MCP has become the de facto standard for connecting AI agents to tools, but protocol adoption alone doesn't make an enterprise architecture AI-ready. The real bottleneck is data readiness: without a structured, schema-driven data layer that encodes entity relationships, enforces field-level access control, and provides a governed, discoverable model, agents can invoke tools but cannot reason about relationships or respect constraints. The author argues for a layered architecture where a federated GraphQL graph serves as the data plane and MCP/A2A protocols sit on top as the coordination plane. Enterprises that invest heavily in coordination protocols before structuring their data plane risk building a well-connected but ungoverned system. Real-world examples from SAP and Microsoft Fabric show this pattern converging in production.
Table of contents
Coordination vs. Data Access: Two Problems Sold as OneEnterprise AI Agent Governance: The Real BottleneckThe Strongest Counterargument — And Why It Falls ShortWhat a Structured Data Layer Provides That a Protocol Does NotMost Agent Workloads Are Data Access, Not CoordinationWhere Protocols Remain EssentialThe Layered AI Agent Architecture: Data Plane + Coordination PlaneThe Strategic QuestionThe AI Agent Architecture Conversation That Should Be HappeningFrequently Asked Questions (FAQ)Sort: