SESSION 8

Session 8 — Building Agents with LangChain

Build your first agent, add memory, streaming, structured output, and multi-agent handoffs.

3 hours•11 exercises · 3 groups

What you'll build by the end

✓ A ReAct agent that autonomously reasons and calls tools
✓ A multi-turn chatbot with persistent memory
✓ A streaming agent that shows progress in real time
✓ An agent gated by human approval for dangerous operations
✓ A sales/support handoff system with state-driven routing
✓ A supervisor that orchestrates research + writing subagents

Prerequisites

Finished Session 6 (LangChain basics) and Session 7 (architecture choice)Session8 folder with .venv, langchain + langchain-openai + langgraph installed OPENAI_API_KEY in Session8/.env Comfortable with @tool, bind_tools, Pydantic from Session 6

The arc

Single agent

00–03

Memory & output

04–07

Advanced

08–10

Exercises

Single agent

Create, inspect, configure, and error-proof your first agent.

Your First Agent (ReAct Pattern)

Messages in the Agent Loop

Static vs Dynamic Tools

Tool Error Handling

Memory & output

Add persistence, handle long chats, stream tokens, return structured data.

Short-Term Memory (Checkpointer)

Managing Long Conversations

Streaming Agent Output

Structured Agent Output

Advanced

Human approval gates, conversational handoffs, supervisor orchestration.

Human-in-the-Loop

Multi-Agent Handoff

Multi-Agent Subagents (Supervisor)

When you get stuck or finish

Troubleshooting

Agent-specific errors: middleware bugs, checkpointer issues, handoff state.

End-of-session quiz

12 questions — score 9+ and head to the capstones.