This dissertation examines how generative AI (genAI) guidance becomes course-level rulemaking during rapid policy emergence. Campus guidance is increasingly available, but is distributed across AI tool pages, cybersecurity rules, academic-integrity statements, teaching-center resources, and syllabus templates, while pedagogical use largely remains subject to faculty judgment. Faculty therefore serve as a practical bridge between institutional intent and the rules governing ordinary coursework. Using an integrated Becher–Weick–Lipsky framework, the study examines how disciplinary cultures, interpretation under ambiguity, and front-line discretion shape classroom AI decisions. The design combines a completed documentary analysis of 300 upper-division undergraduate syllabi from twelve disciplines at five Texas public R1 institutions with a planned faculty survey at one of those institutions. The documentary analysis found visible AI rules in 163 syllabi, with rules appearing more often in soft fields than hard fields but showing little difference between pure and applied fields. Closed-item and open-ended survey responses will examine faculty-reported orientations, perceived guidance, classroom practices, and explanations for concrete AI decisions. Integrating these forms of evidence will clarify how broad AI guidance becomes course policy, why that translation varies, and where institutions can establish clearer student-facing boundaries without eliminating appropriate faculty discretion.

Keywords: generative AI, faculty discretion, sensemaking, disciplinary culture, street-level bureaucracy, syllabus policy, policy emergence

Generative artificial intelligence is already in the classroom. Students are using it, and graduates are entering professional environments where AI fluency is increasingly expected.

Institutions face a genuine governance dilemma. A blanket permission rule would ignore disciplinary differences in knowledge, authorship, assessment, and professional preparation. A blanket prohibition would misrepresent a technology whose implications are still being realized. Broad or principled guidance, and in some cases deliberate silence, may be the most honest institutional response available while the object being governed is still developing. That response transfers part of the governance burden to faculty. Faculty occupy a unique role in ordinary course decisions, including assignment design, permitted tools, disclosure rules, grading judgment, and conversations with students about acceptable work. Individual instructors exercise professional judgment within their disciplinary communities as they address areas where institutional guidance remains broad or unresolved. Analysis of the top fifty United States universities finds that AI policies often develop reactively, without a systematic framework, while instructor discretion remains central to course-level implementation (Alba, McIlwain, & An, 2026).

State policy provides a parallel point of comparison. TRAIGA, enacted in 2025 and effective January 1, 2026, established a statewide AI regulatory framework but excluded institutions of higher education from its definition of governmental entity (Texas Responsible Artificial Intelligence Governance Act, 2025). The exclusion does not prohibit universities from establishing their own policies, but it illustrates a broader governance pattern. Even as the state regulates AI, it leaves ordinary decisions about student use in university classrooms to institutional and faculty-level judgment. The study institution reflects this layered arrangement. Public guidance identifies approved or institutionally supported AI platforms while treating student use of generative AI as a course-level decision and encouraging faculty to communicate whether, when, and how AI may be used.

The stakes of this ambiguity fall most directly on students. At all five institutions in this study’s design, unauthorized AI use is classified as academic dishonesty carrying sanctions that range from a grade penalty through probation, suspension, and expulsion (Texas A&M University, 2025; Texas Tech University, 2024; The University of Texas at Austin, 2025; University of Houston, 2015; University of North Texas, 2025). The institutions vary, however, in how explicitly each has declared that AI misuse falls under the existing conduct framework. The same student act can be permitted practice in one classroom and a disciplinable offense in another.

What that condition produces is visible in the syllabi faculty publish. Statements collected at the study institution in a single semester include a public affairs course permitting AI but not to the point of “minimal thinking”, a two-sentence prohibition within the same college, a nursing rule grounded in privacy and professional standards, and a writing course prohibiting AI because assignments “develop the writer”. Other syllabi say nothing about AI at all, and the silences appear across several disciplinary clusters instead of one field alone. The public record shows no single course-level template resolving classroom AI practice across colleges. Faculty discretion works to address this one course at a time with documents showing different rules, but they reveal little about how faculty convert an evolving policy environment, disciplinary standards, and practical constraints into student-facing decisions. A potential glimpse into that process is the focus of this study.

