Player not loading? Watch on YouTube
Meenakshi Kodati compares agent frameworks by the kind of system a developer wants to build. The explanation covers five designs and uses practical examples to show how their coordination needs differ.
For predictable workflows, a customer support example follows a sequence: search a knowledge base, draft a response and possibly create a ticket. Kodati recommends LangChain for sequential steps and LlamaIndex for retrieval and indexing, with LangGraph as an option for more complex setups.
Autonomous systems receive a goal and work out how to achieve it. A coding assistant illustrates this approach through planner, coder and reviewer agents. The speaker suggests AutoGen, experimental BabyAGI setups and CrewAI. Role-based systems give each agent explicit boundaries; a research, writing and editing example leads to CrewAI, structured AutoGen setups and ChatDev for software development.
For production orchestration, Kodati discusses API, database and business workflow integration. She presents Agent Framework as a combination of Semantic Kernel and AutoGen, and also recommends LangGraph. LangFlow and Flowise cover visual prototyping through connected canvas components.
