The Verifiable Context Layer: The Missing Architectural Commitment in Enterprise Intelligence for Agentic AI

Enterprise AI initiatives continue to stall at the trust boundary despite the increasing maturity of foundation models and agent runtimes. The Enterprise Intelligence Layer (EIL) pattern -- a layer between enterprise systems of record and agentic AI workloads -- is converging across major enterprise vendors under varied branding: Atlan's Context Layer, Snowflake's Agent Context Layer, Microsoft Foundry IQ together with Purview, Salesforce Einstein Trust Layer, IBM watsonx.governance, and Palantir Foundry Ontology. Across these implementations, one architectural commitment is consistently absent or under-specified: an explicit, named layer that turns enterprise context into a substrate for agentic verification. This paper proposes the Verifiable Context Layer (VCL) as a vendor-independent reference architecture for that missing component. The VCL is a named architectural commitment, not a new mechanism. It organises five existing elements -- a semantic layer, a knowledge graph, a governance and policy engine, an agent and tool runtime, and a feedback loop -- around a single load-bearing claim: that enterprise context, properly governed and instrumented, is the substrate that makes agentic AI verifiable at enterprise scale, both against formal verification primitives drawn from the recent literature (VeriGuard, AgentSpec, AgentGuard, MAIF, AAGATE) and against the regulatory obligations imposed by the EU AI Act and the NIST AI RMF Agentic Profile. The paper makes four contributions. First, it names and formalises the VCL as the missing architectural commitment within the emerging EIL pattern, and positions it explicitly against six leading commercial implementations through a fifteen-dimension comparison matrix. Second, it specifies five components with explicit interfaces, illustrated by a worked enterprise use case -- a supplier-contract query traced through every component to a regulator-addressable audit trail. Third, it maps each component to specific EU AI Act articles (9, 10, 12, 13, 14, 15, 17), to NIST AI RMF Agentic Profile functions, and to formal verification primitives from the recent literature. Fourth, it synthesises publicly attributed industry observations from Klarna, Walmart, JPMorgan, Goldman Sachs, Mercado Libre, Siemens, Schneider Electric, Toyota, Notion, and Box, and proposes a phased empirical roadmap with three controlled pilot studies and concrete integration guidance for practitioners. The paper does not claim primary empirical validation. The contribution is a named architectural pattern defended by structural argument, illustrated by a worked example, and grounded in public evidence. The author argues that the regulatory deadlines now in force -- the EU AI Act enters force for high-risk systems on 2 August 2026 -- make naming, implementing, and empirically validating this pattern an urgent priority for the research and practitioner communities.