The AI Agent Stack

The AI agent stack is the set of representative building blocks (layers) needed to turn a generative AI model into an AI agent that works in practice. Terms like RAG, Memory, Tools, MCP, Skills, and Hooks show up constantly across AI agent articles — this page organizes them into a single map.

Why an Overview Matters

There are many terms related to AI agents. Learning each one in isolation, without understanding what problem it solves, makes it easy to lose the overall picture. The four-layer structure covered in Prompt Engineering, Context Engineering, Harness Engineering, and Loop Engineering — instructions, information, execution environment, and repeat structure — can be broken down further into twelve elements commonly used in practice.

The 12 Elements of the AI Agent Stack

Scroll horizontally→

#	Element	Role	Details
1	RAG	Pulls in outside knowledge before answering	What Is RAG?
2	Context	Everything the model can see right now	Context Engineering
3	Memory	What’s remembered across turns and sessions	What Is Memory
4	Agent	An AI that acts, not just answers	What Is an AI Agent?
5	Agentic AI	Systems that plan, decide, and self-correct	What Is Agentic AI
6	Tools	External functions the agent can call	What Is Tool Use
7	MCP	One open standard to connect tools and data	What Is MCP?
8	Skills	Reusable know-how loaded on demand	What Are Skills
9	Hooks	Triggers that run logic at set points	What Are Hooks
10	Subagents	Helpers the main agent hands work to	What Are Subagents
11	Orchestration	Coordinates the whole multi-step workflow	What Is Orchestration
12	Eval	Measures whether the system actually works	What Is AI Evaluation?

This table scrolls horizontally. Keyboard users can focus the table and use the left and right arrow keys.

How the Layers Relate

graph TD
    RAG["RAG\nPull in outside knowledge"] --> Context["Context\nWhat's visible right now"]
    Memory["Memory\nPersists across sessions"] --> Context
    Context --> Agent["Agent / Agentic AI\nDecide and act"]
    Skills["Skills\nReusable procedures"] --> Agent
    Agent --> Tools["Tools\nCall external functions"]
    Tools --> MCP["MCP\nStandard connection"]
    Agent --> Subagents["Subagents"]
    Subagents --> Orchestration["Orchestration\nCoordinate the workflow"]
    Hooks["Hooks\nAutomatic triggers"] -.-> Agent
    Orchestration --> Eval["Eval\nMeasure the outcome"]

• RAG, Context, and Memory are input-side layers that determine what the agent knows
• Agent, Agentic AI, Tools, MCP, and Skills are execution-side layers that determine what the agent can do
• Hooks insert automatic processing at key points in execution
• Subagents and Orchestration handle how work is divided and coordinated across multiple units
• Eval verifies whether all of the above is actually working

Summary

• The AI agent stack maps out the twelve representative layers that make up an AI agent
• It breaks down into input (RAG, Context, Memory), execution (Agent, Tools, MCP, Skills), control (Hooks, Subagents, Orchestration), and verification (Eval)
• No single element works in isolation — together they form one AI agent