Agents105 · Module III · Lesson 10 of 13
Article · 12 min

The framework landscape

LangGraph, AutoGen, CrewAI, and the alternatives.

Summary

The current agent framework landscape: LangGraph (graph-based control flow), AutoGen (Microsoft, conversation-based), CrewAI (role-based teams), LlamaIndex agent workflows, and a growing set of low-code options (Flowise, n8n). Each has a personality; none is a clear winner.

Objectives
  • 01Name the four dominant frameworks and their positioning.
  • 02State when to skip frameworks entirely.
  • 03Recognize framework lock-in risks.
The Lesson

The options

LangGraph: explicit state graphs; production-oriented. AutoGen: multi-agent conversation; research-oriented. CrewAI: role-based agents; ergonomic for simple crews. LlamaIndex: agent workflows integrated with retrieval. Or: skip all of them and write ~200 lines of code, which is often the right choice.

Lock-in

Frameworks change fast, break APIs, and add abstractions you may not want. If your agent is simple, write it yourself. If it is complex enough to need a framework, pick one and expect to migrate within 18 months.

Key Ideas
  • No framework has clearly won; each has a personality.
  • For simple agents, no framework beats a framework.