Approximately four years into widespread access to large language models, faculty responses remain varied, institutional frameworks remain broad or flexible, and the interpretive processes through which faculty arrive at classroom AI decisions remain underexplored. Faculty are not passive recipients of policy. They serve as key translators, exercising considerable discretion over how AI is understood, permitted, and practiced in their courses. Institutions have responded with broad governance frameworks, a rational strategy that preserves flexibility while placing significant interpretive burden on faculty. The literature establishes that faculty responses to generative AI are varied, consequential, and often patterned by discipline with numerous existing studies documenting those differences at scale, though institutional contexts vary (Chirikov, 2026; Buele & Llerena-Aguirre, 2025; Yang et al., 2026). What remains less explained is how institutional ambiguity, disciplinary formation, and faculty discretion combine to produce the classroom rules students encounter.

This mixed methods study examines how broad institutional generative AI guidance becomes student-facing course policy and how faculty explain the judgments involved in that translation. It combines a completed documentary analysis of public institutional guidance and 300 upper-division syllabi from five Texas public R1 institutions with a planned anonymous faculty survey at one of those institutions. This documentary component establishes the patterns of AI governance visible to students across institutional and disciplinary contexts.

The next step is to disseminate a survey to instructors at one of the five institutions. The survey's closed items intend to examine reported orientations, perceived guidance, and discretionary classroom practices, while its open-ended questions elicit faculty explanations of concrete AI-related decisions.

The findings will be integrated to identify areas of convergence, divergence, elaboration, and contextual difference. The study does not assume that the documentary and survey evidence represent the same individual faculty members or directly verify one another. Instead, it examines how patterns visible in public course documents relate to faculty-reported interpretations and practices within overlapping institutional and disciplinary contexts.

The operative research questions are:

The three frameworks follow AI policy from the institution to the student. Becher and Trowler's (2001) academic tribes and territories framework comes first because faculty are not a uniform population. Disciplinary culture is the first filter, shaping what a given field counts as legitimate work and legitimate tool use. Then Weick's (1995) sensemaking theory brings in the individual faculty member, since people encountering something ambiguous construct a working explanation plausible enough to act on. It is this interpretation at the individual level that turns a broad institutional signal into a position a faculty member can act from. Lastly, Lipsky’s (2010) street-level bureaucracy theory explains how faculty discretion converts those interpretations into consequential course-level policy. Faculty do more than execute institutional guidance. Through syllabus rules, assignment boundaries, disclosure expectations, and grading decisions, they determine the operative conditions of a student’s legitimate AI usage.

AI-related matters remain unsettled across higher education. Faculty responses have had time to develop, even as institutional practices and support structures continue to evolve. That response is occurring in a context of widespread student AI use (Rong & Chun, 2024) and substantial faculty concern about its implications. Recent survey data indicate that 83 percent of faculty are concerned about students’ ability to evaluate AI critically, while 82 percent worry that students may become too reliant on it (Digital Education Council, 2025). Yet faculty responses cannot be reduced to resistance or enthusiasm. Familiarity with AI is increasing, but classroom integration has not kept pace. Sixty-one percent of faculty respondents had used AI, but among those who used it in teaching, 88 percent characterized that use as minimal to moderate (Digital Education Council, 2025). Sector surveys similarly identify gaps among institutional strategy, faculty support, and classroom practice (Robert & McCormack, 2025). This uneven engagement may therefore reflect the difficulty of translating evolving institutional expectations into local practice rather than simple compliance failure. Broad institutional AI policies still leave room for interpretation, allowing classroom governance to emerge through professional discretion shaped in part by disciplinary epistemological culture. That pattern predates AI, but the rapid emergence of generative AI makes it unusually visible.

This study works within clear boundaries. The documentary analysis is complete and draws on public syllabi from five Texas public R1 universities within a two-year window covering Fall 2025 and Spring 2026. The faculty survey will be conducted at one of these public institutions and will include faculty with instructional or course-design responsibility, including adjunct instructors. The study does not evaluate pedagogical soundness, student outcomes, actual student AI use, or causal relationships, and it does not reconstruct cognitive sensemaking as it occurs. Documentary claims are drawn from the collected public syllabi only, and survey findings will be bounded to the study institution.

