4.1.1. Agent Architecture Concepts

AI agents go beyond simple prompt-response patterns by combining LLMs with tools, memory, and planning capabilities. An agent can break complex tasks into steps, use tools to gather information, and iterate until the goal is achieved.

What makes something an "agent" vs. a simple chatbot:

• Autonomy: Makes decisions about which tools to call and when
• Reasoning loop: Observe → Think → Act → Observe cycle (ReAct pattern)
• Memory: Maintains state across multiple reasoning steps
• Tool use: Can call external APIs, run code, search databases

Agent patterns tested on the exam:

• ReAct (Reasoning + Acting): Think-aloud approach with tool calls
• Plan-and-Execute: Create full plan first, then execute steps
• Reflection: Self-critique and improve before final answer