AI is fundamentally reshaping how software systems are engineered, and data is no exception. High-quality, fit-for-purpose data has become essential for reliable AI and analytics. Without structured and trustworthy data products, modern AI agents struggle to make sound decisions or perform complex tasks. Yet many organisations still rely on slow, manual data engineering practices that cannot keep pace with these demands.

This session illustrates how AI accelerates the data product lifecycle—from specification and design through implementation, deployment, and governance—while preserving architectural discipline and delivering practical, reusable outcomes.

It begins with data product specification. Capturing domain context, aligning definitions, and enforcing consistency typically require significant manual effort. We’ve evolved a technique that uses AI-assistance to collaboratively design data products with domain experts, applying software engineering practices such as working backwards from use cases and “jobs to be done.” This produces machine-readable data product specifications and contracts that improve consistency, reduce expert effort, and embed standards alignment.

The session then addresses legacy ETL debt. We’ve evolved techniques to use AI to reverse-engineer legacy ETL code into explicit data product specifications, surfacing intent and contracts hidden in existing pipelines. These specifications drive forward engineering, enabling teams to migrate by rebuilding data products and eliminating debt rather than carrying it forward.

With machine-readable specifications in place, AI agents generate production-ready transformation code, tests, and data quality checks following modern architectural patterns such as medallion architecture. A lightweight paved-road platform enables one-click deployment to target environments, including permissions configuration, catalog registration, and agent-friendly output ports.

Finally, the session explores governance at scale. In decentralised data ecosystems, governance only works when automated. AI-driven fitness functions continuously assess data product metadata against standards such as FAIR and DATSIS. A lightweight dashboard provides clear visibility into compliance without slowing delivery.

This session is aimed at technical and non-technical data and architecture leaders navigating modern data ecosystems. Attendees leave with concrete patterns and architectural insights for applying AI to data product design, delivery, and governance—enabling trustworthy, scalable data foundations for AI-driven systems.