This course introduces the foundations of Agentic Artificial Intelligence, namely AI systems that do not merely generate responses, but can observe a context, pursue goals, make decisions, and act autonomously within computational environments. At a time when intelligent agents are reshaping how we interact with software, information, and decision processes, the course provides students with the tools needed to understand what truly makes this behavior possible. Starting from classical agent models, search and planning techniques, and sequential decision-making frameworks, the course builds a rigorous theoretical and algorithmic basis for analyzing how an artificial system can select effective actions with respect to a goal. On this basis, the course introduces Markov Decision Processes and the principles of reinforcement learning, showing how an agent can learn from experience and progressively improve its behavior. The final part of the course connects these foundations to modern agentic AI systems based on large language models, treated not as technologies to be used superficially, but as systems to be examined critically in terms of memory, tool use, planning, evaluation, and operational limits. The goal is to provide a solid, contemporary, and transferable understanding of the principles underlying intelligent agents, both today and in the future evolution of the field. Within the Bachelor’s Degree in Management and Artificial Intelligence, the course is intended not only to provide technical foundations, but also to help students develop the ability to evaluate agentic systems in relation to real decision processes, organizational contexts, and operational constraints. In this sense, the course contributes both to further study in artificial intelligence and to a more informed understanding of how autonomous and semi-autonomous systems can be designed, assessed, and critically discussed in contemporary socio-technical settings.

Students are expected to have basic programming knowledge and a general familiarity with core computer science concepts, in particular algorithms, data structures, and procedural abstraction. In practical terms, students should be able to read simple pseudocode or short programs, follow the logic of an algorithmic procedure, and understand basic notions such as iteration, conditional branching, and structured data manipulation. Introductory mathematical preparation is also expected, especially on functions, basic linear algebra, elementary probability, and basic calculus. Students are not expected to have advanced mathematical training, but they should be comfortable with simple symbolic reasoning, basic vector and matrix notions, elementary probabilistic reasoning, and the interpretation of quantitative relationships relevant to formal models and algorithms. No prior in-depth knowledge of machine learning, deep learning, or large language models is required: these topics will be introduced within the course to the extent needed for understanding modern agentic AI systems. A general familiarity with logical reasoning and quantitative analysis is beneficial, but not a strict prerequisite. These prerequisites should be understood as background normally compatible with the study path of the degree program by the third year. The course does not assume prior specialization in artificial intelligence, but it does require a willingness to engage with formal models, algorithmic reasoning, and conceptually structured technical material.

1. Introduction to Agentic Artificial Intelligence, agent models, and essential elements of machine learning (4.5h) - notion of agent, environment, perception, action, and goals - goal-directed behavior and rationality - main agent models - course overview and connection with modern agentic systems - introductory machine learning notions relevant to the course - difference between supervised learning, reinforcement learning, and other paradigms 2. Problem solving, search, and planning (13.5h) - problem formulation as state spaces - uninformed search strategies - informed search strategies and heuristics - properties of search algorithms: completeness, optimality, and computational cost - introduction to planning and the role of goal decomposition - comparison between classical planning and forms of task decomposition in modern agentic systems - application examples in operational and decision-oriented contexts 3. Uncertainty, sequential decision making, and Markov Decision Processes (7.5h) - decision making under uncertainty - notions of state, transition, reward, and policy - the Markov property - Markov Decision Processes - value functions and the Bellman principle - algorithmic interpretation of optimal choice over time 4. Foundations of reinforcement learning (7.5h) - learning through interaction with the environment - exploration and exploitation - value-based methods and principles of temporal-difference learning - Q-learning as a foundational model for policy learning - distinction between model-based and model-free approaches - limitations, assumptions, and main application contexts 5. Foundations of modern agentic AI systems (12h) - general architecture of an agent loop - large language models as black-box components in agentic systems - tool use, working memory, and context management - context, memory, action generation, and operational limits of the models - planning and task decomposition in LLM-based systems - evaluation criteria for agentic systems - comparison between classical agent models and modern large-language-model-based agents - examples and case studies in informational, organizational, and decision-oriented scenarios - guided discussions and collaborative activities on small problems or application scenarios - final synthesis of the course Total structured contents: 45h. The remaining 3h of the course will be devoted to consolidation activities, case-study discussion, final synthesis, and possible realignment with the effective academic calendar. The course is primarily theoretical and methodological in orientation, but selected topics may be supported by notebooks, computational demonstrations, or small guided exercises aimed at clarifying the operational meaning of the models and methods discussed. These activities are intended to strengthen conceptual understanding rather than to turn the course into a software laboratory. Project work may involve, depending on the nature of the topic and the indications provided by the instructor, one or more of the following forms: conceptual analysis, formalization of a problem, comparative evaluation of methods, simplified implementation of a limited component, discussion of an agentic workflow, or critical assessment of the capabilities and limitations of a contemporary agentic system. The expected level of technical development will remain consistent with the scope of a 6-CFU bachelor-level course.

