Large language models (LLMs) increasingly power sophisticated agents that extract information, orchestrate tools, and solve tasks autonomously. Most AI talks focus on LLM reasoning or tool use, skipping the crucial orchestration layer that actually makes agentic systems reliable.

In this session, we dive into the trenches. We’ll introduce the MCP Protocol (Model-Control-Process), the orchestration engine driving agentic solutions, then walk through the architecture and key code patterns of a lightweight MCP Server implementation. You’ll see how the server routes reasoning, triggers tools, handles errors, and enforces guardrails. Along the way, we’ll cover real-world engineering trade-offs, common pitfalls, and best practices. By the end, you will not only understand why agentic AI works, but also how to implement it yourself, bridging the gap that most talks skip. If you’re responsible for creating autonomous assistants, orchestrated pipelines, or next-gen AI agents, this session gives you both the concept and the code to get started.