May 11
Loading video player...
I built the same AI agent in CrewAI, LangGraph, LlamaIndex, Pydantic AI, and Microsoft Agent Framework to compare them side by side. In this video, you’ll see how each one handles: - setup - tool calling - RAG - MCP - execution - runtime - token usage and cost To make it fair, each framework runs in its own isolated container with its own dependencies. They all connect to the same shared MCP server and the same comparison UI, so everything is tested under the same conditions. Our demo agent is a flight assistant simulator that can check the weather, look up refund rules with RAG, and use MCP booking actions. We'll also look at how long each framework takes to run and how many tokens it uses, so you can get a clearer sense of both performance and potential cost. -- Timeline 00:00 Intro and what this comparison covers 00:07 Test setup and isolated container architecture 00:35 The 5 frameworks in this video 00:53 Demo agent overview and what it can do 01:14 Agent setup in each framework 03:00 Tool calling comparison 04:59 How RAG is handled across frameworks 07:33 MCP integration in all 5 frameworks 11:45 Execution flow and how each agent runs 13:11 Side by side runtime, token usage, and cost comparison 14:56 First test run 16:54 Second test run 18:39 Final behavior check and wrap up Frameworks covered: CrewAI LangGraph LlamaIndex Pydantic AI Microsoft Agent Framework If you are trying to decide which AI agent framework to learn or use in 2026, this should give you a practical side by side look at how they actually feel in code, how long they take to run, and what they may cost in tokens.