Description

I’ve been spending a lot of time thinking about why multi-agent systems feel so promising but so fragile in practice. In this talk I break down how I approach it using A2A for agent collaboration and MCP for standardized context. The focus is on keeping intent intact as work moves between agents, and on building the guardrails that prevent the system from devolving into a game of telephone.

You’ll see how agents can discover each other’s capabilities, coordinate across platforms, and still stay observable and secure. I also walk through a concrete serverless implementation using AWS Lambda, Amazon Bedrock, and Momento Cache to show how these ideas scale in production.