Educational Technology and Change Journal

By Jim Shimabukuro (assisted by Claude)
Editor

The transformation of university pedagogy that agentic AI demands is perhaps the most visible and immediate of the three domains, and it begins with a fundamental rethinking of what learning is supposed to produce. Commentators inside higher education have described the emerging shift as the move “from generative assistant to autonomous agent,” emphasizing that generative models will increasingly sit behind agentic layers that decide when and how to use them.¹ This means that course designs built around the submission of finished products — essays, problem sets, take-home exams — are structurally vulnerable in ways that syllabi policies cannot patch.

As Instructure’s Dani Loble has argued, institutions need to recognize that the solution to AI misuse is not only in the technology, but in the pedagogy, and that it is more beneficial for educators to design assessments that illuminate the learning process rather than just the final output.⁴ The Association of American Colleges and Universities (AAC&U) has been one of the most organizationally ambitious responders to this challenge: its multi-year Institute on AI, Pedagogy, and the Curriculum has engaged more than 316 teams from 296 institutions, working with faculty, staff, and administrators to rethink pedagogical and assessment approaches across disciplines, adopt AI competencies and literacies as learning outcomes, and consider the ethical and equity implications of AI adoption.⁶

The Ohio State University’s Office of Distance Education has gone further in anchoring this conversation to the specific characteristics of agentic systems. Its published analysis establishes that the integration of planning and memory capabilities of agentic AI allows instructors to tailor pedagogical instruction dynamically based on both course goals and student progress, while also cautioning that without institutional frameworks for AI literacy and governance, instructors risk being sidelined from critical pedagogical decisions.² This is a crucial pivot: rather than treating AI merely as a threat to academic integrity, Ohio State is positioning agentic systems as partners in adaptive instruction, while insisting that teacher agency must be protected as a condition of that partnership.

The Modern Language Association (MLA) has approached the same tension from the humanities side, with Matthew Kirschenbaum of the University of Virginia and Anna Mills of the College of Marin leading a task force whose October 2025 statement warned that faculty members are very close to losing control over the instructional experience as ed tech, including agentic AI, absorbs actual university missions and operations.⁵ Their work has been influential precisely because it draws a line between agentic AI as a support tool and agentic AI as a wholesale replacement for the student cognitive act.

What is emerging across these efforts is a convergence on “AI fluency” as a graduation standard rather than a disciplinary elective. The UPCEA’s analysis of the agentic AI university in 2026 describes an “AI-First Curriculum Redesign” that moves beyond academic integrity to treating AI fluency as a graduation standard, with agents helping faculty redesign assessments to focus on process rather than product.³ Northeastern University has moved concretely in this direction, partnering with Anthropic to deploy Claude AI across all its campuses as part of a broader initiative to close the gap between student AI use and professional AI readiness — a gap made vivid by data showing that while around 84% of college students already use AI tools in coursework, only 18% feel prepared to use AI professionally.⁷

Stanford’s Human-Centered AI (HAI) Institute, meanwhile, has been running its SCALE Initiative, which uses research-driven insights to address key educational challenges and is specifically examining how generative and agentic AI can function as tutoring systems while preserving rather than eroding deep learning habits. Stanford President Jon Levin has framed this work directly, stating that AI has enormous potential to accelerate discovery and innovation and that it will reshape education¹² — a signal that institutional leadership at that level is now treating pedagogical redesign as a presidential priority rather than a faculty committee matter.

If pedagogical transformation is the most visible domain, governance transformation is the most urgently underdeveloped. The challenge is structural: generative AI is already a governance challenge, but agentic AI systems capable of taking autonomous actions across research databases, scheduling systems, administrative workflows, and communication platforms will be substantially harder to govern, and current frameworks built around individual human decision-makers do not map cleanly onto multi-step automated pipelines.⁸

The World Economic Forum’s AI Governance Alliance has recommended that institutions begin developing agentic AI governance policies now, before deployment pressure forces universities into reactive policymaking, and early institutional practice is pointing toward three priorities: designating a named human “AI accountability owner” for each deployed agentic system, implementing dynamic and scoped permissions rather than static access grants, and mandating end-to-end logging requirements so that every agent action, handoff, and decision point is recorded in an immutable audit trail that human reviewers can interrogate after the fact.⁸

