8. References

Glossary

Term	Definition
Agent	Autonomous AI entity that uses LLMs, tools, and memory to accomplish goals
Tool	A function an agent can call to perform actions (APIs, code, search)
Handoff	Transferring conversation control from one agent to another
Orchestrator	Manages agent coordination, routing, and workflow sequencing
MCP	Model Context Protocol — standardized interface for agent tools
RAG	Retrieval-Augmented Generation — grounding LLM responses in data
Structured Output	LLM returns typed, schema-validated data instead of free text
Guardrail	Safety policy applied to agent inputs and outputs

Cheatsheet

# Installation Cheatsheet

# OpenAI Agents SDK
pip install openai-agents

# CrewAI
pip install crewai crewai-tools

# LangGraph
pip install langgraph langchain-openai

# AutoGen
pip install pyautogen

# MCP
pip install mcp

# Gradio (UI)
pip install gradio

# Common imports
from agents import Agent, Runner, handoff
from crewai import Agent, Task, Crew, Process
from langgraph.graph import StateGraph, END
from autogen import AssistantAgent, UserProxyAgent

API Reference

Framework	Core Classes / Functions
OpenAI SDK	`Agent`, `Runner`, `Trace`, `handoff`, `function_tool`, `Guardrail`
CrewAI	`Agent`, `Task`, `Crew`, `Process`, `@tool`, `knowledge`
LangGraph	`StateGraph`, `add_node`, `add_edge`, `compile`, `Checkpointer`
AutoGen	`AssistantAgent`, `UserProxyAgent`, `GroupChat`, `GroupChatManager`
MCP	`Server`, `ClientSession`, `Tool`, `list_tools`, `call_tool`

Key Lessons Learned

Start simple — add complexity only when needed
Use structured outputs for reliable data exchange
Always add guardrails for production agents
Memory is critical for multi-turn conversations
MCP makes tools reusable across frameworks
Monitor and trace agent decisions in production

📘 Congratulations! You've completed the AI Engineer Agentic Track. You now have hands-on experience with all major agent frameworks and can build sophisticated multi-agent systems for real-world applications.