We have become accustomed to generative artificial intelligence in the past couple of years. That will not go away, but increasingly, it will serve in support of agents.

“Where generative AI creates, agentic AI acts.” That’s how my trusted assistant, Gemini 2.5 Pro deep research, describes the difference. By the way, I commonly use Gemini 2.5 Pro as one of my research tools, as I have in this column, however, it is I who writes the column.

Agents, unlike generative tools, create and perform multistep goals with minimal human supervision. The essential difference is found in its proactive nature. Rather than waiting for a specific, step-by-step command, agentic systems take a high-level objective and independently create and execute a plan to achieve that goal. This triggers a continuous, iterative workflow that is much like a cognitive loop. The typical agentic process involves six key steps, as described by Nvidia:

An early version of a general agent was released last week by OpenAI to their paid subscribers of ChatGPT. The message accompanying the release explains the potential for power and productivity as well as the care one must take to ensure privacy:

I tried the new agent for an update on an ongoing research project I have been conducting this year. It was faster than the ChatGPT-o3 deep research product I have used previously. The report was more concise but included all the data I expected for my weekly update. It also condensed and formatted relevant material in tables. I was careful with the way in which I handled sharing personal information with the agent. Over time, I am confident that more secure ways will be found to protect users and their privacy.

Inherently, the agentic AI is different from the generative AI. Generative AI is like a brilliant but rather passive research assistant that requires constant, explicit direction. You must provide a series of precise, individual prompts to get it to complete your real objective. Agentic AI, on the other hand, functions more like an experienced project leader. You provide it with a high-level, strategic objective such as “Prepare a report for the provost that outlines the potential of offering a number of relevant new online AI certificate programs this fall targeted to large regional corporations.”

The agent then autonomously deconstructs this goal into a multistep workflow. It will search for relevant topics and targets, identify potential programs, compare and contrast current and potential offerings with those at competing institutions, generate a ROI over time analysis, synthesize the findings, draft the briefing document, access the provost’s calendar, identify available meeting times, and send a calendar invitation with the briefing attached.

That’s just one example. Agentic AI will be useful in many aspects of the university operation. It will promote efficiency, accuracy and save significant money through its round-the-clock productivity. Here are some key areas where agentic AI may be useful in the year ahead.

• Student recruitment, admissions and support: We are already seeing agentic AI transforming recruitment from a high-volume, nonpersonalized process into a deeply individualized and proactive process. Engaging prospective students 24-7 across multiple communication channels, agents tailor their outreach with the promise of personalized learning that has been a central goal of educational technology. Agentic AI is poised to make this vision a reality at scale.
• Teaching and learning: At last, agentic AI can personalize the learning process. These systems function as autonomous, 24-7 AI tutors that adapt to each student’s unique learning pace and style. The agentic tutor can assess a student’s understanding of a concept, identify any knowledge gaps and adapt the materials for each learner to create a personalized learning path. By employing techniques such as Socratic questioning, an agent can guide a student through a problem-solving process, adapting to the learner’s understanding of the topic and prompting them to think critically, rather than simply providing the correct answer. This can lead to mastery learning, where all learners master the key concepts of a class before they are awarded credit. No learner is left behind.
• Administrative support: Agentic AI can create enhanced, annotated grade books and continuously updated, enhanced course plans for faculty; predictive analytic reports for deans and directors; individualized retention and advancement recommendations; marketing and public relations materials and plans; library recommendations for acquisitions and student engagement; and many more functions across the spectrum of administration.

AI agents will offer the next level of artificial intelligence to higher education. We can anticipate embodied agents becoming available in a year or so. Meanwhile, I encourage us all to experiment with agentic AI as it becomes available. In doing so, we can begin to create our own personalized, proactive, professional assistant that can anticipate our needs and implement our preferences.

Who at your university is leading the move to agentic AI? Perhaps you may be in a position to model the efficiency and professionalism of AI agents.