TOOLS TO SUPPORT PROGRAMMING EDUCATION WITH THE HELP OF MULTI-ROLE AI AGENTS

Abstract

The article theoretically substantiates and experimentally validates a multi-role system of AI agents—Tutor, Evaluator, and Coach—for supporting programming instruction in higher education institutions. It is shown that the ad hoc use of generative models by students without pedagogical design creates risks of substituting independent work, violating academic integrity, and making assessment non-transparent. The aim of the study is to design and test an architecture of a multi-role agent system with controlled pedagogical support density, which ensures the individualization of learning while preserving the leading role of the instructor. The methodological basis comprises an analysis of scholarly literature and regulatory documents on the use of AI in education, the design of an agent architecture structured as “routing – tools – evaluation,” as well as a pedagogical experiment at Lviv Polytechnic National University. Within the experiment, telemetry data, results of rubric-based code assessment (Correctness, C# idioms, Clarity, Robustness, Efficiency), and survey responses on students’ perceptions of the agents’ work were collected. The findings demonstrate high acceptability of the system: most students evaluated interactions with the agents as useful, the Tutor’s explanations as clear, the Evaluator’s grading as fair, and the Coach’s recommendations as helpful for planning their learning. An increase in the proportion of successful solutions, a reduction in task completion time, and a decrease in repeated errors were recorded. The scientific novelty of the work lies in combining a multi-role AI architecture with rubric-based formative assessment and coaching support integrated into a real programming course.