Artificial intelligence refers to large language model and generative systems such as ChatGPT, Claude, Gemini, and Copilot. The term does not include ordinary search engines, grammar checkers, plagiarism detection software, or older forms of algorithmic assistance. The disciplinary quadrant refers to the four-part Becher and Trowler classification, which crosses hard and soft knowledge with pure and applied orientations and is assigned through the survey distribution structure and the disciplinary classification used in the documentary analysis. Sensemaking is the process by which people construct a workable interpretation of an ambiguous situation so they can act (Weick, 1995). In this study, sensemaking will be examined through retrospective reports and related indicators rather than treated as directly observable cognition. It also does not imply movement toward AI adoption, since integration, prohibition, limitation, irrelevance, and continuity may each reflect how faculty interpret their circumstances. Discretionary practice refers to the classroom decisions faculty make when institutional guidance remains broad, incomplete, or still developing. Policy emergence describes the broader process through which those unsettled conditions become increasingly specified through institutional rules and local practice.

Chapter 2 explains why faculty interpretation and discretionary classroom practice are appropriate objects of inquiry. The review follows AI policy through the governance conditions faculty read, the disciplinary culture that shapes how they read it, the individual sensemaking that produces a working position, and the front-line discretion that turns it into classroom rules.

Higher education AI policies have tended to emphasize data privacy, security, academic integrity, and broad principles, while leaving many pedagogical decisions to departments and individual faculty. That pattern appears in syllabus evidence, faculty survey research, and sector-level governance analyses (Chirikov, 2026; Robert & McCormack, 2025; Jiang et al., 2025; Digital Education Council, 2025). Longitudinal evidence shows that movement away from blanket prohibition can coexist with continued assignment-level restriction. At one institution, university-required syllabus language shifted from 57% course-wide prohibition in Fall 2023 to 23% in Fall 2025, while most collected syllabi still prohibited AI for at least some assignments (Sabourin, 2026). Across more than 31,000 syllabi at one large public university, explicit AI regulation reached 55% of courses by Fall 2025, fully restrictive policies declined about five percentage points per year among courses with AI policies, and framing shifted from integrity toward learning effects and attribution (Chirikov, 2026).

This governance condition remains uneven and still developing. Formal policy can sit apart from classroom practice, a pattern institutional theory explains through the decoupling of formal structures from everyday organizational activity (Meyer & Rowan, 1977). Higher education is especially susceptible to such gaps when goals are ambiguous, expertise is distributed, environments are changing, and responsibility is contested (Kezar & Eckel, 2002). Recent research suggests that AI governance is developing under many of these same conditions. One study of professional practice in a heavily regulated setting finds that professionals respond to AI disruption through forms of interactive governance shaped simultaneously by institutional, organizational, and social influences (Chen et al., 2025), while broad expectations for AI adoption without precise prescriptions produce uneven faculty responses (Chan & Colloton, 2024). Together, this work helps explain why institutional direction may coexist with substantial variation in what faculty ultimately do in their classrooms.

Three interpretations of institutional inaction are possible. Silence may represent a deliberate effort to preserve flexibility or postpone conflict, consistent with the idea that non-decision-making can itself function as a form of decision-making (Bachrach & Baratz, 1962; Matland, 1995). It may also reflect traditions of academic freedom that leave substantial pedagogical authority with faculty (Tierney, 2004). A third possibility is that policy simply lags behind technological change, particularly when AI is treated primarily as a technology, risk, or compliance issue while instructional questions remain unresolved. This dissertation does not presume which explanation is correct. Instead, it examines the consequence of uncertain policy for faculty who must make course-level AI decisions without fully specified guidance.

