Code Labs - 01: Agent Types

Build single agents from scratch across three complexity levels in all four frameworks. Each level adds a new capability layer on top of the previous.

← Back to Overview: Agents · Back to Concepts: Agent Fundamentals · Agent Patterns

Complexity Progression

Level 1: Simple           Level 2: Intermediate      Level 3: Complex
─────────────────         ─────────────────────       ────────────────────
ReAct loop                + Memory                    + Planning
Tool binding              + Multi-tool                + Reflection
Basic Gemini call         + Structured output         + Streaming

Labs by Framework and Level

Google ADK

Level	File	What You Learn
Simple	`ADK/01-simple/agent.ipynb`	ReAct agent, tool binding, Gemini model config
Intermediate	`ADK/02-intermediate/agent.ipynb`	Session memory, multi-tool orchestration, output schema
Complex	`ADK/03-complex/agent.ipynb`	Planning steps, self-reflection, streaming response
Real-world	`ADK/vertex-ai-real-world/agent.ipynb`	Production ADK on Vertex AI with live APIs

LangChain

Level	File	What You Learn
Simple	`LangChain/01-simple/agent.ipynb`	LCEL chain, tool calling, message history
Intermediate	`LangChain/02-intermediate/agent.ipynb`	ConversationBufferMemory, multi-tool routing
Complex	`LangChain/03-complex/agent.ipynb`	Plan-and-execute, output parsers, callbacks
Real-world	`LangChain/real-world/agent.ipynb`	Tavily web search, live data integration

LangGraph

Level	File	What You Learn
Simple	`LangGraph/01-simple/agent.ipynb`	Graph nodes, edges, StateGraph, basic tool call
Intermediate	`LangGraph/02-intermediate/agent.ipynb`	Checkpointer memory, conditional edges, tool router
Complex	`LangGraph/03-complex/agent.ipynb`	Subgraphs, reflection loop, streaming events
Real-world	`LangGraph/real-world/agent.ipynb`	Stateful conversation with web retrieval

CrewAI

Level	File	What You Learn
Simple	`CrewAI/01-simple/agent.ipynb`	Agent + Task + Crew, role assignment
Intermediate	`CrewAI/02-intermediate/agent.ipynb`	Shared crew memory, multi-tool agents
Complex	`CrewAI/03-complex/agent.ipynb`	Sequential task chaining, delegation, callbacks
Real-world	`CrewAI/real-world/agent.ipynb`	Research crew with live web tools

Framework Comparison at Each Level

Capability	ADK	LangChain	LangGraph	CrewAI
Tool binding	`@tool` decorator	`@tool` + bind_tools	ToolNode	Agent tools list
Memory	Session service	ConversationBuffer	Checkpointer	Crew memory
State model	Session dict	Message list	TypedDict graph state	Shared crew context
Streaming	`stream_query`	`.stream()`	`.stream_events()`	Callback handlers

Getting Started

# Start with the simplest ADK notebook
jupyter notebook 05-Agents/CodeLabs/01-Agent-Types/ADK/01-simple/agent.ipynb

# Or LangChain if you prefer
jupyter notebook 05-Agents/CodeLabs/01-Agent-Types/LangChain/01-simple/agent.ipynb

After completing all four simple agents, compare: same task, four different approaches. The differences in state management, tool binding, and model calls become immediately apparent.

What to Read Alongside

Agent Fundamentals - the ReAct loop, tool use, memory types
Agent Patterns - planning and reflection patterns used at Level 3
Framework-specific deep-dives: GCP ADK · LangChain · LangGraph · CrewAI

Next: Architecture Patterns

Once you've built single agents, move to 02 - Architecture Patterns to see how multiple agents coordinate.