AI Agents In-Depth – Function Calling, MCP and Tool Use Under the Hood
summary
Shows how GPT-4o function calling and SDKs (OpenAI Foundry, LangChain, Semantic Kernel, Agent Framework) actually drive tool selection, not tool execution, and demos concrete failures and fixes: a pizza agent orders (resulting examples: 12 large vegetarian, 10 pepperoni, 8 extra-cheese in one run), a RAG pipeline drops ~18,000 input tokens down to ~1,435 by summarization, a calculator tool once returns 14+16=31, a QR code generator fails until a code-interpreter runs Python, and an MCP server on 127.0.0.1:8000 shows distributed tool/notification flows.
AI-written summary · about
chapters (22)
- 0:00Introduction to AI agents
- 2:47Tool use and function calling
- 7:31Fine-tuning for tool calling
- 8:54Frameworks for AI agents
- 11:19Pizza ordering agent demo
- 15:33Vibe coding with file tools
- 20:58Multi-agent systems
- 21:19Retrieval Augmented Generation with tools
- 25:28RAG demo with Wikipedia
- 30:24Challenges in tool use
- 37:59Security in agent apps
- 40:42Model Context Protocol intro
- 43:03What MCP is not
- 44:26MCP server capabilities
- 45:30MCP vs function calling
- 45:51MCP music playback demo
- 49:27Multi-agent project manager demo
- 52:22MCP in VS Code
- 54:05GitHub Copilot customization
- 55:31Skills vs agents in VS Code
- 56:11Vibe coded mashup editor demo
- 59:45Summary
related talks
01h 28mAI Coding Agents and how to code them
45mAgents, Tools, and MCP, oh my! Next-level AI concepts for developers
01h 03mAI Coding Agents and how to code them
48mAI Coding Agents and how to code them
48mAI Coding Agents and How to Code Them
37mRAG, Tooling, MCP, Agents ... and What is Missing?
40mAgents in Action: Advancing Open Source AI Missions
34mAI Agents: Transforming Autonomous Decision-Making