gains, and 87% described themselves as self-taught, while 71% reported Privacy, 43% reported lacking policy, 37% reported lack of training, and 46% reported substantial influence within their immediate team against 11% at the institutional level (Slotnick et al., 2026).

Miller then argues that existing adoption theories are incomplete here. The technology acceptance model (Davis, 1989) and the unified theory of acceptance and use of technology (Venkatesh et al., 2003) explain whether people see value and whether the environment permits use, and diffusion of innovations theory (Rogers, 2003) explains how an innovation spreads. What they do not explain is the professional burden of deciding whether a GenAI-assisted output is valid, fair, transparent, and appropriate — a judgment that sits inside validity theory and the evidence claims an assessment office is accountable for.

The four conditions

The two axes are deliberately simple: time capacity from low to high, and disposition from preservationist to generative. The labels describe work conditions and a professional may occupy different conditions across a year, a project, or a task. The Calendar Trap pairs low capacity with preservationist caution: the calendar is full, the risk feels high, and the safest move appears to be staying with familiar processes, so legitimate concern hardens into avoidance. The support move is to begin with assessment values — validity, fairness, transparency — and small tests of non-sensitive tasks, not with persuasion. The Margin Lab pairs a generative disposition with low capacity: experimentation happens in the margins, informally, under-documented, and often uncompensated, which is why these professionals become informal translators for colleagues until the work depends on personal energy alone. The move is institutionalization: approved tools, written procedures, documentation standards, modest release time.

The Evidence Gate pairs high capacity with preservationist caution. These professionals may have tools, training, and support yet remain cautious, asking what would count as adequate evidence that GenAI-assisted coding is trustworthy, or what human review is required. The condition becomes unproductive only when every uncertainty becomes a reason to postpone any test, and the move is disciplined experimentation: give skepticism a constructive task. The Practice Commons pairs high capacity with a generative disposition, producing stewards who test workflows, document decisions, share resources, invite critique, and protect student data. Its risk is collapsing back into one person who becomes responsible for everything, policy, training, faculty consultation, tool testing, and ethical review.

Applying the framework, and what it asks of leaders

The framework is worked through ordinary assessment tasks — outcomes and curriculum mapping, rubric development, qualitative analysis, annual reporting and accreditation, faculty development, and student GenAI use — with each task answered differently depending on the condition. Its central claim for practice is that evidence quality, not report fluency, is the test: a rubric that reads well may not represent the intended construct, a summary of student comments may flatten important differences, and a GenAI-generated findings list may exaggerate patterns or miss discrepant cases, so fluent text still requires review before it can support an educational claim.

For leaders, the first implication is to assess the conditions before judging the people, auditing workload so that GenAI does not become an excuse to do more low-value assessment faster. Leaders should create protected experimentation structures — GenAI assessment labs, prompt review sessions, summer pilot cohorts, internal mini-grants, cross-functional governance teams — each producing something reusable such as a tested workflow, a disclosure template, a risk matrix, or an item-review protocol. They should pair dispositions intentionally, putting someone who sees possibilities beside someone who sees risks, and they should normalize disclosure and risk tiering, matching oversight to consequence. The article closes with a five-part research agenda: survey work, longitudinal studies of movement between conditions, design-based evaluation of interventions, validation studies of specific workflows, and scholarship on the labor politics of who is asked to learn GenAI.

What this means for practice

• Teaching Audit the conditions before the people: ask whether an assessment professional or a colleague has protected time, approved tools, and policy clarity before treating low GenAI engagement as resistance.
• Instructors. Reserve real time for validation rather than more production — a scheduled review of one generated outcome or rubric criterion beats an unfunded redesign of the reporting cycle.
• Assessment designers. Write rubrics and outcomes so they can be examined against evidence: compare generated descriptors with actual student work and intended performance levels, and keep records of GenAI's contribution and the human review applied to it.
• Faculty developers. Package GenAI work as repeated, job-embedded practice with a community of practice, because a single workshop can introduce vocabulary but cannot build judgment.
• Administrators Create risk tiers that match oversight to consequence, and recognize the labor of testing, documenting, and translating GenAI policy, or the institution benefits from innovation while denying its cost.

Limitations

• The article is conceptual and non-empirical: there is no sample, no setting, no comparison, and no outcome data, and the framework's predicted differences in evidence quality are explicitly left as an open empirical question.
• The supporting figures come from surveys with weak designs: the Elon/AAC&U faculty sample was nonprobability-based, and the pulse survey of assessment professionals used a convenience sample of 278, so both describe their respondents rather than the wider population.
• The author names moderators the framework omits — policy clarity, tool access, AI Literacy, data governance, role authority, leadership support, risk tolerance, prior technical skill, and local assessment culture — and treats them as supports and constraints rather than dimensions.
• The proposed interventions (assessment labs, pilot cohorts, fellowships, cross-functional governance teams) are suggestions, not evaluated treatments; the article itself frames them as examples rather than established effects.

Citation

Miller, W. (2026). Capacity and Disposition: A Framework for Understanding GenAI Engagement in Assessment Practice. Intersection: A Journal at the Intersection of Assessment and Learning, 7(5), 115–134.