.NET Rocks! Thinking Agentic AI with Seth Juarez
Jul 3, 2025
Seth Juarez, a machine learning expert from Microsoft, dives deep into the world of agentic AI. He explains the Model Context Protocol (MCP) and how it allows AI agents to collaborate more effectively. The discussion highlights the challenges of managing these advanced tools and emphasizes the need for clear governance. Listeners will enjoy insights into the evolution of AI, including the balance needed when working with user instructions in AI outputs, and the importance of narrowing tasks for better performance in real-world applications.
AI Snips
Chapters
Transcript
Episode notes
Basics of Agentic AI
- An agent is an LLM with the ability to choose, reason over data, and act based on that reasoning.
- This creates a new programming control structure where the agent decides which function to run next.
Narrow Tasks Reduce Hallucinations
- Restrict tool calls and narrowly define tasks to reduce LLM hallucinations and increase accuracy.
- Design prompts carefully to maximize the probability of correct outputs.
Use MCP to Connect Tools
- Use protocols like MCP to expose function calls as tools to LLMs for executing actions.
- Let the LLM generate function calls and handle responses to bridge AI reasoning and code execution.