The Partnership on AI has identified the core governance gap for agentic systems as the problem of non-reversibility and accountability attribution: agentic systems introduce the potential for non-reversible actions, open-ended decision-making pathways, and privacy vulnerabilities from expanded data access, and to enable responsible adoption in 2026, prioritizing evaluation frameworks that scale and accountability infrastructure for attribution and remediation is necessary.⁹ Singapore’s Infocomm Media Development Authority (IMDA) has published what is currently the most detailed model governance framework for agentic AI, emphasizing that agents should be bounded by design — granted only the minimum permissions required for specific tasks — and that humans must remain meaningfully accountable at every layer of a multi-agent architecture.

UC Berkeley has begun developing institutional AI standards that draw on similar principles, recognizing that campus deployments of agentic advising, tutoring, and administrative systems will require a new kind of governance instrument: not a policy document but an operating protocol that travels with each agent deployment and can be audited continuously. The 2026 State of AI Agents report offers an important empirical point for motivating this work: companies that actively practice AI governance put twelve times more AI projects into production than those that do not¹¹ — a finding that reframes governance not as a brake on innovation but as its precondition.

In the United States, compliance pressure is also accelerating governance timelines. The EU AI Act, which came into force in August 2024, classifies several AI applications common in universities — including AI-assisted admissions systems and student performance analytics — as high-risk, and European universities must now demonstrate compliance or face penalties.⁸ American institutions face an analogous pressure through FERPA, which imposes strict constraints on how student data can be processed by third-party AI systems, and through the Department of Education’s guidance explicitly calling for ethical frameworks for AI use in student services and assessment.

The Chronicle of Higher Education’s coverage of “Einstein,” the agentic tool that could complete whole courses for students, has served as an accelerant for governance conversations: the episode forced institutions to confront the possibility that agentic systems could quietly handle readings, quizzes, and discussion posts end-to-end unless pedagogy and assessment are redesigned.¹ Faculty governance bodies, accreditation agencies, and state higher education boards are now beginning — belatedly, in many cases — to ask whether existing academic integrity policies, faculty handbook provisions, and vendor contract standards are adequate for an environment in which software agents can act autonomously inside institutional systems.

The third domain — data infrastructure — may ultimately determine how much of the pedagogical and governance ambition universities can actually realize. Agentic AI depends on unified, high-quality, interoperable data far more critically than generative AI does, because agents must be able to retrieve and act on student records, learning analytics, advising histories, and institutional research data in real time, often across systems that were never designed to talk to one another. Instructure’s Loble has been direct about this dependency: to harness the opportunities of agentic AI, universities need an underlying AI-friendly architecture that includes flexible and transparent interfaces and unified access to data, and 2026 must be the year of structural reconfiguration in which educators redesign teaching and assessment and unite their fragmented technological systems.⁴

This is a significant infrastructure investment, and it is one that most institutions have not yet made. Thomas Davenport and Randy Bean, writing in MIT Sloan Management Review, have noted that companies that don’t have internal AI infrastructure force their data scientists and AI-focused businesspeople to each replicate the hard work of figuring out what tools to use, what data is available, and what methods and algorithms to employ, making it both more expensive and more time-consuming to build AI at scale.¹⁴ The same dynamic applies with even greater force to universities, whose data environments are typically more fragmented than those of large enterprises.

Stanford’s investments offer a partial model for what the data infrastructure transformation looks like in practice: throughout the 2024–2025 academic year, Stanford expanded its infrastructure, boosted computing capacity, and convened new symposiums, committees, and conversations — including bringing online Marlowe, a GPU-based supercomputer — to shape the future of AI across the institution.¹² Its new Computing and Data Science (CoDa) building is explicitly designed to change the way faculty, students, and researchers think by changing the physical and technical space in which computing and data science work occurs. These are not merely hardware investments; they represent an architectural commitment to treating data infrastructure as the substrate on which pedagogical and governance innovation rests.

The UPCEA’s vision of the agentic AI university points in the same direction, describing how institutions are moving from scattered pilots to governed, agentic workflows, with the work of these agents expected to be personalized, proactive, and persistent³ across the full student lifecycle — a goal that cannot be met without data architectures capable of sustaining continuous, adaptive, cross-platform agent activity.

