References: Choosing an agent framework

Source material

Source curriculum (structural mirror, cited as further study):
• Microsoft, "Explore AI Agent Frameworks" (AI Agents for Beginners, Lesson 02)
  Author: Microsoft Cloud Advocates
  Lesson page: https://github.com/microsoft/ai-agents-for-beginners/tree/main/02-explore-agentic-frameworks
  License: MIT
• Berkeley CS294/194-196, Lecture 3 (Agentic frameworks), two talks:
  - Chi Wang, "Agentic AI Frameworks & AutoGen"
  - Jerry Liu, "Building a Multimodal Knowledge Assistant"
  Course page: https://rdi.berkeley.edu/llm-agents/f24
  Video: https://www.youtube.com/watch?v=OOdtmCMSOo4
  License: YouTube standard (link-out only)
Clawdemy's lessons are original prose that follows the pedagogical arc of these
sources. We do not reproduce or transcribe them; we cite them as recommended
companions. All rights to the original materials remain with their authors.

Read or watch this next

• Explore AI Agent Frameworks (Microsoft) by Microsoft Cloud Advocates. The practitioner version of this lesson. Its contribution to our framing is the fit-question pattern: it treats choosing between options as a “which fits this scenario” question, not a “which wins” question. MIT-licensed.
• Berkeley CS294 Lecture 3: Agentic AI Frameworks (Wang on AutoGen + Liu on LlamaIndex). Two consecutive talks in the same Berkeley class session: Chi Wang (creator of AutoGen) and Jerry Liu (creator of LlamaIndex), each presenting their own design. The fastest way to hear, firsthand, what each framework is built for.

The frameworks named here

Each project’s own documentation, as a primary source for what it is built for. Listed by category, not by preference.

• Microsoft Agent Framework (orchestration / multi-agent). Microsoft’s current framework for coordinating one or more agents and multi-agent conversations; positioned as the enterprise-ready successor to AutoGen.
• AutoGen (orchestration / multi-agent; maintenance mode). The predecessor to Microsoft Agent Framework. Per its own README, AutoGen is now in maintenance mode (no new features, community-managed), and new projects should start on MAF; kept here for reference and for projects already on it.
• LlamaIndex (retrieval-first). A data framework that treats retrieval over your own documents as a first-class primitive.
• LangGraph (graph / state-machine). A framework for expressing an agent’s control flow as an explicit graph of steps and transitions.

Adjacent topics

Where this leads inside this track.

• The tool-use design pattern in depth. The next lesson. This one was about choosing the machinery; the next goes underneath it, into how to define a single tool well so the model reliably knows when and how to use it.
• How tool use turns a model into an agent. The previous lesson. The four-step exchange every framework here automates. Worth re-reading if “what a framework removes” felt abstract.
• Multi-agent systems. Later in the track. The orchestration category named here gets its own lesson, on when splitting work across several agents actually beats one agent and what coordination costs.