Core Concepts

The goal of this chapter is not to make readers memorize a pile of AI terms, but to establish one clear main line:

Start with a large language model that can only predict the next piece of text, then understand how it gradually becomes a system that can converse, remember, search, call tools, run workflows, and collaborate.

Once this main line is clear, terms like LLM, Prompt, Context, Memory, Agent, RAG, MCP, Workflow, and Skill stop feeling disconnected.

What This Chapter Covers

This chapter is organized into 5 progressive stages:

1. Understand the model itself: LLM, Prompt, Context, Memory, Token
2. Understand external capabilities: Agent, Search, RAG
3. Understand tool conventions: Function Calling, MCP
4. Understand process structuring: Workflow, Skill
5. Understand collaboration and expansion: SubAgent, A2A

Reading Order

1. Large Language Model (LLM)
2. Prompt
3. Context
4. Memory
5. Token
6. Agent
7. Search
8. Retrieval-Augmented Generation (RAG)
9. Function Calling
10. Model Context Protocol (MCP)
11. Workflow

Learning Framework

The most important takeaway in this chapter is not the definition of any single term, but this unifying idea:

Most AI application-layer concepts are ultimately solving the same problem: how to place better information into the context.

You can read them like this:

• Prompt tells the model what you want it to do
• Context gives the model the background of the current task
• Memory carries forward prior dialogue and rules
• Search brings in fresh public information
• RAG brings in private knowledge-base content
• Function Calling lets the model express which tool it wants to use in a structured way
• MCP lets tools connect to the system in a standardized way
• Agent orchestrates the model, tools, retrieval, and process
• Workflow solidifies repeated steps
• Skill packages reusable know-how, instructions, and scripts
• SubAgent isolates context through task decomposition
• A2A lets multiple agents collaborate and pass results

How to Study This Chapter

• Do not memorize these terms as isolated glossary entries. Focus on what problem each one solves from the previous stage.
• Always look at them from the perspective of 「system evolution」: how the model is gradually extended with memory, retrieval, tools, and process capabilities.
• Focus especially on a few easily confused pairs: Prompt vs Context, Context vs Memory, Search vs RAG, and Function Calling vs MCP

What You Can Gain After Finishing This Chapter

• Know which layer the common AI terms belong to
• Understand that these concepts are not a flat list, but a connected system evolution
• Understand the relationships among models, context, memory, retrieval, tools, and process
• Clearly distinguish between 「the model's own capabilities」 and 「system-level external capabilities」
• Quickly judge what problem a new concept is trying to solve