Every conversation about AI in the enterprise eventually arrives at the same place: agents. Autonomous systems that can reason, plan and act on behalf of an organisation without constant human direction. The capability is real and the commercial implications are significant.
But there is a pattern emerging that deserves attention. Organisations are deploying AI agents on top of the same fragmented, inconsistent and poorly integrated data estates that made their previous technology investments underperform. The agent is sophisticated. The foundation it operates on is not. The result is sophisticated AI producing unreliable outputs, because the data it is reasoning over is incomplete, stale or contradicted by a system it cannot see.
This is not an AI problem. It is an infrastructure problem. AI agents require clean, consistent and well-governed data to reason over. They require integration with the systems they are supposed to act within. They require audit trails, governance frameworks and the kind of data architecture that allows their outputs to be verified and their actions to be traced.
The organisations that will extract sustained value from AI are not the ones that deploy agents first. They are the ones that build the infrastructure first, the data foundation, the integration layer, the governance architecture, and then deploy AI into a system that is ready for it.
The sequencing matters. Infrastructure before agents. Foundation before capability. That is not caution. That is engineering.