Agent Frameworks Compared: LangChain, LlamaIndex and Others
For engineers choosing an LLM application framework: understand what LangChain, LangGraph, LlamaIndex, Pydantic AI, DSPy, Haystack, CrewAI and Microsoft Agent Framework actually do, build the same app in several, and decide which to adopt, or none.
The curriculum is being prepared and reviewed. Enrollment is not open yet.
About this course
Framework choice shapes an LLM codebase for years, yet most comparisons are feature checklists. This course compares frameworks by building with them. Each module looks at a framework's core abstractions and runtime model: LangChain 1.x's create_agent and middleware on the LangGraph runtime, LangGraph's state graphs, checkpointers and interrupts; LlamaIndex's data connectors, indexes, query engines and event-driven Workflows; Pydantic AI's typed agents and dependency injection; Microsoft Agent Framework, the successor to Semantic Kernel and AutoGen; DSPy's signatures and optimizers; Haystack pipelines; CrewAI's crews and flows; and Mastra for TypeScript teams.
The last module turns comparison into a decision: lock-in and abstraction cost, observability and debugging, upgrade churn, and when a thin in-house layer over vendor SDKs is the better answer. Agent architecture and MCP are covered in bld-320, which compares vendor agent SDKs; this course concentrates on general-purpose application frameworks and their trade-offs. Version details are stated as of October 2026. The course ends in an automated exam; passing earns a Wirewalk Institute of AI certificate of completion and assessment.
What you will be able to do
Explain the core abstractions and execution model of LangChain, LangGraph, LlamaIndex, Pydantic AI, Microsoft Agent Framework, DSPy, Haystack and CrewAI
Build the same tool-using, document-grounded assistant in at least three frameworks and compare code size, control and debuggability
Use LangGraph checkpointing and interrupts to add persistence and human approval to a workflow
Use LlamaIndex Workflows and query engines to orchestrate data-heavy retrieval steps
Optimize a prompt pipeline with DSPy against a metric and explain when compiled prompts beat hand-written ones
Assess a framework's lock-in, observability, release churn and license before adopting it
Write a framework decision record and a migration plan from a legacy chain-based codebase
Syllabus
6 modules · 17 lessons · about 14 hours
1
Module 1: What Frameworks Are For
The problems frameworks solve, the costs they add, and the reference app used throughout.2 lessons
Framework, SDK or Plain Code: The Abstraction Trade-off
The Reference Application: Spec, Test Set and Scoring
2
Module 2: LangChain and LangGraph
The largest ecosystem and its graph runtime.3 lessons
LangChain 1.x: create_agent, Middleware and Integrations
LangGraph: State Graphs, Checkpointers and Interrupts
Migrating from Legacy Chains and LCEL Code
3
Module 3: LlamaIndex
A data-first framework for retrieval-heavy applications.3 lessons
Connectors, Indexes and Query Engines
LlamaIndex Workflows: Events, Steps and State
LlamaParse and Document-Heavy Pipelines
4
Module 4: Typed and Enterprise Frameworks
Type-safe Python agents and Microsoft's unified framework.3 lessons
Pydantic AI: Typed Agents, Dependencies and Structured Results
Microsoft Agent Framework: From Semantic Kernel and AutoGen to One SDK
Haystack Pipelines for Production Search Applications
5
Module 5: Programmatic Prompting and Multi-Agent Frameworks
Optimizing prompts as programs, and role-based agent teams.3 lessons
DSPy: Signatures, Modules and Optimizers
CrewAI: Crews, Flows and When Role-Play Helps
Mastra and the TypeScript Framework Landscape
6
Module 6: Choosing, Operating and Leaving a Framework
Turning the comparison into a decision you can defend and reverse.3 lessons
Observability Hooks: LangSmith, Langfuse and OpenTelemetry Instrumentation
Release Churn, Licenses and Lock-In Assessment
Writing a Framework Decision Record and an Exit Plan
Exam and certificate
The exam unlocks once you have completed every lesson. It draws 20 questions from a larger bank, spread across the modules, with the answer options shuffled for each attempt. You have 40 minutes and need 80% to pass. After a failed attempt you can try again after 24 hours, up to 3 attempts in total. Feedback is given per module, never as the answer key.
Passing earns a certificate of completion and assessment from the Wirewalk Institute of AI with a unique ID that anyone can check at institute.wirewalk.com/verify. It is not an accredited qualification or a vendor certification. How certification works.
The Wirewalk Institute of AI is independent. This course teaches publicly documented products and is not affiliated with or endorsed by LangChain, LlamaIndex, Pydantic, DSPy, deepset Haystack, CrewAI, Microsoft, or Mastra. Product and company names are trademarks of their respective owners.
A hands-on course on building no-code assistants with the builders already in your workplace tools: ChatGPT projects, custom GPTs and their successor plugins, Claude Projects and Skills, Gemini Gems, and Microsoft 365 Copilot Agent Builder and Copilot Studio. You leave with a tested, documented assistant and a plan to share it safely.
A hands-on, code-first course on the three major LLM APIs side by side: messages, streaming, structured outputs, tool calling, vision, cost, caching, batch, retries, abstraction layers, cloud routes and testing. You finish by building and hardening a small multi-vendor service.
A practitioner's course on retrieval-augmented generation: when to use it, how to ingest, chunk, embed and index documents, how to combine BM25 and vector search with reranking, and how to measure, debug and operate the result. Every module includes working Python sketches.