Agentic AI and the future of universities: Why teachers must rethink their role

Unlike the generative AI models of the yester years that merely responded to user prompts, Agentic AI systems are emerging as autonomous tutors, redefining India’s higher education landscape. To harness their potential and achieve desired learning outcomes, especially in campus educational settings, the teacher’s role needs a radical shift from the traditional lecturing.

How is agentic AI different from generative AI ?

Generative AI is similar to a vast digital library, where students can access the knowledge by posing specific questions or prompts. In contrast, agentic AI is more like an autonomous tutor, which does not wait for a question to be asked. Instead, when it notices that a student is unable to understand a concept, it proactively explains the same, using a simple example. Through ‘cognitive scaffolding’, it provides step-by-step guidance and gradually, phases out the assistance, as students grasp the concepts.

Evolution of the agentic AI in academic delivery

In 2026, emergent agentic AI in education is a multi-agent orchestration system that co-ordinates and manages multiple agentic processes, across student life cycle, from admissions to evaluation. This entails curriculum design, pedagogical planning, virtual teaching, intelligent tutoring, academic counselling, and adaptive assessment.

Recently, Anthropic launched Claude for Education agentic AI system, which guides the students’ self-learning process through reasoning pathways, rather than directly providing the answers, thereby enabling development of critical thinking skills in students. A teacher can create rubrics aligned to specific learning outcomes, design the evaluation process, with varying difficulty levels, to provide individualised feedback, and facilitate personalised adaptive leaning.

Experience of global educational institutions in using agentic AI

Arizona State University (ASU), U.S. is credited with, one of the most comprehensive and effective implementations of Agentic AI during 2025. In partnership with OpenAI, it built the proprietary CreateAI platform, with agentic capabilities, enabling its students and faculty to build, customise, and share AI-powered applications.

ASU developed and deployed a decentralised network of over 500 specific AI agents, which work based on the goals set by faculty and students. The “Study Buddy” agent analyses the performance of students and prepares them for the upcoming examination by helping with revisions of weak chapters and providing practice questions. Using Mixed Reality (MR) and AI, the system tracks eye movements and logical steps taken by students in a virtual lab and awards a grade, not for the final answer, but for the reasoning.

Northeastern University, U.S., partnered with Anthropic to deploy Claude for Education for over 50,000 users across 13 campuses providing personalised study plans, data visualisation, workflow optimisation, and critical thinking through reasoning walkthroughs.

London School of Economics (U.K.) has integrated the Claude platform to support classroom engagement, research and workforce-ready competencies.

Indian experience

The members of faculty at IIM Nagpur used AI for framing question papers. The aim was to test higher-order-thinking skills of MBA students, rather than rote memory, by setting the difficulty level and the reasoning pathways. The teachers at IIT Delhi have been using AI agents as co-teachers to create personalised learning pathways.

Key benefits of AI deployment in classrooms

IIM Nagpur faculty found that the assessment, which typically took over two weeks in the manual system can now be completed in one to two days, enabling a faster feedback loop for students. Recent studies indicate a 15-20% improvement in student performance through personalised learning paths, which boosted the level of student engagement. By using predictive risk scoring, the AI agents identified disengaged students and slow learners, well in advance, allowing for taking timely remedial action.

The risks of AI deployment in academic delivery

AI models are generally pre-trained by AI product developers, using the existing data. However, when a query is asked, for which factual data is not available in the knowledge base, most of the AI models guess or extrapolate based on the information available, though incomplete, and confidently fabricate an answer, which may not be accurate. Thereby, it misleads the students, who lack skills to question the output critically. This is called the hallucination error. In order to mitigate this situation, teachers must set guardrails for AI agents by restricting system access only to approved libraries and curating reliable content.

AI agents pre-trained on specific datasets may penalise unconventional answers that do not match the pre-set answers, thereby discouraging out-of-the-box creative thinking of the students. At IIM Nagpur, AI-generated scores are reviewed by the professors, ensuring that innovative and divergent thinkers are recognised and rewarded, not penalised.

There is a risk that the current generation students may over-rely on AI, eroding their native thinking and analytical skills. To avoid this, faculty must design the evaluations with safeguards that discourage over-use of AI and promote critical thinking.

Complexities in designing and training educational AI agents

Unlike in the case of business processes like banking and insurance, designing an agentic AI workflow for a university is significantly more complex. While agentic business processes prioritise efficiency in terms of time and human efforts, the focus of academic delivery is to ensure the process of learning, which varies depending on individual students. While knowledge acquisition can be automated easily, judicious application of the acquired knowledge for real life problem solving is a more complex process and calls for active collaboration between human teachers and AI, in the form of Human-in-the-Loop (HITL) framework.

New role of the teachers in the agentic AI world

In the Human-in-the-Loop framework, teachers design AI agents’ personas and rules, tailoring them to classroom ground realities. They curate their personalised pedagogical datasets, like lectures, case studies, evaluation methodologies and research papers to ensure that the AI agents reflect their unique teaching style. Teachers also set guardrails for ethical and responsible AI use.

As noted by experts at Stanford University, while AI can handle personalised instruction, only human teachers can provide emotional support, mentoring, empathy, and encouragement, which are critical for effective learning.

The way forward

Agentic AI is revolutionising the education landscape with its ability to facilitate personalised and experiential learning, with minimal human intervention. However, it cannot take away the role of the teacher in a campus environment. On the other hand, the teacher remains pivotal to ensure effective academic delivery. Their role will evolve to orchestrating multiple educational AI agents, focusing on enabling students to apply their learnings to solve real-world problems rather than merely transmitting knowledge. Institutions must invest in equipping the faculty members with the necessary skills to harness the potential of agentic AI.

To succeed in this new agentic AI world, teachers must unlearn traditional methods of teaching and relearn to leverage the power of AI, as a partner, while preserving the human touch that motivates learners. It is this judicious balance between technology and human being that will redefine teaching in the new generative AI era.

(Prof. O.R.S. Rao is the Chancellor of ICFAI University, Sikkim. Views are personal.)