AI Agents Roadmap

A structured path from Python and LLM fundamentals through building, evaluating, and deploying production-grade AI agents and multi-agent systems.

✓ Every resource link below is verified live.

1. Stage 1: Foundations

Python for AI Engineering
Agents are built and orchestrated primarily in Python
docPython Official Docs coursefreeCodeCamp Python Full Course
LLM Concepts & Prompt Engineering
Agents are powered by LLMs; prompting controls their behavior
docOpenAI Prompt Engineering Guide tutorialPrompting Guide (promptingguide.ai)
APIs & HTTP Basics
Agents call external APIs as tools to act on the world
docMDN HTTP Overview tutorialRequests Library Quickstart

OpenAI & Anthropic API Usage
Most agent frameworks wrap these canonical LLM APIs
docOpenAI API Reference docAnthropic API Docs
Function Calling & Tool Use
Tool calling lets agents take actions beyond text generation
docOpenAI Function Calling Guide docAnthropic Tool Use Guide
Structured Outputs
Reliable JSON output is critical for agent decision pipelines
docOpenAI Structured Outputs tutorialInstructor Library Docs

ReAct & Chain-of-Thought Patterns
ReAct is the foundational reasoning loop most agents implement
tutorialLangChain ReAct Agent Docs docPrompting Guide: ReAct
LangChain Agents
LangChain is the most widely adopted agent framework in industry
docLangChain Agents Conceptual Docs tutorialLangChain Quickstart Tutorial
LlamaIndex Agent Framework
LlamaIndex specializes in data-grounded agents and RAG pipelines
docLlamaIndex Agents Docs tutorialLlamaIndex Starter Tutorial
OpenAI Agents SDK
Official lightweight SDK models the agent loop with minimal abstraction
docOpenAI Agents SDK Docs tutorialOpenAI Agents Quickstart

Retrieval-Augmented Generation (RAG)
RAG gives agents grounded, up-to-date domain knowledge
docLangChain RAG Tutorial tutorialfreeCodeCamp RAG Tutorial
Vector Databases
Vector stores are the persistence layer for agent long-term memory
docChroma Docs docPinecone Docs
Agent Memory Patterns
Short and long-term memory determines agent coherence over sessions
docLangChain Memory Concepts docLlamaIndex Memory Docs

Multi-Agent Orchestration with LangGraph
Graph-based orchestration handles complex multi-step agent workflows
docLangGraph Docs tutorialLangGraph Quickstart
Model Context Protocol (MCP)
MCP is the emerging standard for agent-to-tool interoperability
docMCP Official Docs docAnthropic MCP Announcement
AutoGen Multi-Agent Framework
AutoGen enables conversational collaboration between multiple agents
docAutoGen Docs tutorialAutoGen Getting Started

Agent Evaluation & Benchmarking
Systematic evals are required to ship reliable agents to production
docLangSmith Evaluation Docs docOpenAI Evals Framework
Guardrails & Output Validation
Guardrails prevent harmful, off-topic, or malformed agent outputs
docGuardrails AI Docs docOpenAI Agents SDK Guardrails
Prompt Injection & Security
Agents acting on tools make them a high-value attack surface
docOWASP LLM Top 10

Serving Agents with FastAPI
FastAPI is the standard for exposing agents as production HTTP services
docFastAPI Official Docs tutorialFastAPI Tutorial
Observability & Tracing
Distributed tracing makes agent reasoning debuggable in production
docLangSmith Tracing Docs docArize Phoenix Docs
Containerization & Deployment
Docker and cloud deployment make agents scalable and reproducible
docDocker Official Docs tutorialfreeCodeCamp Docker Handbook
Portfolio & Job Readiness
Tangible projects and OSS contributions signal real agent engineering skill
tutorialGitHub Docs: Creating a Portfolio docLangChain Use Case Templates

Want this taught by an AI tutor — with lessons, quizzes, flashcards, and progress tracking?