Introduction: Agentic AI as the Next Frontier of AI Development

Artificial intelligence is advancing quickly, and many IR/IE practitioners and data analytics professionals may feel as though the ground is shifting beneath their feet. Just 3-4 years ago, tools like ChatGPT introduced many of us to the many possibilities offered by the large language models and generative AI. Today, the conversation has already shifted to something new: agentic AI. According to Amazon Web Services, agentic AI is an autonomous AI system that can act independently to achieve a pre-determined goal. Traditional artificial intelligence requires prompting and step-by-step guidance. However, agentic AI is proactive and can perform complex tasks without constant human oversight. The term “Agentic" means agency, the ability of these systems to act independently in a goal-driven manner.

As the capabilities and features of Generative AI advance at such a rapid pace, many institutions and data analytics professionals are struggling to keep up with new terminology and technologies that could be added to their work environment. There is a lot of hype about AI in the press and popular culture, but there are often few concrete examples of how these new technologies apply to the workplace or large institutions. Even fewer concrete examples show how these new technologies fit into the higher education data ecosystem. 

In this article, we outline the spectrum of AI technologies and explain the range of AI tools and use cases from chatbots to agentic systems. There is a broad spectrum of how AI can be applied to IR/IE. However, less frequently discussed is what is required to implement these new technologies that will provide reliable and trustworthy answers and outputs. We explore how agentic AI might be used in the context of higher education, and discuss the implications of agentic AI development on institutional researchers and data analytics professionals in higher education. 

Mapping the AI Spectrum: Bots, Assistants, and Agents

There is a range of implementations for AI applications, from those that require direct inputs and oversight from the human using the software to those that run autonomously and automate business processes and workflows. The terms used for AI across this spectrum can be confusing and inconsistent across industries and applications. The graph below, adapted from the Digital Education Council’s Agentic AI typology (DEC, 2025), offers a succinct yet effective description of the progression from Chatbot implementation to a full agentic AI system.

Chatbot

At this point, most data analytics professionals will be familiar with and have used an AI chatbot. These are tools that use a conversational interface to iteratively respond to questions or prompts, and generate an output based on that conversation. While they can be useful for many tasks, such as text summarization, generating draft emails, or creating icons and stock images for presentation slides, they typically do not allow for the reproduction and automation of tasks. The user has to manually submit the prompts to that chatbot and interact with the outputs.

AI Assistant

AI Assistants extend chatbot functionality by embedding a chat prompt into other tools used for work, or by incorporating specific requirements, outputs, or processes into the chat interface. Examples include AI-drafted email responses, presentation software that can reformat slides and create stock images, or assistants that can autocomplete Python or SQL syntax while writing in a code editor. Because AI assistants are embedded within other tools, they can often be more helpful for productivity as they take into context information from the document or process that the user is actively working on. However, they still require manual inputs and prompts from the user and are not fully automated workflows.

AI Agent

AI agents extend the capabilities of chatbots and assistants by automating the process of achieving a result. Unlike tools that require manual iteration or specific workflows, agents are goal-oriented. This shift from human-directed to algorithm-directed execution is a fundamental change. The user defines the objective, and the agent determines the necessary steps to reach it and executes those steps. While these systems can follow specific instructions or steps to achieve the provided goal, they do not require them. To ensure security and accuracy, management teams must pre-configure system prompts and robust guardrails that remain invisible to the end user. When properly implemented, these guardrails define the agent’s scope and knowledge base and permissions, making it far more powerful than a standard assistant. Without such oversight, AI agents risk unintended consequences such as altering database code, changing variable definitions, or exposing sensitive information. Without strong oversight, system governance, and data security, AI Agents may be vulnerable to malicious users seeking unauthorized data access or accidentally revealing sensitive or personal information.

AI Agentic Systems

To prevent AI agents from exceeding their scope or permissions, complex workflows are often broken down into discrete tasks managed by multiple specialized agents. In these architectures, an agentic system serves as an orchestration layer, coordinating individual agents to ensure greater control and oversight. This modular approach allows for clearer business processes, as each task is governed by specific guardrails tailored to that agent's unique role. The entire system works together to direct user inputs to the appropriate agent that is designed to handle a specific task.

While agentic systems have dominated AI discussions over the last six months, many of these concepts are still in the early stages of implementation. Their full realization depends on a growing institutional understanding of how to structure data to support autonomous processes. In the meantime, software providers and vendors are increasingly embedding AI solutions directly into their platforms. As these tools become staples of enterprise-wide applications, it is critical for IR/IE professionals to understand their capabilities, limitations, and inherent risks.

Agentic AI in Action: Some Use Cases in Higher Education

Admission and Recruitment - the “Digital Concierge”: AI agents can significantly streamline the admissions process, particularly in the labor-intensive review of high school transcripts. Because transcript formats vary widely across institutions, admissions and registrar staff often spend hours manually evaluating documents to determine eligibility and transfer credits. AI agents can automate these workflows by extracting data from raw text into structured databases and identifying missing documentation. When a file is incomplete, the agent can automatically contact applicants or school counselors to request the necessary information. Furthermore, for students who have stopped responding, the agent can send targeted "nudges" to re-engage them. Once all requirements are met, the agent provides an initial assessment and routes the file to an admissions staff member for final human review.

