- EDUCAUSE '26: Top 10 Reflects 'Age of Perpetual Change' — The 2027 EDUCAUSE Top 10 IT Issues, unveiled Thursday at the annual conference in Denver, puts artificial intelligence in the No. 1 slot for the first time in the list's history. The list is organized around a theme of "perpetual change": higher education leaders are prioritizing change management amid shifting enrollment, funding pressure, and rapid technology cycles. EDUCAUSE frames the coming year as one where AI experimentation gives way to thoughtful evaluation of, and investment in, uses that help fulfill institutional missions.
📌 Key takeaways:
- With AI at the top of the sector's priority list for the first time, every college and university should expect AI governance, funding, and staffing decisions to be board-level conversations in the next budget cycle — technology leaders who arrive with an evaluation framework will set the terms.
- The "perpetual change" framing is a signal that change management capacity — not any single tool — is becoming the core IT competency the sector is being judged on.
- The shift from experimentation to value evaluation matches what most institutional AI programs are already feeling: the pilot era is ending and portfolio decisions are due.
- EDUCAUSE '26: Reflecting on AI in Higher Ed, 4 Years Into ChatGPT — Reporting from the conference floor in Denver, GovTech captures EDUCAUSE President John O'Brien's assessment that the initial hype around generative AI is dying down and higher education is doing a better job of putting AI into broader context. Four years in, the sector's question has matured from "should we use this" to where AI genuinely adds value — in teaching, research, and administrative operations. The piece is a useful temperature check on sector sentiment as strategic plans come up for revision.
📌 Key takeaways:
- The hype-to-value transition is now the sector's official posture; campus leaders should reframe AI roadmaps around demonstrated use cases rather than capability promises.
- Institutional context-setting — asking how AI serves mission rather than chasing features — is emerging as the differentiator between mature programs and reactive ones.
- Controlling agents a joint priority for data, AI governance teams — TechTarget's Craig Stedman examines why data governance and AI governance teams now share responsibility for keeping AI agents inside their lanes: agents that complete tasks accurately but reach for unauthorized data or take unapproved actions are a new class of governance failure. The piece walks through setting limits on what agents can do, starting with a clear business case and a well-defined scope for each deployment. It is one of the clearest recent treatments of the organizational — not technical — side of agent governance.
📌 Key takeaways:
- Institutions deploying agents against systems of record should treat data access scoping as a joint data-governance/AI-governance decision made before deployment, not an audit finding after it.
- A defined business case and explicit per-agent scope is emerging as the baseline control pattern for agentic AI in enterprises — campus data stewardship groups should expect to be pulled into AI project reviews.
- 🏛️ UCSD angle: The campus's data governance and AI governance structures already intersect at the LiteLLM gateway, where policy and audit for agentic workflows like the Enterprise Data Agent are enforced centrally.
- Cohere's North 2 puts AI agents on a budget and gives them a memory — Cohere's North 2 overhaul of its enterprise agent platform takes aim at two problems every large agent deployment hits: runaway token spend and amnesia between sessions. The release adds user quotas, rate limits, organization-wide spending caps, and a redesigned agent harness, plus memory that keeps an agent's context across sessions. Giving administrators financial controls over agentic work — the same way they control cloud spend — is a meaningful step toward agents being managed as an operating cost rather than an experiment.
📌 Key takeaways:
- AI consumption controls — per-unit quotas, org-wide caps, chargeback-ready spend visibility — are becoming standard platform features; institutions without a cost model for agentic AI should build one before the invoices force the issue.
- Persistent agent memory across sessions moves agents from one-shot task execution toward durable operational roles, which raises both the value and the governance stakes.
- 🏛️ UCSD angle: Usage-based cost accountability is already the campus model — the AI program runs on a recharge structure with gateway-level spend tracking, which anticipates exactly this vendor direction.
- The great enterprise AI rewrite is starting — Constellation Research argues that enterprises are beginning to rewrite themselves to become AI-native — restructuring workflows, org charts, and system architectures around agents rather than bolting AI onto existing processes. The analysis points to early signs including factories running experiments at scale and companies saving on token expenditure by controlling their own models and data. For institutions, the provocative question is whether higher education's process inventory — registration, advising, procurement, IT service — gets redesigned around agents or merely augmented.
📌 Key takeaways:
- The competitive frontier is moving from deploying AI tools to redesigning processes as AI-native; institutions that only augment existing workflows may find the efficiency ceiling lower than expected.
- Owning the model layer — controlling models and data rather than renting capability piecemeal — is surfacing as a cost lever, which strengthens the case for institution-run gateways and private deployments.
- 🏛️ UCSD angle: The campus's citizen developer program and AI-native app building on the TritonAI platform are early instances of exactly this rewrite pattern — rebuilding service delivery workflows around AI rather than adding it at the edges.
- A model guide for the GPT-6 family — OpenAI published a practical guide to its GPT-6 model family: how to choose among the models, tune reasoning effort, improve prompts and skills, coordinate tools, and prepare agent workflows for production. For institutions that now juggle multiple frontier model families behind institutional gateways, vendor-published selection guides like this are becoming required operational reading — model choice is turning into a tunable engineering decision with real cost and quality tradeoffs, not a one-time procurement.
📌 Key takeaways:
- Reasoning-effort tuning means the same model family can serve cheap high-volume use cases and expensive deep-reasoning ones — platform teams should expose these knobs to application owners rather than picking one setting for everyone.
- Preparing agent workflows for production (tool coordination, skills, prompt structure) is now vendor-documented practice, which lowers the barrier for institutions building agentic services.
- 🏛️ UCSD angle: The campus gateway already routes model families per workload, so guidance like this feeds directly into the model-agnostic evaluation process the AI program runs when new releases land.