Enterprises deploying AI agents across business functions face a new systems problem: managing a portfolio that includes agents built in-house, composed from open frameworks, and bought from commercial platforms. This report offers practical guidance for structuring and governing a heterogeneous agent portfolio. It addresses build-time architectural decisions like zoning, design patterns, and interoperability and runtime considerations like lifecycle oversight, telemetry, and policy enforcement to ensure consistency, scalability, and control across agent types and environments.