The shape of Data Stores in the age of AI

For years , we designed data systems with one assumption .. “structure and model first”

Raw data came in .. ETL pipelines cleaned it .. schemas were defined .. warehouses were built .. then and only then could we analyse.

This made sense in a world where machines needed structure to understand anything. But that assumption is quietly breaking. AI is changing the direction of flow.

Today, models are remarkably good at creating structure from unstructured data. They can read documents , conversations , logs , images , reviews and extract meaning , summaries , tables .. even insights.

But the reverse is not true. Once data is compressed into rigid schemas , much of its richness is gone .. context disappears .. edge cases vanish .. nuance gets flattened.

This raises a fundamental question : Why are we still forcing structure too early?

Perhaps the future of data stores looks different.

Instead of storing structured data as the primary layer , we are better off storing raw and unstructured data as the source of truth .. namely documents , conversations , events , logs , media and all webpages etc.

.. then create structured views on demand , not fixed schemas used for dynamic interpretations. Imagine :

• ask a question .. get a table.
• ask differently .. get a summary.
• ask deeper .. get a chart.

Same data. Different shapes .. we are already seeing early signals.

• Swiggy’s chat experience doesn’t rely on rigid query interfaces. It interprets intent from conversation.
• Weather apps don’t just show raw data .. they translate it into contextual narratives: “feels like”, “likely rain”, “plan your day” etc.

The interface is no longer tied to the storage format .. “AI sits in between”

Another subtle shift is in how we treat pipelines. ETL systems were designed to clean data. But in doing so , they often discard information that doesn’t fit the schema such as :

• outliers.
• text fragments.
• irregular patterns.

In an AI-first storage world , these “imperfections” often carry vital signals. By preserving raw data , we allow intelligence to revisit , reinterpret and reframe information as models improve.

• data doesn’t age as quickly .. it evolves.
• structure doesn’t disappears. .. it becomes a layer, not the foundation.
• output becomes a view .. not just the truth.

In the age of AI-powered discovery , the shape of data stores is no longer fixed .. it is fluid. Because intelligence today is not limited by structure. It is defined by its ability to create structure when needed. And that may quietly be one of the biggest shifts in how we build systems going forward.

Until next time… happy thinking.

Author – Sumit Rajwade, Co-founder: mPrompto