The Building Blocks of Production AI Agents
Every organization moving from AI pilots to production hits the same three walls:
- Token spend they can't see or control: usage is opaque, budgets are aspirational, and finance finds out last
- Ungoverned MCP sprawl: servers multiplying across teams with no registry, no policy, no identity
- A platform that wasn't built for agents: Kubernetes is where the workloads live, but it was never designed to run agents as first-class citizens
These aren't independent problems. They're the symptom of treating agents like just another workload on infrastructure built for something else.
Solving the challenges unlocks the operating model AI leaders actually want: ship agents faster, run them safer, and give platform teams a way to govern without putting friction in front of developers. Solo partnered with Google on Agent Substrate to make Kubernetes a first-class agentic platform, and we built agentgateway, agentregistry, and kagent as the build-and-run stack on top of it. Purpose-built around MCP and A2A, not retrofitted onto yesterday's gateways.
This workshop walks through how the pieces fit together. We'll cover:
- Why token spend visibility is an infrastructure problem, not a billing problem, and how to enforce budgets at the data plane
- How to bring MCP sprawl under governance without slowing down the teams shipping agents
- How Solo and Google made Kubernetes a first-class agentic platform with Agent Substrate and kagent
- How agentgateway, agentregistry, and kagent compose into a complete build/run experience
- Field patterns from enterprises already running this stack in production
