About

Data architecture has always been about control — making messy data usable through standardisation and transformation. But AI models now thrive on the very messiness we’ve spent years taming. In this provocative discussion, leaders from Lloyds Banking Group, Mars, and Google debate whether our obsession with structure is holding us back. Is it time to replace rigid ETL workflows with adaptive, AI-native data platforms that learn context rather than enforce it?

• AI shifts the value from structure to context.
  Traditional ETL and modelling pipelines were built to impose order on data; AI can now interpret meaning from disorder. The future of data engineering is less about cleaning and more about capturing rich, contextual signals.
• The modern data platform isn’t dead — it’s evolving.
  Instead of dismantling what we’ve built, AI invites a rethink of architecture: blending structured and unstructured layers, integrating vector stores, and enabling semantic search and retrieval-augmented pipelines.
• Governance, trust, and lineage still matter — maybe more than ever.
  As AI consumes raw data, we need new ways to measure quality, provenance, and bias. “Unstructured” doesn’t mean “ungoverned.”
• Data teams must evolve from pipeline builders to data product designers.
  The role of data architecture expands beyond moving data — to designing experiences that make it discoverable, interpretable, and safe for AI-driven systems.