Skip to content
Ayan Putatunda
All writing
August 25, 2026Data platforms1 min read

What AI-ready data actually means

A maturity model I use with teams: from ground data to platforms an agent can be trusted with. Most teams are further behind on governance than on retrieval.

"AI-ready" gets used to mean "we have a vector database". That is the easy part. When I assess a platform I score it on four things, in order, because each depends on the one before it.

1. Ground data

Ingestion is reliable, late data is handled, and someone gets paged when it breaks. If your Fivetran connectors silently stall for a day, no agent built on top will notice either.

2. Modelled data

There is a layer (dbt for most of us) where raw tables become entities with names a person in the business would recognise. Customer, account, subscription. One definition each.

3. Governed context

This is where most teams stop and it is where agents actually need help. Metric definitions, semantic views, lineage, and domain notes live in version control with owners and review. When two agents ask "what is active revenue", they get one answer because there is only one place the answer lives.

4. Agent access

Only now do you expose the vocabulary to models: semantic views behind Cortex Analyst, or an MCP server that serves governed context to Claude Code and your own agents. Retrieval is a solved problem. Deciding what is allowed to be retrieved is not.

The test

Ask an agent a question you already know the answer to. Then ask it to show you where the definition it used came from. If it cannot point at a file with an owner, you are at level two, whatever the vector database says.