Agentic Large Language Models represent a rapidly emerging frontier in artificial intelligence, integrating generative modeling with the capacity to reason, act, and interact autonomously. This graduate‑level textbook provides the first systematic and comprehensive treatment of agentic LLMs, offering a unified framework that spans foundational principles, computational mechanisms, and advanced applications.

The book surveys the full methodological landscape: the architecture and training pipeline of LLMs; scaling laws; multi-step reasoning and self-reflection; multimodal vision-language-action models; world-model construction; interaction styles and memory architectures; and agentic tool-chains, including the Model Context Protocol and the robustness, safety, and security concerns that arise once agents can act in the world. It further examines behavioral and cognitive dimensions of agentic systems, including Theory of Mind, strategic behavior, emergent norms, and collective intelligence in multi-agent environments. Advanced chapters address synthetic data generation, mechanistic interpretability, and cognitive architectures, highlighting open challenges and research opportunities.

Four hands-on case studies take readers from training a small language model from scratch, through benchmarking reasoning and self-reflection on games and puzzles, to building a multi-agent medical diagnosis system and simulating a society of interacting LLM agents. Written for graduate students and LLM developers with foundational AI knowledge and supported by Python-based examples and open code repositories, this textbook serves as both a course text and a reference for the theoretical foundations, computational models, and future directions of agentic LLM research.

Max van Duijn is Assistant Professor at Leiden University, specializing in the intersection of cognitive science, linguistics, and artificial intelligence. His research examines the mechanisms underlying social intelligence—empathy, perspective‑taking, and Theory of Mind—and models these capacities in both humans and AI systems, including Large Language Models. He previously edited a Springer LNCS volume on misinformation in online media.

Michiel van der Meer is a Researcher at Leiden University whose work focuses on open, agentic LLMs that reason over human values, opinions, and arguments—engineered end-to-end across the pipeline (data, training, alignment, inference, evaluation) to faithfully represent the diversity of a society.

Aske Plaat is Full Professor of Artificial Intelligence at Leiden University. His research spans large reasoning models, machine learning, and AI‑generated music, with a particular focus on computational strategies for complex decision‑making. He is the author of two Springer textbooks on reinforcement learning—Learning to Play: Reinforcement Learning and Games (2020) and Deep Reinforcement Learning (2022)—both widely used in graduate‑level AI education.

Niki van Stein is Associate Professor at Leiden University, where she works on explainable AI for machine learning and optimization. Her research combines interpretability with metaheuristic and evolutionary optimization, including the use of large language models for automated algorithm design and code generation. She co-edited the Springer volume Explainable AI for Evolutionary Computation and develops open-source tools for analyzing and benchmarking black-box optimization algorithms.