Module 8: AI Agents & Agentic RAG

Learning Objectives

• Build AI agents that reason and use tools
• Implement agentic RAG with dynamic retrieval
• Design multi-agent systems for complex tasks
• Use LangGraph for agent orchestration

Why This Matters

Basic RAG answers simple questions. Agentic RAG handles complex, multi-step questions that require reasoning, multiple data sources, and tool use. This is where AI applications become genuinely useful for enterprise workflows.

Lesson Content

Basic RAG retrieves and generates. Agentic RAG reasons about WHAT to retrieve, WHEN to use tools, and HOW to break complex questions into steps. The AI agent decides the retrieval strategy dynamically based on the query.

From Pipeline to Agent

In basic RAG, the pipeline is fixed: embed → retrieve → generate. In agentic RAG, the LLM decides: should I search the knowledge base? Query the database? Search the web? Calculate something? The agent orchestrates tools based on reasoning.

Tool Calling

tools = [
    {"name": "search_docs", "description": "Search internal documentation"},
    {"name": "query_database", "description": "Run SQL query on the product database"},
    {"name": "web_search", "description": "Search the internet for recent information"},
]

# The agent decides which tool to use based on the question
# "What is our refund policy?" → search_docs
# "How many orders last month?" → query_database
# "What is the latest Python version?" → web_search

Multi-Agent Architectures

Complex tasks are split across specialized agents: a Research Agent retrieves information, an Analysis Agent processes data, a Writing Agent generates the final response. LangGraph orchestrates the flow between agents.

Memory Systems

• Short-term memory: Conversation history within a session
• Long-term memory: Persistent storage of user preferences, past interactions
• Working memory: Intermediate results during multi-step reasoning

Common Mistakes

• Building agents without reliable basic RAG first
• Not limiting agent tool calls (runaway agents are expensive)
• Not implementing guardrails for tool access (agent with SQL access can drop tables)
• Using agents for simple questions (overkill adds latency and cost)

Key Terms

Agentic RAG

RAG where the AI decides retrieval strategy dynamically

Tool Calling

LLM capability to invoke external functions (search, SQL, APIs)

LangGraph

Framework for building stateful multi-agent workflows

Multi-Agent

Architecture with specialized agents collaborating on tasks

Hands-On Labs

1. Build a Multi-Tool Agent

   Create an AI agent that reasons about tool selection.

   40 min - Advanced

   • Define tools: search, database, calculator
   • Implement the agent loop (reason → act → observe)
   • Test with queries requiring different tools
   • Handle multi-step queries requiring multiple tools

   View lab files on GitHub

2. Multi-Agent Orchestration with LangGraph

   Build a multi-agent system for complex tasks.

   45 min - Advanced

   • Design agent roles: researcher, analyst, writer
   • Build a LangGraph workflow connecting agents
   • Test with complex questions requiring collaboration
   • Add memory for conversation continuity

   View lab files on GitHub

Key Takeaways

• Agentic RAG decides WHAT to retrieve dynamically — not a fixed pipeline
• Tool calling gives agents access to databases, search, APIs, and calculators
• Multi-agent systems split complex tasks across specialized agents
• LangGraph orchestrates agent workflows with state management
• Memory systems enable context-aware multi-turn interactions