What an AI agent is made of

The survey first defines the components of an AI agent:

• The LLM, with a defined role and task, serving as the reasoning engine and dialogue interface.
• Memory, both short-term (the immediate conversation state) and long-term (accumulated knowledge and past experience). This connects directly to the agentic memory topic.
• Planning, including reflection and self-critique, which guides iterative reasoning and breaks complex tasks down.
• Tools, such as vector search, web search, and APIs, which extend the agent beyond text generation.

With those parts in mind, the survey introduces four design patterns that turn a plain model into an agent.