Module 10: Agent Frameworks
The Framework Paradox
LangChain has over 90,000 GitHub stars. It also has a reputation for being the hardest LLM framework to debug. Both things are true - and that tension tells you something important about the state of agent frameworks in 2025.
Frameworks promise to save you time. They deliver pre-built integrations, opinionated patterns, and community support. But every abstraction is a bet: the framework authors bet that their model of the problem matches yours. When it does, you move fast. When it does not, you spend hours unwrapping abstractions to understand what is actually happening.
This module gives you the knowledge to make that choice deliberately - not because you saw a framework trending on Twitter, but because you understand the trade-offs.
Module Map
Lesson Guide
| Lesson | Topic | Key Skills |
|---|---|---|
| 01 | When to Use a Framework | Choose the right tool for the job with a decision framework |
| 02 | LangChain | Understand LCEL, chains, agents, and LangSmith tracing |
| 03 | LangGraph | Build stateful agents with graphs, checkpointing, and streaming |
| 04 | LlamaIndex | Connect agents to complex data with advanced retrieval |
| 05 | CrewAI | Build multi-agent teams with Flows and production memory |
| 06 | AutoGen | Orchestrate conversational agent teams with MagenticOne |
| 07 | Raw API | Build complete agents directly on the Anthropic SDK |
| 08 | Framework Comparison | Score frameworks across 12 dimensions for your use case |
| 09 | Production Lessons | The 10 lessons only production experience teaches |
What You Will Build
Across this module, you will implement the same core use case - a research-and-summarize agent - in multiple frameworks and in raw API style. Seeing the same problem solved five different ways gives you the comparison you cannot get from reading documentation alone.
By the end, you will be able to:
- Choose the right framework (or no framework) for a given agent task
- Navigate LangChain, LangGraph, LlamaIndex, CrewAI, and AutoGen with confidence
- Build production-grade agents directly on the Anthropic SDK
- Identify when your framework is working against you and how to escape it
The final lesson distills a year of production agent experience into ten hard-won lessons. Read it before you deploy anything.