The data infrastructure challenge is inseparable from the equity challenge. Institutions that cannot afford to build or acquire unified AI-ready data systems will be unable to deploy the kinds of agentic tutoring, advising, and mental health triage tools that are already showing results at well-resourced universities. Aviva Legatt, writing in Forbes, has articulated the stakes with unusual directness: the shift from AI as a tool to AI as institutional infrastructure has become unmistakable, and in 2026, institutions that operationalize AI will widen their performance gap, while those that don’t will inherit a shadow system they can’t control.³

The shadow system she describes — informal, ungoverned, inequitable student use of commercial AI agents — is already present on most campuses. The choice universities now face is not between an AI future and a pre-AI past, but between a future in which agentic systems are shaped by deliberate pedagogical, governance, and infrastructure choices, and one in which those systems are shaped entirely by the market logic of ed-tech vendors. The institutions and leaders doing this work most seriously — Ohio State on pedagogy, the MLA on faculty agency, the AAC&U on curricular redesign, Stanford on data infrastructure, Singapore’s IMDA on governance frameworks, and the Partnership on AI on accountability standards — are not yet writing the final draft of what an agentic university looks like. But they are writing the first ones, and the norms they establish in this liminal moment will be far harder to revise once agentic AI has moved from the margins of campus life to its operational center.

References

1. “Status of Agentic AI in Higher Ed: A Liminal Moment” — Educational Technology and Change Journal (2026). https://etcjournal.com/2026/03/06/status-of-agentic-ai-in-higher-ed-a-liminal-moment/
2. “Agentic AI in Higher Education” — Ohio State University, ASC Office of Distance Education (2025). https://ascode.osu.edu/news/agentic-ai-higher-education
3. “The Rise of the Agentic AI University in 2026” — UPCEA / Inside Higher Ed (2026). https://upcea.edu/the-rise-of-the-agentic-ai-university-in-2026/
4. “Higher Education Needs Structural Changes to Flourish in the AI Era” — Times Higher Education Campus (2026). https://www.timeshighereducation.com/campus/higher-education-needs-structural-changes-flourish-ai-era
5. “Educational Technologies and AI Agents” — News from the MLA (2026). https://news.mla.hcommons.org/2026/01/30/educational-technologies-and-ai-agents/
6. “2026–27 Institute on AI, Pedagogy, and the Curriculum” — AAC&U (2026). https://www.aacu.org/event/2026-27-institute-ai-pedagogy-curriculum
7. “Agentic AI in Education: Use Cases, 2026 Trends, Playbook” — 8allocate (2026). https://8allocate.com/blog/agentic-ai-in-education-use-cases-trends-and-implementation-playbook/
8. “AI Governance in Higher Education: The 2026 Framework for Policy & Risk” — The Education Magazine (2026). https://www.theeducationmagazine.com/ai-governance-in-higher-education/
9. “Six AI Governance Priorities for 2026” — Partnership on AI (2026). https://partnershiponai.org/resource/six-ai-governance-priorities/
10. “Model AI Governance Framework for Agentic AI” — Singapore IMDA (2025). https://www.imda.gov.sg/-/media/imda/files/about/emerging-tech-and-research/artificial-intelligence/mgf-for-agentic-ai.pdf
11. “Databricks Named a Leader in the IDC MarketScape: Worldwide Unified AI Governance Platforms 2025–2026” — Databricks Blog (2026). https://www.databricks.com/blog/databricks-named-leader-idc-marketscape-worldwide-unified-ai-governance-platforms-2025-2026
12. “How Stanford Is Advancing Responsible AI” — Stanford Report (2025). https://news.stanford.edu/stories/2025/06/stanford-collaborative-responsible-ai-initiatives
13. “Stanford AI Experts Predict What Will Happen in 2026” — Stanford HAI (2025). https://hai.stanford.edu/news/stanford-ai-experts-predict-what-will-happen-in-2026
14. “Five Trends in AI and Data Science for 2026” — MIT Sloan Management Review (2025). https://sloanreview.mit.edu/article/five-trends-in-ai-and-data-science-for-2026/

[End]

Filed under: Uncategorized |