From understanding what an AI agent is to building production multi-agent systems. Three tracks — pick the one that matches where you are right now.
✓Understand what agentic AI is and how it differs from standard AI
✓Know the four major frameworks: LangChain, CrewAI, AutoGen, LlamaIndex
✓Understand agent safety, MCP, and production deployment considerations
Free foreverNo signup requiredProgress saved in your browser8 full reference guides included
Who this is forNo prior AI knowledge required. You use AI tools but don't know what "agents" means.
Time~3 hours across 5 lessons
Next stepIntermediate track
Your progress0 of 5 lessons
Lesson 1
What is AI? The starting point
Before agents, you need a clear picture of what AI actually is — not the hype version, the real one. What language models do, how they generate text, and why they're different from software you've used before.
Lesson 2
What is Agentic AI?
The single biggest shift in AI right now. Standard AI answers questions. Agentic AI completes projects. Learn exactly what that difference means, where you already encounter agents, and why it matters.
Lesson 3
How AI Agents Work
The agent loop — perceive, plan, act, observe — explained from first principles. What tools are, why they matter, and the three most common ways an agent can fail.
Lesson 4
Multi-Agent Systems
One agent doing everything has limits. A team of agents — each with a role, coordinated by an orchestrator — can handle far more complex tasks. Learn how the team model works.
Lesson 5
AI tools that use agents today
ChatGPT with tools on, Claude with tool use, Perplexity — you may already be using agents without knowing it. This lesson connects the concepts to the tools you know.
Question 1
What is the key difference between a standard AI chatbot and an AI agent?
Correct. Standard AI responds to prompts and waits for the next input. An agent receives a goal, plans and executes the steps required to achieve it, and continues until the task is complete — without requiring a prompt at every step.
Question 2
What are the four core components every AI agent has?
Correct. Every agent — regardless of which framework built it — has: Perception (what it can observe), Memory (what it can remember), Planning (how it decides what to do next), and Action (what it can do with tools).
Question 3
In a multi-agent system, what is the role of the orchestrator agent?
Correct. The orchestrator coordinates the team — it does not necessarily do the detailed work itself. It breaks the goal into sub-tasks, assigns them, monitors progress, and assembles the final result.
Question 4
Which of these is an example of an agentic AI tool you can use today?
Correct. Perplexity is an agent — it decides which sources to search, retrieves them, and synthesises results without you directing each step. The calculator, FAQ page, and spell checker are all non-agentic: they respond to inputs but do not plan or take autonomous actions.
Question 5
What does MCP (Model Context Protocol) standardise?
Correct. MCP is the "USB port" standard for AI agents — it defines a universal interface so any tool built to the MCP standard works with any agent built to the MCP standard, regardless of which framework or model is used.
Ready for the next level?
The Intermediate track covers the four major frameworks: LangChain, CrewAI, AutoGen, and LlamaIndex.
Who this is forYou understand what AI agents are. You want to learn the frameworks used to build them.
Time~5 hours across 5 lessons
PrerequisiteBeginner track or equivalent knowledge
Your progress0 of 5 lessons
Lesson 1
LangChain & LangGraph
The most widely used framework. LangChain provides the ecosystem of 1,000+ integrations; LangGraph provides the graph-based state machine for reliable multi-step agent workflows. Start here for complex pipelines.
Lesson 2
CrewAI
The easiest framework to get started with. Define agents with a role, goal, and backstory — like hiring a team. Sequential or hierarchical task execution. The fastest path from idea to working multi-agent system.
Lesson 3
Microsoft AutoGen
Conversation-based multi-agent framework from Microsoft Research. Agents communicate by sending messages to each other. Important: AutoGen is in maintenance mode and merging into AG2 — understand what this means before building with it.
Lesson 4
LlamaIndex
Specialises in data-augmented agents — AI that reasons over your own documents, databases, and knowledge bases using RAG. The go-to choice when your agent needs to work with proprietary or large-scale data.
Lesson 5
AI Automation tools for agents
n8n, Make, and Zapier provide no-code and low-code ways to connect AI agents to real-world workflows — CRM updates, email sending, Slack notifications, database writes. The operational layer underneath your agents.
Question 1
What is the key difference between LangChain and LangGraph?
