AI Agents In-Depth – Function Calling, MCP and Tool Use Under the Hood - Alan Smith - NDC AI 2026

Value of the Information & Strength of the Argument

The talk provides valuable insights into the practical implementation of AI agents, with a focus on function calling and tool use. The argumentation is solid, based on live demonstrations and clear explanations of the underlying concepts. The speaker effectively illustrates how models infer tool usage, the role of system prompts, and the importance of tool definitions. He also addresses common pitfalls such as non-deterministic behavior and the need for careful prompt engineering. The value lies in the practical, hands-on approach, making complex concepts accessible to developers.

Scientific Rigor, Source Quality, Title Accuracy

The talk demonstrates scientific rigor through its structured explanation and live demos. However, it lacks formal citations or references to academic sources, relying instead on practical experience and official documentation. The title accurately reflects the content, which is a technical tutorial on AI agents. The speaker’s credibility is enhanced by his 30 years of coding experience and MVP status. The content is well-organized and technically accurate, but the absence of external references limits its scholarly depth.

The title accurately reflects the content, which covers function calling, MCP, and tool use in AI agents.

8/10

The talk is a technical tutorial by an experienced developer (MVP) with live demos and clear explanations. The content is accurate and well-structured, but it lacks formal citations and peer-reviewed sources, relying on practical experience and official documentation.

Markers derived by PSI from the transcript: the creator did not define chapters.

• 02:00Introduction to the session and overview of topics: function calling, MCP, and tool use.
• 05:00Explanation of how LLMs work and the concept of tool calling.
• 10:00Demo of function calling with GPT-4o in Postman, showing JSON requests and responses.
• 15:00Discussion on model fine-tuning for function calling and user approval flows.
• 20:00Comparison of frameworks: LangChain, Semantic Kernel, and Agent Framework.
• 25:00Demo of a pizza ordering agent, showing tool inference and multi-language support.
• 30:00Demo of a vibe coding assistant that creates a website using file tools.
• 35:00Explanation of agentic RAG and how search services are used as tools.
• 40:00Demo of agentic RAG with Wikipedia, comparing internal knowledge vs. external search.
• 45:00Conclusion and summary of key takeaways.

• NDC AI Conference — The talk was recorded at NDC AI in Oslo, Norway.
• NDC Conferences — Information about upcoming NDC conferences.

• OpenAI Function Calling Documentation — Official documentation on function calling, which aligns with the talk's content.
• Model Context Protocol (MCP) Official Site — Official site for MCP, a protocol mentioned in the talk.