A practical guide to agentic AI in higher education

How can universities use agentic AI?

The clearest opportunities are often not futuristic ones, but processes that already consume large amounts of staff time.

Automating complex administrative workflows

University administration is full of processes spread across several systems and teams. Gee gave the example of organising campus visits, where staff had previously spent considerable time coordinating room availability, accommodation and other requirements. An agent can instead gather information across different sources and bring the options together.

Rice University has also explored agentic approaches to processes such as tenure, where information needs to be collected from several internal and external databases.

Not every task needs an AI agent, however. Fixed and predictable processes may already be better suited to conventional automation. As Gee put it during the webinar: “Don't bring an agent to an RPA fight.” The starting point should be the workflow and the problem that needs solving, rather than the technology itself.

Supporting students more personally

AI agents could also help students navigate information that currently sits across multiple university systems. An agent might bring together course information, careers resources, learning platforms and university services to help a student understand their options and what to do next. For example, a student asking about possible careers from their degree could receive information on relevant pathways, associated skills, useful modules and support available through the university. That does not mean replacing advisers. In many cases, the value is in making routine information easier to access so staff can focus more of their time on students who need individual support. From the QS International Student Survey, we know advisers are a key source of information, but that they also increasingly expect quick responses to enquiries.

Creating more responsive learning experiences

Agentic AI could also change how students interact with course materials. AI-supported learning environments can already generate summaries, explanations and quizzes. An agent could go further by responding to a student's progress, identifying new course materials, suggesting relevant revision activities or highlighting areas where more support may be useful.

The role of faculty remains important here. Universities will need to decide where AI genuinely supports learning and where students need to undertake the intellectual work themselves.Pedagogy, assessment and critical thinking therefore need to be considered as part of the design of an agent, rather than after it has been introduced.

Giving researchers more time

Some of the most useful applications may have relatively little to do with conducting research itself. Researchers spend significant time on funding searches, administration and information gathering. Gee described the objective at Rice University simply as trying to “find that hour”: giving researchers back even a small amount of time each week for their core work.

One application being explored at Rice uses an AI agent to match researchers' expertise with relevant grant opportunities. Similar approaches could support grant monitoring, institutional knowledge searches and routine research administration. Removing small amounts of friction across many researchers can create meaningful additional capacity for research.

How should universities implement agentic AI?

As AI systems become capable of taking actions across university systems, governance becomes more important.

Universities need to be clear about which systems an agent can access, what data it can use, which actions it is allowed to take and where human approval is required.

A practical approach is to begin with a small number of clearly defined use cases. Rather than accumulating large numbers of disconnected pilots, institutions can focus on workflows where there is an identifiable problem and a measurable benefit, whether that is reducing processing time, improving response times or freeing staff capacity.

Human oversight also needs to reflect the consequences of the task. Processes involving admissions, assessment, academic progression or other high-impact decisions require considerably more scrutiny than routine administrative work.

During the QS webinar, speakers also warned about the risk of “agent sprawl”, where departments create large numbers of independent agents without sufficient central oversight.

This points to the need for a common institutional foundation. Identity management, permissions, data architecture, security and interoperability all become more important as agents begin working across university systems.

From AI pilots to the responsive campus

The most useful measure of agentic AI in higher education may not be how many agents an institution deploys, but whether they make the university more responsive. That could mean helping a student reach the right support more quickly, shortening an administrative process or giving a researcher more time to focus on their work.

With 67% of academics and 62% of students already using Generative AI at least weekly, the question for many institutions is increasingly shifting from whether AI will be used to how it should be integrated responsibly and effectively. Getting there will require more than access to increasingly powerful AI models. Universities will need to identify the right problems, redesign existing workflows and put the appropriate governance around how agents operate. That also requires broader institutional AI capability. Leadership, infrastructure, governance, staff skills and culture will all influence whether AI agents remain isolated experiments or become useful parts of how the university works.