Toward Self-Driving Universities: Can Universities Drive Themselves with Agentic AI?

TL;DR

The chapter addresses the administrative and quality-assurance bottlenecks in higher education, including accreditation overhead and data fragmentation. It proposes a framework that maps agentic AI capabilities to an autonomy-level taxonomy, with Phase 1 data infrastructure as prerequisite. Through a MUST University pilot, it demonstrates AI-assisted data ingestion, ExamGPT-based assessment and CLO/PLO analytics, and automated reporting. It discusses governance, ethics, change management, and a phased roadmap toward autonomous operations, highlighting practical impact and remaining challenges.

Abstract

The rapid evolution of Agentic AI and large language models (LLMs) presents transformative opportunities for higher education institutions. This chapter introduces the concept of self-driving universities, a vision in which AI-enabled systems progressively automate administrative, academic, and quality-assurance processes through a staged autonomy model inspired by self-driving systems. We examine the current challenges facing traditional universities, including bureaucratic overload, fragmented information systems, and the disproportionate amount of time faculty spend on clerical tasks, which diverts effort away from timely feedback, curricular improvement, student mentorship, and research productivity. While prior AI-in-education research has focused primarily on learning support, tutoring, and analytics, there remains a lack of system-level frameworks for automating institutional quality assurance and accreditation workflows using agentic AI. We address this gap by presenting a framework for progressive automation, detailing how agentic AI can transform course design, assessment alignment, accreditation documentation, and institutional reporting. Through case studies of pilot deployments, we demonstrate that AI-assisted workflows can substantially reduce task completion times while enabling capabilities previously considered infeasible. The chapter's originality lies in introducing an autonomy-level framework for higher education operations grounded in agentic AI architectures rather than prompt-based LLM assistance. Finally, we discuss the critical infrastructure requirements, ethical considerations, and a strategic roadmap for universities to transition toward higher levels of academic autonomy.

Figures (10)

• Figure 1: Self-attention mechanism in transformer architecture. Each input token attends to all other tokens, enabling contextual understanding of relationships within academic text.
• Figure 2: Retrieval-Augmented Generation architecture for university systems. User queries trigger document retrieval from the institutional knowledge base, grounding LLM responses in verified policies and specifications.
• Figure 3: Generative AI versus Agentic AI. While generative systems follow simple prompt-response patterns, agentic systems execute iterative perception-planning-action-memory loops with external tool integration.
• Figure 4: University autonomy progression from Level 0 (fully manual) to Level 5 (full autonomy). Each level represents increasing AI capability and decreasing human intervention requirements.
• Figure 5: Vision-language model processing of handwritten examination. The interface shows the original scanned document alongside OCR-corrected text, CLO mapping, chain-of-thought reasoning, and AI-generated feedback.