Multi-agent systems built on MCP are rapidly moving from prototypes into production and the ecosystem is discovering that the hardest problems are not in the agent logic, but in the infrastructure surrounding it. This talk directly addresses one of the most common gaps practitioners report: how do you govern traffic between agents, models, and MCP servers at scale? We'll walk through a live demo built around a fictional company, NexusWork, shipping a new release of their project management product. Their multi-agent system processes customer feedback arriving from global social media channels, correlates it with GitHub issue data, and autonomously takes action: filing pull requests and generating region-specific sales briefs all without human intervention. What makes this demo instructive isn't the happy path. It's what's happening underneath: every LLM call, every MCP tool invocation, every agent-to-agent handoff traverses a governed data plane. We'll show concretely how observability, authentication, rate limiting, prompt injection defense, and multi-provider routing are enforced at the infrastructure layer not bolted on per-agent.
Juhi is the Developer Evangelist (APAC) at Kong. She’s passionate about helping the developer community and building strong connections. Outside of work, Juhi enjoys reading, staying fit, trekking, traveling, and cooking. She describes herself as a "product of a community" with many hobbies and passions that she loves to explore, even though keeping up with them can be a challeng
