The chairman and CEO of Everpure said today it’s now only a matter of time before IT architects will need to revisit how data is stored across the enterprise to meet the challenges of the artificial intelligence (AI) era.

Charles Giancarlo said, historically, multiple copies of the same data sets have been stored in different systems by the applications used to create them. The challenge is that approach creates different versions of the same data that winds up being strewn across the enterprise. Whenever an AI agent encounters those conflicting data sets, the chances the output it generates will be flawed only increase, said Giancarlo.

Resolving that issue will require IT architects to create a data plane through which data becomes accessible to any application or AI agent versus continuing to allow every application to store data in a way that cannot be easily accessed by what will soon be thousands of AI agents that organizations are deploying, noted Giancarlo. “The relationship between data and the application needs to be inverted,” he says.

Formerly known as Pure Storage, Everpure is making a case to provide an integrated platform that enables IT teams to centrally manage data as a service. That approach makes it feasible to rationalize existing platforms while at the same time reducing total costs.

The overall goal is to streamline IT operations by defining a smaller number of systems of record that become the single source of truth for agentic AI workflows, said Giancarlo. In effect, as more backend applications become headless services, the management of data and storage will need to become more integrated, he added.

It’s not clear how quickly IT teams will be moving in that direction, but as data becomes more accessible, there is likely to be less reliance on some providers of software-as-a-service (SaaS) applications that organizations might replace by using AI to create a custom equivalent that more narrowly addresses tasks and functions they would like to automate. Achieving that goal, however, will require more control over how data is stored and managed to make it more self-describing, said Giancarlo.

Additionally, IT teams will need to deploy routers that make it possible to shift requests to process prompts between multiple AI models as it becomes apparent that at any given time one of them might be better suited to handle one task versus another, he added. In many instances, those AI models are likely to be lower-cost, open-weight alternatives that are capable of handling the tasks that the average enterprise is looking to automate, noted Giancarlo.

Exactly how long it might take to isolate data and the schemas used to describe it from applications is at this point difficult to assess. One thing that is certain is that, in time, the data silos that characterize most IT environments will need to be greatly reduced, if for no other reason than to minimize the amount of IT resources that AI agents consume today to reason across massive amounts of duplicate and conflicting data.