Why might faculty in different fields interpret AI differently? Becher and Trowler (2001), building on Biglan (1973), classify academic fields along two dimensions: whether knowledge is hard or soft and whether the field is pure or applied. Disciplines function as epistemic cultures, carrying shared assumptions about evidence, rigor, authorship, method, and the legitimacy of particular tools. These norms can shape how AI is evaluated. Hard pure fields may place greater emphasis on verifiable method and reproducible reasoning, while soft applied fields may give more weight to professional judgment, ethics, and situated practice. Research predating genAI shows that disciplinary context and institutional commitment can shape faculty technology use (Waggoner, 1994; Shelton, 2014), while recent studies report disciplinary differences in AI orientation, with STEM and applied fields often showing greater openness and non-STEM fields more frequently emphasizing ethical or cultural concerns (Yang et al., 2026). Becher and Trowler therefore provide a structure for examining patterned disciplinary differences without assuming that individual faculty within a field will respond uniformly.

Weick’s sensemaking theory explains how faculty interpret unsettled institutional conditions and determine how to act. Sensemaking involves constructing plausible understandings that allow action to proceed despite ambiguity (Louis, 1980; Weick, 1995; Weick, Sutcliffe, & Obstfeld, 2005). In the context of AI, faculty confront contested benefits, discipline-specific risks, and still-developing standards for appropriate use. Their judgments are shaped by prior experience, disciplinary identity, peer interaction, and available institutional cues. Sensemaking is therefore social and situated rather than purely individual, which helps explain why the same institutional guidance can lead to different classroom decisions even within a single department. It also does not imply movement toward adoption, since integration, restriction, prohibition, and continuity with existing practice may each represent a coherent response to ambiguity.

The focus on interpretation under ambiguity also connects to the subjective tradition in educational management associated with Greenfield (1973) and later Bush (2003), which emphasizes that organizational reality is constructed through the meanings members attach to their circumstances. Its limitation for this study is that it offers less leverage for explaining why interpretations may vary systematically across disciplinary settings or how those interpretations become consequential classroom rules. The framework developed here extends that tradition by locating faculty within disciplinary cultures through Becher and Trowler, explaining interpretation through Weick, and tracing discretionary implementation through Lipsky. Weick’s remaining properties of sensemaking, including its ongoing character, reliance on extracted cues, and emphasis on plausibility over accuracy, remain relevant. Loose coupling in educational organizations further creates the local space in which such interpretation and discretion can operate (Weick, 1976).

Governance conditions, disciplinary norms, and individual interpretations become consequential for students when faculty decide what their courses require. Lipsky’s (2010) street-level bureaucracy framework explains this point of implementation: front-line professionals exercise judgment under conditions of ambiguity, limited resources, and incomplete rules, and those judgments help determine how policy is experienced in practice. For students, AI governance is encountered through syllabi, assignment requirements, grading practices, and instructor explanations, where broad institutional guidance is translated into specific expectations.

Faculty fit this framework with some modification. Academic freedom, disciplinary expertise, and shared governance make professional discretion an expected feature of academic work (Lovell, 2023; Khelifi, 2025). Grading offers a useful comparison. Institutions require grades and establish academic standards, yet they do not prescribe what an A must mean in every course. Faculty exercise professional judgment within those broader boundaries and AI decisions operate through a similar discretionary channel: bounded by institutional expectations, interpreted through professional expertise, and consequential for students.

The three frameworks do different work at cultural, interpretive, and implementation levels. Becher and Trowler locate faculty within disciplinary cultures. Weick explains how ambiguity is interpreted in ways that support action. Lipsky explains why those interpretations become consequential once they are expressed through classroom decisions. Together, they address several gaps in the existing literature. Faculty AI studies describe attitudes and practices but leave open how broad institutional guidance becomes course-level policy. Disciplinary research shows that fields respond differently, while the discretion literature shows that faculty hold substantial classroom authority, yet the relationship between disciplinary context and discretionary action remains underdeveloped. Existing research also tends to treat policy ambiguity as context and faculty response as outcome. This study focuses on what occurs between the two by examining how faculty interpret institutional signals and disciplinary assumptions and translate them into the rules students encounter.

