Mofesola Babalola & Hannah Olukoye

When rolling out agentic AI and Model Context Protocol (MCP) servers across an enterprise, a natural tension emerges. Engineering Managers want velocity. They push to integrate multi-agent swarms with internal APIs to accelerate feature delivery. Meanwhile, Infrastructure Engineers are staring down an operational nightmare. Rogue agents hammer internal databases, token spend becomes unpredictable, and massive new attack vectors bypass traditional network perimeters. In this dual-perspective session, an Engineering Manager and a Lead Infrastructure Engineer will debate and dissect the "Day 2" realities of scaling agentic systems. We will show how to resolve the velocity versus control paradox by building an Agentic Paved Road powered by Istio Ambient Mesh. Together, we will demonstrate how to leverage Istio and eBPF to enforce zero-trust security on agent-to-agent traffic without the dreaded sidecar tax that starves GPU nodes of memory. The Engineering Manager will cover managing agentic debt, tracking ROI, and developer enablement. The Infrastructure Engineer will dive into the technical implementation. We will show how to deploy node-level ztunnels and L7 Waypoint proxies alongside the new Gateway API Inference Extension. You will learn how to rate-limit, audit, and secure MCP tool calls entirely at the platform layer, ensuring intelligent LLM inference routing without slowing down the development teams.