Beyond AI Tutors: A Human-Centred Framework for Goal-Driven Learning Agents in Education

Abstract

Artificial intelligence is changing education by offering new ways to explain concepts, support teachers, and personalise student learning. However, many conventional educational chatbots focus on answering individual questions rather than coordinating the complete learning process. Students often need more than an answer: they need a learning plan, suitable practice activities, feedback on mistakes, and guidance about what to study next. Agentic artificial intelligence offers a possible way to coordinate these activities around a defined learning objective. This paper reviews selected research published between 2024 and 2026 on pedagogical AI agents, large language model agents, and agentic systems in education. The reviewed literature highlights opportunities for personalized tutoring, adaptive feedback, assignment support, and teacher assistance. It also identifies concerns involving inaccurate outputs, student dependency, privacy, fairness, accountability, and limited real-world evaluation. To address these challenges, this paper proposes a human-centred framework that connects learning-goal identification, task planning, trusted educational resources, instructional support, progress evaluation, and adaptive follow-up. The framework centres student independence, teacher judgement, and responsible data use. It is a conceptual proposal and has not yet been experimentally evaluated. Future research should assess its educational effectiveness, reliability, accessibility, and suitability for different learning environments