The study design follows from this framework by tracing how disciplinary context, interpretation, and discretion appear in faculty practice and course materials. Disciplinary placement is established through the survey distribution structure and organized within the four Becher and Trowler quadrants. Reported orientations, cues, and explanations provide evidence of sensemaking, while course-level decisions such as AI policies, assignment changes, student discussions, enforcement practices, and reported uncertainty show how discretion is exercised. Public course documents add a second perspective by showing whether AI is prohibited, permitted, conditionally allowed, subject to disclosure, left unaddressed, or governed through template language. Comparing those materials with faculty self-report makes it possible to identify where faculty accounts align with, or differ from, the rules students encounter. The aim is not to judge compliance, but to understand how broad institutional conditions become specific classroom practices.

The study uses a convergent mixed methods design with embedded documentary analysis (Creswell & Creswell, 2023; Creswell & Plano Clark, 2018). The anonymous faculty survey supplies closed-item and open-ended responses from one respondent pool, while the documentary component supplies cross-institutional, student-facing course-policy evidence. The two components are analyzed separately and integrated at the interpretation stage. The documentary analysis draws on publicly available HB 2504 syllabi and is substantially complete, while the faculty survey will proceed following proposal approval and IRB clearance.

The documentary analysis examines 300 upper-division undergraduate syllabi from five Texas public R1 institutions across twelve disciplines, three in each Becher and Trowler quadrant, with five syllabi per institution-by-discipline cell. Eligibility required a complete public syllabus, upper-division status under each institution’s numbering convention, membership in the discipline-prefix crosswalk, and a Fall 2025 or Spring 2026 term. Selection was outcome-blind and deterministic. Within each cell, eligible records were ordered by prefix, catalog number, term, section, and source URL before AI governance outcomes were examined, and the first five eligible records were retained. Each retained row preserves its source URL, original document, extracted text, and document and text hashes, and an independent verification script rechecks retained hashes, matrix balance, counts, and exact tests.

Coding first distinguished explicit AI language from an operative student-facing AI-use rule. A syllabus could mention AI as course content without stating what students may or may not do. Visible rules were classified as Full Prohibition, Partial Restriction, Disclosure Required, Full Permission, or Scaffolded Integration, while AI-Silent indicated that no operative rule was visible. Two independent coders applied the same codebook to a blind 60-syllabus subset, one from each institution-by-discipline cell. Agreement was 59 of 60 for AI-language presence (κ = .963), 50 of 60 for governance position (κ = .773), and 54 of 60 for course-specific rationale (κ = .833). Among the 38 blind-subset rows classified as governance-visible, rationale agreement was 32 of 38 (κ = .582), a weaker and less precise result that is reported conservatively and interpreted cautiously.

The completed analysis found visible student-facing rules in 163 syllabi (54.3%) and no visible rule in 137. Among governance-visible syllabi, 49 used Full Prohibition, 84 Partial Restriction, 20 Disclosure Required, 3 Full Permission, and 7 Scaffolded Integration. The clearest disciplinary difference was between hard and soft fields: visible governance appeared in 56 of 150 hard-field syllabi (37.3%) and 107 of 150 soft-field syllabi (71.3%), with an exact discipline-profile test of p = .006 and a Bonferroni-adjusted p = .013. Pure and applied fields showed little difference (52.7% versus 56.0%, p = .851). The four-quadrant comparison was weaker and is treated as exploratory (p = .045).

The evidentiary boundary remains constant. The documentary record supports claims about visible public course governance, not instructor motive, enforcement, or actual student use. Template and source-system cautions also apply. Repeated syllabus language may reflect institutional templates rather than independent faculty reasoning, and a missing statement may reflect a practice considered settled enough to go unwritten.

The researcher holds an Assistant Provost role at the study institution. That position provides institutional knowledge relevant to the research problem while also creating the possibility that prior experience and organizational familiarity could shape interpretation. Recruitment and data collection will be separated from the researcher’s administrative responsibilities, and the survey will not collect unnecessary identifying information. During analysis, open-ended responses will first be reviewed closely on their own terms before being interpreted through the theoretical framework. Analytic decisions and assumptions that could materially affect interpretation will be documented as needed, and findings will be compared across documentary, closed-item, and open-ended evidence to identify areas of consistency and difference.

