The enterprise data architect in the AI era

For decades the enterprise data architect made data available, consistent, integrated and governed. That job hasn't gone away. But now that AI systems read, reason over and act on enterprise data, the architect's question has changed from "is the data there?" to "can an intelligent system use it safely and correctly?"

The classic role

The enterprise data architect has always been the person who sees the whole picture: how information is represented, where it lives, how it moves between systems, and who is allowed to use it. Typical responsibilities include:

• Defining enterprise data models and shared business definitions.
• Designing integration patterns between operational systems and analytics.
• Setting data standards, principles and architecture guardrails.
• Making sure platforms and projects fit the business and technology strategy.

All of this still matters. AI raises the stakes on every one of them.

What changed with AI

Reports and dashboards are read by people, who can notice when a number looks wrong. AI assistants and agents read data at machine speed, combine it with other sources, and answer in confident natural language. Agents may even take actions. So the architect now has to ask:

• Is this information trustworthy?
• Is its business meaning understood and written down?
• Can it be accessed securely, by the right system, for the right user?
• Is its context preserved when it's retrieved?
• Can the resulting decisions and actions be monitored and audited?

In other words, the job moves from data availability to information usability, meaning, trust and intelligence.

The model is only one component

Take a customer-service assistant. The language model can understand questions and write answers, but it is only useful if the architecture around it connects:

1. Customer information the user is authorized to see.
2. Product and service definitions.
3. Business policies and knowledge.
4. Data platforms and integration services.
5. Retrieval and semantic capabilities.
6. Security, authorization and audit.
7. Human oversight and the business workflow.

That is an architecture problem, not a model problem. It's why the enterprise data architect sits at the center of AI transformation.

The expanded skill set

The modern role reaches into areas that used to belong to other disciplines:

• Cloud platform and lakehouse architecture.
• Data product design and domain ownership.
• Semantic models, knowledge graphs and enterprise context.
• AI data readiness, retrieval and context architecture.
• AI governance, security and responsible use.

Nobody needs to be the implementation expert in all of these. The architect's job is to understand how they relate, and to set the principles, boundaries and decisions that let them work together.

A way to think about it: EDA-IQ

In my book I use a framework called EDA-IQ (Enterprise Data Architecture for Intelligence and Quality). It has five connected dimensions, each depending on the ones below it:

1. Trusted data foundation: quality, lineage, security and governance.
2. Information and data product architecture: clear models, ownership and shared meaning.
3. Data platform architecture: the cloud, lakehouse and integration platforms that store and move data.
4. Semantic and knowledge architecture: the business context that makes data understandable to people and machines.
5. Intelligence architecture: analytics, AI and agents that turn information into decisions and action.

The point of the layers is simple: you can't build reliable intelligence on data nobody trusts or understands.

Where to start

• If you're an architect today: pick one AI use case and trace every piece of data it needs back to its source. The gaps you find are your roadmap.
• If you're a senior engineer: start owning decisions across systems, not just inside one pipeline. Write them down.
• If you lead a team: treat data readiness as part of every AI project from day one, not a clean-up job afterwards.

An intelligent enterprise isn't created by adding AI to existing data platforms. It's architected, deliberately, from trusted information up.