Teaching and Learning—the Socratic Tutor: Learning Management Systems (LMS) contain a wealth of complex, high-value data that often remains underutilized due to inconsistencies across courses and the technical challenges of data cleaning. Consequently, analytics professionals frequently rely on static markers, such as midterm or final grades, rather than the dynamic, real-time data available within the LMS. AI agents can bridge this gap by harnessing this data to provide tailored, student-specific interventions. Beyond simple quizzing, an agent can monitor a student’s progress and send automated reminders for missing assignments or upcoming assessments. 

Student Success and Outcomes: If IT systems are in place to record student interactions on campus throughout the day, AI agents can monitor these engagements over time and alert advisors when participation drops. By analyzing class attendance, library and meal plan usage, assignment completion, and recreational activities, an agent can identify patterns correlated with attrition or academic risk. The agent could then recommend a course of action, notify an advisor, and automatically schedule a meeting on the advisor’s calendar while informing the student of available success resources. This could even trigger a secondary agent to deploy a multi-channel outreach campaign promoting university programming to specific student populations. While such a comprehensive system requires years of training data and robust IT infrastructure, it can be built in stages as institutions refine the processes that drive student success.

Implications for IR/IE and other Data Analytics Professionals

When an agentic AI is applied to handle data analytics-specific tasks, it is often referred to as agentic analytics. Agentic analytics is an approach to data analytics where AI-powered agents autonomously sense, analyze, decide, and act on a user's behalf. Building on the operational perspective of agentic AI discussed earlier, it represents a fundamental shift from treating analytics as a passive reporting layer to an active decision engine (Kulseth, 2025; Undru, 2025). The example mentioned above about student success and outcomes relies heavily on agentic analytics.

Traditional data analytics tools rely on static reports or retrospective dashboards that describe what happened in the past, leaving it up to humans to interpret the data, validate hypotheses, and figure out what to do next. In contrast, agentic analytics bridges the gap between insight and action by operating proactively and independently. Agentic analytics can democratize data by allowing non-technical business users to ask questions in natural language. At CMU, we have successfully developed a pilot application to allow users to ask data questions using a natural language dialogue. A dean can now simply ask, "how many international students have we had in the last 10 years?" and receive an immediate, plain-language explanation backed by data, rather than having to wait for a data analyst to retrieve the data.

AI workflows and agents will change the responsibilities of data analytics professionals over the coming years. However, AI agents will not be able to perform well without strong data quality and data preparation. Data analytics professionals, particularly in Institutional Effectiveness or Institutional Research offices, are often the bridge between the business users and campus leaders who need data to make decisions, and the technical colleagues in IT that manage the institution's data architecture and enterprise software systems. IR/IT staff will play a critical role in translating the needs of running the university into technical specifications for how to define metrics that matter to campus leaders. 

Conclusion: Be Prepared for the Agentic Campus

Leadership priorities on many campuses are going to force IR/IE to adapt AI tools and agentic systems sooner rather than later, even if IR/IE staff members are still learning how to implement and use these new tools. Teams that are not yet prepared to incorporate AI into their workflows and processes may be at a disadvantage and may struggle to meet the demands of increasing reliance on data for planning and decision-making.

IR/IE offices need to begin implementing strong data quality and data governance programs to prepare for the use of AI and agentic AI systems. Having strong data preparation and processing workflows that consistently work and prepare raw data for analytical questions needs to be put in place now. IR/IE professionals need to be the navigators between the metadata and the semantic layers of the data resources to provide the right business context for implementing AI solutions. Clearly defined metrics that match the business needs and logic used by campus leadership need to be defined, established, and agreed upon before attempting to implement any agentic AI workflows. Ethical considerations and discussions about data access and permissions need to be considered when tasks can be delegated to AI, when a human-in-the-loop fits into the overall architecture to maintain oversight, and what tasks or workflows are too critical or sensitive to be performed by AI at this time. 

References:

Digital Education Council. (2025, October). The agentic AI playbook: Use cases for higher education. https://www.digitaleducationcouncil.com/executive-briefings-event/the-agentic-ai-playbook-use-cases-for-higher-education-dec-executive-briefing-022

Gao, Y. (2025, October 14). Agentic AI in higher education. ASC Office of Distance Education, The Ohio State University.

Kulseth, D. (2025, December 4). Agent analytics: From dashboards to proactive insights.

Undru, A. (2025, July 17). Agentic analytics explained: AI that acts on your data. ThoughtSpot. https://www.thoughtspot.com/data-trends/analytics/agentic-analytics

 

---

Dr. Henry Zheng is Carnegie Mellon University’s Vice Provost for Institutional Effectiveness and Planning. He provides strategic leadership for the university’s data reporting, enterprise data strategy, strategic analytics, and organizational assessment functions.

 

Matthew Hoolsema is Director of Data Science and Advanced Analytics in Carnegie Mellon University's Institutional Effectiveness and Planning division. He oversees the development of strategic analytics and dashboards, data strategy implementation, predictive modeling initiatives, and development of data models for use with AI applications and AI agents. mhoolsema@cmu.edu.