The site is a large public research university whose scale, disciplinary range, and decentralized governance make faculty-level interpretation visible. All four disciplinary quadrants are represented within a common institutional governance environment, allowing disciplinary variation to be examined while holding the institutional setting constant. The population includes faculty with instructional or course-design responsibility, including tenured, tenure-track, non-tenure-track, and adjunct instructors. Research-only faculty, postdoctoral fellows, and visiting faculty without instructional responsibility are excluded.

The Qualtrics instrument contains twenty respondent-facing questions: five context and classification items, twelve closed construct items organized into four lean groupings (guidance environment, disciplinary interpretation, classroom discretion, and iterative sensemaking), and three open-ended prompts. The prompts ask respondents to describe one concrete AI-related classroom decision, what shaped it, and what institutional, departmental, disciplinary, or peer guidance would help. Closed indicator items use a five-point agreement scale from strongly disagree to strongly agree, allowing a neutral response where faculty have not formed a stable position. One item focuses specifically on how faculty communicate AI expectations to students, allowing comparison with student-facing patterns identified in the documentary analysis. The instrument does not treat sensemaking as a validated psychometric scale. Instead, the survey examines reported orientations and classroom responses through sensemaking theory, with connections to Weick’s properties examined rather than assumed.

Data collection will proceed through instrument finalization, IRB review, and target mid-semester deployment. Exempt review is the anticipated route because the survey is anonymous, adult, minimal-risk, and designed without direct identifiers. The administrative role will be disclosed in the IRB application, and department chairs will not serve as recruitment intermediaries unless an approved neutral distribution process permits it. The eligible invitation pool will be established from faculty with current instructional or course-design responsibility. Because the study compares disciplinary quadrants, the strength of subgroup claims will depend on achieved cell sizes rather than the aggregate response count. Quantitative analysis is exploratory and will identify patterns in faculty-reported orientations and classroom practices without making causal claims. Closed-item responses will be summarized using appropriate descriptive statistics and compared across the four disciplinary quadrants. Nonparametric tests will be used where appropriate to examine disciplinary differences and associations among ordinal survey items. Missing data will be reported transparently and handled using available responses for each analysis. Discipline will be carried into Qualtrics through the survey distribution structure and mapped to the corresponding Becher and Trowler quadrant.

Open-ended responses will be analyzed thematically, beginning with faculty wording before interpretation through the study’s theoretical framework. The analysis will identify recurring explanations for classroom AI decisions while remaining attentive to responses that do not fit Becher, Weick, or Lipsky. Because the data consist of brief written responses, qualitative findings will be used to interpret reported orientations and decisions rather than to verify behavior or reconstruct cognition.

The documentary, closed-item, and open-ended findings will then be integrated by research question to identify areas of convergence, divergence, and elaboration. Integration will be used to compare what is visible in course documents with how faculty describe their orientations, decisions, and institutional context, while recognizing the different limitations of each source.

The study has several limitations. The faculty survey will be conducted at one institution, so its findings are bounded to that governance context. The documentary analysis is limited to public course materials and cannot establish instructor motive, enforcement, or actual student behavior.

Survey responses also reflect retrospective accounts rather than sensemaking as it occurs, and brief open-ended responses may limit qualitative depth. For that reason, sensemaking is treated as an interpretive lens rather than a directly measured psychological construct.

Finally, the study captures one period of rapid AI policy development. Nonresponse, uneven participation across disciplinary quadrants, and changes in institutional guidance may affect the patterns observed. Findings will therefore be interpreted as bounded evidence of faculty-reported sensemaking and discretionary practice rather than as causal or broadly generalizable conclusions.

Chapters 1–3 establish the need, framework, and design for examining how generative AI governance becomes course-level practice. Institutions, systems, teaching centers, and instructors are already producing AI guidance, yet students encounter that guidance through uneven course-level translation. The Becher–Weick–Lipsky architecture treats this translation as layered governance. The documentary analysis shows what course-level governance looks like when it is visible, while the survey will examine how faculty at the study institution describe making comparable classroom AI judgments. Together, the components can clarify how broad guidance is translated into classroom rules, how faculty explain the judgments behind those decisions, and where institutions might improve transparency and support without flattening disciplinary differences or treating faculty discretion as a problem to eliminate.