This chapter sets the foundation for constructing LLM applications designed for autonomous operation. We will first clarify what defines 'agency' within the context of Large Language Models. Then, we will break down the typical components that form an autonomous system, including the core model, memory structures, planning processes, and action execution mechanisms.

You will learn about the inherent difficulties in designing these systems, such as mitigating inaccurate outputs (hallucinations), enabling effective long-range planning, and achieving reliable interaction with external tools. To conclude, we provide a brief overview of prevalent frameworks like LangChain, AutoGen, and CrewAI, viewed from an architectural standpoint, which are often used to assemble these complex systems.