As agencies start to figure out the best mission areas to test out agentic artificial intelligence, it likely will come down to the data.

As in the early days of using large language models, the quality of data will drive success and confidence.

“Most organizations are trying to move slowly and work with the more curated sets of data that they have,” said Andrew Churchill, vice president of public sector at Qlik during Federal News Network’s Federal Leader’s Guide to the CAIO & CDO.

“Finance tends to be an area that’s highly curated. It’s built into data warehouses, marts, lakes and reports, and that’s an easy area for them to begin to build and adopt AI capabilities. Whereas you know some of these policy things, which rely a lot on the human involvement and assessment of what they’re seeing, that’s going to take more time.”

Despite AI being around for several years, agencies still find full implementation challenging because employees still are learning how best to use the tools and they have to be confident that their data will produce the best results.

In the Office of Management and Budget’s 2025 AI inventory, issued in April 2026, agencies reported 3,611 use cases at various stages of deployment, with about 440 still in the pilot stage and another 1,479 in the pre-deployment stage.

Churchill said as agentic AI tools mature, agencies need to improve their employees’ data literacy as well as establish a culture of willingness to test and improve their data sources.

“The culture is going to say we’re all going to participate in this effort to make things better. But also, to give people the opportunity — whether you say crawl, walk and run, or just trust but verify — let’s enable our organization to go out and start to use these systems and to accept the fact that there might be some errors in what we’re doing and build a structure around the way that we use that information to inform the final decision,” he said.

“We’ve got to be nimble. We’ve got to build contracts that allow things to pivot, change direction and adjust for this changing world from an AI tools perspective.”

This is why creating a culture of test and try is easier with back-office systems like finance and procurement, where agencies have trusted data sources and have confidence in the processes that AI can simplify and accelerate.

Churchill said back-office systems also are easier to add zero trust principles to as a way to instill more faith in the tools.

“Zero trust is the biggest piece, other than getting data right, that stands in the way of agentic scaling, and it’s simply because you may have access to these three systems, but that agent, in order to do its work for you, needs to get to 12 systems, and you don’t have access to that. How can I take the data from those other systems, put it into this context layer and then field that data for these responses through these agentic systems much more efficiently and easily?” he said.

“One of the things that we’re seeing a lot right now is we’re doing a lot of modernization of older reporting processes, and one of the reasons that agencies are doing it is because we will be able to put, for example, a model context protocol interface on the front end of what we do, making it available to any chatbot, to Palantir, to any company that might want to consume that data more effectively, and letting them move faster with, again, more trust and authority.”

The agentic AI tool needs its own identity and authentication approvals.

Churchill said as agencies expand their zero trust architectures, current obstacles to expanding agentic AI will come down.

“In most of our large enterprise deployments, we’re interfacing with dozens of systems on the back end. We’re taking data, characterizing that data and then we’re building out an access control for that data for an individual user. They’re being assigned the availability to get to this dashboard or what we call an app, it’s now the same thing,” he said.

“That agentic system is making a request on behalf of Jason Miller. Jason Miller has access to these things within Qlik that are predefined that are going to X number of systems on the back end, many of which would be very difficult to plug into the agentic system itself.”

Discover more Federal Leader’s Guide to the CAIO & CDO articles and videos now on the Federal News Network.

Copyright © 2026 Federal News Network. All rights reserved. This website is not intended for users located within the European Economic Area.