.NET Rocks! Thinking Agentic AI with Seth Juarez
14 snips
Jul 3, 2025 Seth Juarez, an AI expert from Microsoft, breaks down the intriguing concept of agentic AI, sharing how it can revolutionize development. He discusses the power of tools like MCP and NLWeb for inter-agent collaboration while addressing the challenges of managing these systems. The conversation dives into the technical underpinnings of agentic AI, touching on the limitations and the necessity for proper governance. They also explore the innovative ways prompts can influence AI responses, showcasing the evolving landscape of coding with AI.
AI Snips
Chapters
Transcript
Episode notes
Understanding AI Agents
- An AI agent is essentially an LLM that chooses actions based on reasoning and outputs structured commands to execute tasks.
- This transforms language into a giant decision structure, much like an expansive switch statement in programming.
Reasoning Happens in LLMs Only
- AI reasoning is performed exclusively by the LLM; the protocol (MCP) only enables calling functions.
- Reasoning and control happen in prompts and outputs, not in the protocol layer itself.
Control AI for Safety
- Restrict agent tool calls and carefully write response handling code to ensure trustworthy outputs.
- This controlled approach confines AI agents to safe actions and minimizes errors.