Skip to content

01 — Agent Types

What This Section Teaches

A single AI agent, built four different ways, growing in complexity across three levels. The goal is to understand how each framework approaches the same core problem — and how the agent itself evolves as you add memory, planning, and reflection.

The Three Levels

SIMPLE                    INTERMEDIATE               COMPLEX
──────                    ────────────               ───────
One goal                  Multi-tool                 Goal decomposition
1–2 mock tools            Short-term memory          Planning step
No memory                 Structured output          Self-critique + retry
One-shot response         Tool chaining              Streaming output

Concepts: ReAct loop      Concepts: memory,          Concepts: planning,
tool binding, prompt      output parsers             reflection, autonomy

Framework Comparison

Feature LangChain LangGraph CrewAI ADK
Mental model Chain of callables State machine Role-playing crew Configurable agent object
State Message history Typed graph state Shared crew context Session-based
Tool binding .bind_tools() Node functions @tool decorator FunctionTool
Memory ConversationBufferMemory State dict Crew context session.state
Streaming .stream() .stream() Built-in runner.run_async()
GCP / Vertex AI Via LangChain-Google Manual Manual Native

Folder Structure

01-Agent-Types/
├── LangChain/
│   ├── 01-simple/         agent.ipynb + agent.py
│   ├── 02-intermediate/   agent.ipynb + agent.py
│   ├── 03-complex/        agent.ipynb + agent.py
│   └── real-world/        Tavily web search example
├── LangGraph/             same structure
├── CrewAI/                same structure
└── ADK/
    ├── 01-simple/
    ├── 02-intermediate/
    ├── 03-complex/
    └── vertex-ai-real-world/    ADK + Vertex AI on GCP
  1. Pick one framework (recommended: ADK for GCP focus)
  2. Work through 01-simple → 02-intermediate → 03-complex in that framework
  3. Then read the same level in another framework to compare
  4. Finish with the real-world example of your primary framework