Скачать или смотреть Generative and Agentic AI: Concepts, Frameworks, and Engineering

This is a comprehensive educational overview of the transition from Generative AI (GenAI), exemplified by Transformer-based Large Language Models (LLMs), to Agentic AI, which focuses on autonomous action and goal-directed decision-making. The material contrasts these concepts, highlighting that agentic systems embed LLMs (as the "brains") within a broader framework that allows for tool use, environmental interaction, and iterative planning through mechanisms like the "think-act-observe" loop. A significant portion of the sources is dedicated to outlining practical aspects of engineering these systems, including various prompt engineering techniques (e.g., Chain-of-Thought), distinct agent architectures (Reactive, Planning, Self-Reflective), and the complexities of coordinating Multi-Agent Systems (MAS) using defined communication protocols. Finally, the notes summarise key frameworks (LangChain, AutoGen, crewAI) and common design patterns (Orchestrator–Worker, Router, Evaluator–Optimiser) essential for building reliable agentic workflows, concluding with discussions on real-world data science applications and crucial safety/governance risks associated with increased autonomy.