Is Technical Debt a Barrier to AI App Deployment?

Nearly every IT leader today is in the midst of moving the next generation of AI apps from the design phase into deployment, and they are finding that they must grapple with the problems that arise when those apps are dependent on legacy data or infrastructure. And according to the attendees of our CIO Roundtables, there is no one-size-fits-all answer. Each case of legacy dependence must be evaluated separately.

Once IT leaders have evaluated how specific AI projects are impacted by technical debt, they then check the organization’s existing plan for addressing technical debt, possibly choosing to accelerate the parts of that plan that will best help meet the goals of the specific AI project.

Particularly tricky are AI apps that are dependent on resources that are trapped by technical debt, usually because data is stuck in a system with substantial issues. Er There are two common problems.  In some cases, it is not possible to extract the data from the legacy environment in a way that will support the goals and functionality of the AI app. That might require a rebuild or a total scrapping of the app, with a new design needed to replace it. That’s expensive and time-consuming, but there might be no other option. The second scenario occurs when the new app can get at the data, but the data cannot be delivered at the speed necessary to support a real-time or somewhat real-time AI app. Addressing this issue is possible, but the solution will depend on what the particular legacy system can technically support. Again, one size does not fit all.

Many CIO Roundtable attendees were blindsided by unexpected technical debt in the storage infrastructure.  Many AI use cases require data to be stored on premises because of data sovereignty, statutory, governance, or privacy requirements. And AI apps can depend on many more petabytes of data than what legacy storage infrastructure can support. Attendees reported that AI apps can run up against storage technical debt impacting every aspect of storage: management, capacity, and throughput. Invariably, they have had to invest in modern, capable technology. In some cases, storage that is cloud-like but remains on premises has been the best solution.

The bottom line is that many of the AI apps being designed today will have some level of dependence on legacy data or infrastructure. Unfortunately, there is no single broad framework that can help IT leaders foresee the technical debt issues that might arise with each individual AI app. The senior IT professionals who have attended recent CIO Roundtables agree that evaluating the particular technical debt issues that result from each specific new AI app or system is the only truly effective approach. If technical debt becomes a roadblock to making AI advances that could keep the organization competitive, it will become necessary to invest in the resources needed to eliminate it. Identifying potential issues as early as possible is the best way to ensure that delays don’t occur.