Enterprise Architecture in an AI World

There is a quiet assumption in some quarters that AI will route around the messiness of enterprise systems — that a capable enough model makes underlying architecture matter less. The opposite is closer to the truth. AI inherits whatever it is built on, and it inherits it faster and at greater scale than anything before.

An AI capability sits atop data whose meaning must be clear, systems whose behaviour must be predictable, and boundaries that decide what it may and may not touch. Where those foundations are sound, AI compounds their value. Where they are confused, AI industrialises the confusion — producing wrong answers with impressive confidence.

Architecture as the boundary of trust

Enterprise architecture in this context is less about diagrams and more about deciding, deliberately, where trust begins and ends: which data is authoritative, which systems are systems of record, and where a model's output may flow. Those decisions are what let an organisation adopt AI without losing control of it.

This does not call for more architecture for its own sake — I remain wary of complexity dressed up as rigour. It calls for the right architecture: enough structure to make AI dependable, and no more. As data and AI become central to how institutions operate, that discipline stops being a technical nicety and becomes a condition of trust. It pairs naturally with strong AI governance.