Agentic Data Architecture

About this book

Second Edition, August 2026.

Enterprise AI is moving beyond answering questions. Agents now retrieve data, interpret business meaning, coordinate with other agents and take actions across enterprise systems. That shift changes what the data architecture underneath them must provide.

A warehouse designed for dashboards is not automatically ready for autonomous workflows. An agent needs governed meaning, authoritative context, traceable lineage, controlled memory, secure tools, explicit permissions and evidence that explains what it retrieved, decided and changed.

Rather than treating RAG, vector databases, semantic layers, MCP, multi-agent systems and AI governance as separate technologies, the book connects them into one enterprise architecture, following Meridian Retail Bank from conventional analytics infrastructure to a platform that can support governed AI agents.

The book shows how to

• redesign enterprise data platforms for non-human users and AI agents
• make metadata, lineage and data contracts part of runtime reasoning
• use the semantic layer as the governed interface between business language and enterprise data
• design long-term memory, vector retrieval and document ingestion without creating a shadow data estate
• build secure agent tooling with MCP, delegated identity, short-lived credentials, policy boundaries and controlled write-back
• combine structured data and documents through Agentic RAG
• coordinate multi-agent workflows with A2A and enterprise handoff contracts
• prevent permission laundering, uncontrolled delegation and endless agent loops
• create audit-ready evidence packets linking requests, sources, policy decisions, tool calls, approvals and actions
• align agentic systems with ISO/IEC 42001, the EU AI Act, DORA and agentic security guidance
• plan a practical 90-day path to an agent-ready enterprise data platform

New in the Second Edition

Deeper coverage of secure agent tooling, delegated identity, execution receipts, idempotency and recovery; a document-ingestion layer before RAG; updated A2A and MCP treatment; and a complete worked-artefact appendix with a Data Contract, Semantic Object, Tool Definition, Policy Rule, A2A Handoff Packet, Evidence Packet and Golden Evaluation Case.

Who it’s for

Enterprise, data and AI architects, data platform and engineering leaders, governance professionals and practitioners moving AI agents from demos into dependable enterprise systems.