What you are looking at
An AI agent is a program that reads a request, decides which tools to use (for example "look up an order"), uses them, and then writes an answer. Companies build agents with different frameworks. Here, the same agent job is built eight ways (LangGraph, Google ADK, OpenAI Agents SDK, Microsoft Agent Framework, AutoGen, Semantic Kernel, CrewAI, and one that fetches its tools over the MCP protocol). They all give the same answer, which proves they are interchangeable behind one interface.
- Choose an example question.
- Choose a framework (or run all of them).
- The flow diagram shows the steps: the question, the model planning, the tool calls, and the final answer.
1. Pick a question
2. Pick a framework
3. Result
Comparison
| Framework | Answer | Tool calls | Time |
|---|
What is real and what is simulated
- Real: each framework's own agent loop, tool wiring, and the MCP and A2A protocol servers.
- Simulated: the language model is a rule-based stand-in so the demo is free and repeatable. The same code can talk to a real model by changing one setting.
- This page is a recorded replay. Each result shown was produced by actually running that framework (recordings are stored in
recordings.json). Free-text questions need a running server; use the four examples here, or run the Docker image locally.