Main reference books: - Stuart Russell, Peter Norvig, Artificial Intelligence: A Modern Approach, 4th ed., Pearson, 2021. - David L. Poole, Alan K. Mackworth, Artificial Intelligence: Foundations of Computational Agents, 3rd ed., Cambridge University Press, 2023. Also available online: https://artint.info/3e/html/ArtInt3e.html - Richard S. Sutton, Andrew G. Barto, Reinforcement Learning: An Introduction, 2nd ed., MIT Press, 2018. Also available online: https://web.stanford.edu/class/psych209/Readings/SuttonBartoIPRLBook2ndEd.pdf - Simon J.D. Prince, Understanding Deep Learning, MIT Press, 2023. Also available online: http://udlbook.com Supplementary and supporting material: - slides, notes, and teaching material provided by the instructor during the course; - notebooks, computational examples, and additional readings distributed through the course GitHub repository; - curated teaching material for the final part devoted to modern agentic AI systems, made available by the instructor through the course repository and related course channels.

The course will be delivered through plenary lectures. Teaching will follow a foundations-first approach, in which concepts are introduced from motivating problems and reference scenarios, and then progressively formalized through rigorous models, algorithms, and representations. Lectures will combine theoretical explanation, analysis of examples, guided discussion, and short applied activities aimed at clarifying the functioning of the methods under study. Some topics will be accompanied by notebooks, demonstrations, or small computational examples, used not as an end in themselves, but as support for understanding the core concepts. Particular attention will be devoted to connecting classical agent models, sequential decision making, and modern agentic AI systems based on large language models. In the final part of the course, collaborative analysis activities, case-study discussions, and comparative synthesis sessions may be included in order to consolidate the relationship between theoretical foundations and contemporary manifestations of agentic systems. Depending on the effective organization of the course, specific sessions may also be devoted to orienting students in the development of the project work, clarifying expectations, discussing possible formats, and providing guidance on the connection between project topics and the course contents.

Assessment will consist of two components: - an individual written examination, worth 50% of the final grade; - a small project, worth 50% of the final grade, to be developed individually or in groups according to the modalities specified by the instructor at the beginning of the course. The written examination will assess understanding of the main concepts, models, and methods covered in the course, as well as the ability to formalize problems, interpret algorithmic schemes, reason on simple decision-making settings, and discuss the behavior and limitations of agentic systems rigorously. The written examination may include conceptual questions, formal reasoning exercises, and the interpretation of simplified algorithmic or modeling scenarios consistent with the course contents. The project will aim to explore, on a limited scale, one or more course topics through an activity involving analysis, design, simplified implementation, comparative study, or structured evaluation of an agentic system or related method. Depending on the type of project proposed, students may be asked to frame a problem, motivate modeling or design choices, analyze the strengths and limitations of an approach, and connect their work to the theoretical contents of the course. Where appropriate, projects may also include discussion of application, organizational, or decision-oriented implications consistent with the degree program. The project will be accompanied by a short presentation and discussion, intended to assess understanding of the choices made, the ability to connect the work to the course contents, clarity of exposition, and the capacity to discuss the work critically and consciously. In the case of group projects, the assessment will take into account each student’s individual contribution, according to criteria and modalities communicated by the instructor at the beginning of the course. The overall evaluation will therefore take into account conceptual correctness, the ability to formalize and reason about problems, command of scientific language, the quality and coherence of the project work, and the ability to justify and discuss the adopted choices. Detailed operational criteria, including project format, any group-work rules, and the evaluation modalities for the presentation and discussion, will be communicated by the instructor at the beginning of the course.

Where applicable within the degree program and in accordance with current regulations, final thesis assignment may concern topics consistent with the course contents, such as agent models, search and planning techniques, sequential decision making, reinforcement learning, and the analysis of modern agentic AI systems. Depending on the profile of the proposed work, thesis topics may be theoretical, analytical, evaluative, or application-oriented, provided that they remain coherent with the course contents and the student’s study path. Assignment will be based on the student’s interest, coherence with the study path, feasibility of the proposed work, and the instructor’s availability. In cases where the topic has interdisciplinary relevance, proposals connecting technical foundations with organizational, decision-oriented, or socio-technical questions may also be considered, where consistent with the degree program and applicable regulations.