Beyond Vibe Coding: Building Multi-Agent Systems with MCP & A2A

Everyone's talking about AI agents. Frameworks are multiplying. Demos are everywhere. But how do these systems actually work under the hood? Val Andrei Fajardo dismantles the magic by building a multi-agent system from first principles. No framework dependency, no black boxes. Just the core mechanics that every agent framework is built on, exposed and explained.

What you'll learn:

• The foundational agent building blocks: how an LLM + tools + a reasoning loop becomes the basic unit of every agent framework
• MCP (Model Context Protocol) in practice: how agents discover and use external tools, turning isolated agents into composed systems
• Skills: Composable, reusable behaviours that bundle related agent capabilities (think SOPs). While business users are buzzing about Claude Cowork workflows, we’ll explore the engineering behind that same instinct by building the underlying "Skill" architecture from scratch.
• A2A (Agent-to-Agent protocol): how multiple agents collaborate, delegate, and coordinate work across service boundaries
• Why understanding the internals makes you better at using ANY production framework (Google ADK, LangGraph, Anthropic Agent SDK, or your own)

Why this matters if you're an enterprise or solutions architect:
The flow shapes in modern agent systems (sequential chains, parallel fan-out, routing, orchestration) are not new. What's new is that every processing node now makes probabilistic decisions instead of deterministic transformations. Understanding what's inside each node is how you debug, optimize, and trust these systems when they hit production.

About the book
Andrei's "Build a Multi-Agent System (from Scratch) - With MCP and A2A" is available through Manning's Early Access Program. The book helps you deeply understand how agent frameworks work by having you build one yourself, from scratch. All code lives in the book's companion library: masfromscratch.com. Ebook coupons will be raffled at the event.

About the speaker
Val Andrei Fajardo is an ML/AI Researcher and Engineer based in Canada. As a founding engineer at LlamaIndex (one of the most widely adopted agent/RAG frameworks), he built core components used by millions of developers monthly. He holds a PhD in Statistics from the University of Waterloo and has conducted research at the Vector Institute. He is currently writing "Build a Multi-Agent System (from Scratch)" for Manning Publications.

Serverless Toronto has been bridging the gap between IT and business needs since 2018. Past sessions: youtube.serverlesstoronto.org