RAG Fundamentals for Everyone
Master the core building blocks of RAG, from embeddings to agentic retrieval.
BeginnerFree
Start learning
Loading...
Loading...
An AI agent is an LLM that can call tools, reflect on results, and decide what to do next. The interesting engineering lives in the decide-what-to-do-next loop. Failure recovery, tool selection, context management, and knowing when to stop.
Curated by Param Harrison
These courses teach agents two ways: from scratch so you understand the loop, and with frameworks (LangGraph, AutoGen) so you can ship quickly. Multi-agent patterns, voice agents, and design patterns for production systems are covered separately, so you can pick the angle that matches your use case.
Showing 20 of 20 courses
Common questions
ReAct is a pattern, LangGraph is a framework that implements that pattern plus observability, checkpointing, and branching. Start with ReAct from scratch to understand the loop, then adopt LangGraph when you need the operational features.
LangGraph is the most production-ready and has the best observability hooks. AutoGen is strong for multi-agent conversation patterns. CrewAI is easy to get started but the abstractions can obscure debugging. Courses cover LangGraph and AutoGen because those are what teams actually deploy.
Record traces in development, build a fixture set of representative user turns, and assert on tool-call sequences and final outputs. The observability course covers this setup. Agent testing is closer to integration testing than unit testing.
One agent is almost always enough to start. Multi-agent helps when tasks naturally decompose (research + write, plan + execute) or when you need different models per role. Don’t add agents to solve what’s really a prompt problem.
MCP is a standard for how an agent calls tools. Most agent courses here use direct function calling; the MCP course shows the portable version. Ship agents with direct tools first, migrate to MCP when you need multi-client reuse.