Correct. LangChain provides the components (model connectors, tools, document loaders). LangGraph provides the architectural pattern for assembling those components into reliable, stateful, cyclical agent workflows. The two work together.
Question 2
In CrewAI, what are the three fields that define every agent?
Correct. CrewAI's defining feature is role-based agent definition: each agent has a Role (what it is), a Goal (what it is trying to achieve), and a Backstory (how it approaches its work). These three fields shape every decision the agent makes.
Question 3
Why should you think carefully before starting a new project with AutoGen?
Correct. AutoGen is actively maintained for bug fixes only. It is being merged into Microsoft's Agent Framework (AG2). New projects should evaluate AG2 or actively developed alternatives like CrewAI or LangGraph before committing to AutoGen.
Question 4
What does RAG stand for and what problem does LlamaIndex help solve with it?
Correct. RAG solves the problem of AI models not knowing about your specific data (internal documents, product info, etc.). Instead of retraining the model, you retrieve relevant document chunks at query time and pass them as context. LlamaIndex provides the most capable tooling for building and optimising RAG pipelines.
Question 5
Which framework would you choose for an agent that needs to query a large internal document knowledge base?
Correct. LlamaIndex is specifically designed for agents that reason over large document sets. It provides advanced retrieval features (hybrid search, reranking, HyDE, sentence window retrieval) that the other frameworks do not offer out of the box.
Ready to go deeper?
The Advanced track covers multi-agent safety, MCP in depth, and production deployment.
Who this is forYou understand the frameworks. You want to understand agent safety, MCP, and production deployment.
Time~8 hours across 5 lessons
PrerequisiteIntermediate track or equivalent experience
Your progress0 of 5 lessons
Lesson 1
Multi-Agent Systems — architecture in depth
Graph architectures, DAGs, the actor model, hierarchical task networks. Cascading failure, prompt injection in multi-agent systems, and the inter-agent communication standards (MCP, A2A) emerging in 2026.
Lesson 2
Agent safety — OWASP Top 10 for Agentic AI
The first formal taxonomy of agentic AI security risks. Prompt injection, insecure tool permissions, uncontrolled resource consumption, goal misalignment, memory poisoning. What each means and how to mitigate it in production systems.
Lesson 3
MCP — Model Context Protocol in depth
How MCP works at the protocol level — JSON-RPC over stdio and SSE, tool schema definitions, server and client architecture. How LangGraph, CrewAI, and Claude implement MCP natively. How to build an MCP server.
Lesson 4
LangGraph in production — checkpointing, streaming, evaluation
Persistence backends (SQLite, PostgreSQL), time travel debugging with LangSmith, streaming intermediate outputs, RAGAS evaluation framework, and cost management strategies for multi-step agent workflows.
Lesson 5
EU AI Act and regulatory compliance for agents
High-risk classifications, mandatory documentation requirements, human oversight obligations. Which agentic deployments require conformity assessments and what that means for teams building in the EU. Provisions taking full effect August 2026.
Question 1
According to the OWASP Top 10 for Agentic AI, what is the primary risk category?
Correct. Prompt injection is the top-ranked risk — malicious instructions embedded in content the agent reads (web pages, documents, emails) can override the agent's goals and cause it to take unintended actions. The attack surface is large: every external source any agent reads is a potential vector.
Question 2
What does LangGraph's checkpointing system enable that simple agent loops do not?
Correct. Checkpointing saves the full state at every node execution. This enables: resuming interrupted workflows from the last checkpoint, pausing for human review and modifying state before continuing, and replaying exact failure conditions from historical checkpoints for debugging.
Question 3
Under the EU AI Act, when do high-risk provisions for agentic AI deployments take full effect?
Correct. The EU AI Act (Regulation 2024/1689) came into force in August 2024. High-risk provisions — including mandatory conformity assessments and human oversight requirements for agentic systems in high-risk domains — take full effect from August 2026.
Question 4
What transport protocols does MCP use for communication between agent clients and tool servers?
Correct. MCP defines two transport options: stdio (for local tool servers running as child processes) and HTTP with Server-Sent Events (for remote tool servers). Both use JSON-RPC as the underlying message format. This is specified in the official MCP specification at modelcontextprotocol.io.
Course complete
You've covered the full agentic AI stack — concepts, frameworks, safety, and production. Explore the full reference guides for anything you want to go deeper on.